release: v2.14.1

MIN_STABILITY_TIME_MS is 5000ms in implementation, but test timeout was only 5000ms.
Extended to 10000ms to allow proper polling completion.

AGENTS.md

@@ -1,29 +1,30 @@
 # PROJECT KNOWLEDGE BASE
 
-**Generated:** 2026-01-02T22:41:22+09:00
-**Commit:** d0694e5
+**Generated:** 2026-01-09T15:38:00+09:00
+**Commit:** 0581793
 **Branch:** dev
 
 ## OVERVIEW
 
-OpenCode plugin: multi-model agent orchestration (Claude Opus 4.5, GPT-5.2, Gemini 3, Grok), 11 LSP tools, AST-Grep, Claude Code compatibility layer. "oh-my-zsh" for OpenCode.
+OpenCode plugin implementing Claude Code/AmpCode features. Multi-model agent orchestration (GPT-5.2, Claude, Gemini, Grok), LSP tools (11), AST-Grep search, MCP integrations (context7, websearch_exa, grep_app). "oh-my-zsh" for OpenCode.
 
 ## STRUCTURE
 
 ```
 oh-my-opencode/
 ├── src/
-│   ├── agents/        # 7 AI agents - see src/agents/AGENTS.md
+│   ├── agents/        # AI agents (7): Sisyphus, oracle, librarian, explore, frontend, document-writer, multimodal-looker
 │   ├── hooks/         # 22 lifecycle hooks - see src/hooks/AGENTS.md
-│   ├── tools/         # LSP, AST-Grep, session mgmt - see src/tools/AGENTS.md
+│   ├── tools/         # LSP, AST-Grep, Grep, Glob, session mgmt - see src/tools/AGENTS.md
 │   ├── features/      # Claude Code compat layer - see src/features/AGENTS.md
 │   ├── auth/          # Google Antigravity OAuth - see src/auth/AGENTS.md
 │   ├── shared/        # Cross-cutting utilities - see src/shared/AGENTS.md
 │   ├── cli/           # CLI installer, doctor - see src/cli/AGENTS.md
 │   ├── mcp/           # MCP configs: context7, grep_app
 │   ├── config/        # Zod schema, TypeScript types
-│   └── index.ts       # Main plugin entry (464 lines)
+│   └── index.ts       # Main plugin entry (548 lines)
 ├── script/            # build-schema.ts, publish.ts, generate-changelog.ts
+├── assets/            # JSON schema
 └── dist/              # Build output (ESM + .d.ts)
 ```
 
@@ -31,13 +32,25 @@ oh-my-opencode/
 
 | Task | Location | Notes |
 |------|----------|-------|
-| Add agent | `src/agents/` | Create .ts, add to builtinAgents, update types.ts |
-| Add hook | `src/hooks/` | Dir with createXXXHook(), export from index.ts |
-| Add tool | `src/tools/` | Dir with constants/types/tools.ts, add to builtinTools |
-| Add MCP | `src/mcp/` | Create config, add to index.ts |
-| Add skill | `src/features/builtin-skills/` | Dir with SKILL.md |
-| Config schema | `src/config/schema.ts` | Run `bun run build:schema` after |
+| Add agent | `src/agents/` | Create .ts, add to builtinAgents in index.ts, update types.ts |
+| Add hook | `src/hooks/` | Create dir with createXXXHook(), export from index.ts |
+| Add tool | `src/tools/` | Dir with index/types/constants/tools.ts, add to builtinTools |
+| Add MCP | `src/mcp/` | Create config, add to index.ts and types.ts |
+| Add skill | `src/features/builtin-skills/` | Create skill dir with SKILL.md |
+| LSP behavior | `src/tools/lsp/` | client.ts (connection), tools.ts (handlers) |
+| AST-Grep | `src/tools/ast-grep/` | napi.ts for @ast-grep/napi binding |
+| Google OAuth | `src/auth/antigravity/` | OAuth plugin for Google/Gemini models |
+| Config schema | `src/config/schema.ts` | Zod schema, run `bun run build:schema` after changes |
 | Claude Code compat | `src/features/claude-code-*-loader/` | Command, skill, agent, mcp loaders |
+| Background agents | `src/features/background-agent/` | manager.ts for task management |
+| Skill MCP | `src/features/skill-mcp-manager/` | MCP servers embedded in skills |
+| Interactive terminal | `src/tools/interactive-bash/` | tmux session management |
+| CLI installer | `src/cli/install.ts` | Interactive TUI installation |
+| Doctor checks | `src/cli/doctor/checks/` | Health checks for environment |
+| Shared utilities | `src/shared/` | Cross-cutting utilities |
+| Slash commands | `src/hooks/auto-slash-command/` | Auto-detect and execute `/command` patterns |
+| Ralph Loop | `src/hooks/ralph-loop/` | Self-referential dev loop until completion |
+| Orchestrator | `src/hooks/sisyphus-orchestrator/` | Main orchestration hook (660 lines) |
 
 ## TDD (Test-Driven Development)
 
@@ -52,46 +65,67 @@ oh-my-opencode/
 
 | Phase | Action | Verification |
 |-------|--------|--------------|
-| **RED** | Write test describing expected behavior | `bun test` → FAIL (expected) |
-| **GREEN** | Implement minimum code to pass | `bun test` → PASS |
-| **REFACTOR** | Improve code quality, remove duplication | `bun test` → PASS (must stay green) |
+| **RED** | Write test describing expected behavior | `bun test` -> FAIL (expected) |
+| **GREEN** | Implement minimum code to pass | `bun test` -> PASS |
+| **REFACTOR** | Improve code quality, remove duplication | `bun test` -> PASS (must stay green) |
 
 **Rules:**
 - NEVER write implementation before test
 - NEVER delete failing tests to "pass" - fix the code
 - One test at a time - don't batch
 - Test file naming: `*.test.ts` alongside source
+- BDD comments: `#given`, `#when`, `#then` (same as AAA)
 
 ## CONVENTIONS
 
-- **Bun only**: `bun run`, `bun test`, `bunx` (NEVER npm/npx)
+- **Package manager**: Bun only (`bun run`, `bun build`, `bunx`)
 - **Types**: bun-types (not @types/node)
 - **Build**: `bun build` (ESM) + `tsc --emitDeclarationOnly`
 - **Exports**: Barrel pattern in index.ts; explicit named exports for tools/hooks
 - **Naming**: kebab-case directories, createXXXHook/createXXXTool factories
-- **Testing**: BDD comments `#given`, `#when`, `#then` (same as AAA); TDD workflow (RED-GREEN-REFACTOR)
+- **Testing**: BDD comments `#given/#when/#then`, TDD workflow (RED-GREEN-REFACTOR)
 - **Temperature**: 0.1 for code agents, max 0.3
 
-## ANTI-PATTERNS
+## ANTI-PATTERNS (THIS PROJECT)
 
-| Category | Forbidden |
-|----------|-----------|
-| Type Safety | `as any`, `@ts-ignore`, `@ts-expect-error` |
-| Package Manager | npm, yarn, npx |
-| File Ops | Bash mkdir/touch/rm for code file creation |
-| Publishing | Direct `bun publish`, local version bump |
-| Agent Behavior | High temp (>0.3), broad tool access, sequential agent calls |
-| Hooks | Heavy PreToolUse logic, blocking without reason |
-| Year | 2024 in code/prompts (use current year) |
+- **npm/yarn**: Use bun exclusively
+- **@types/node**: Use bun-types
+- **Bash file ops**: Never mkdir/touch/rm/cp/mv for file creation in code
+- **Direct bun publish**: GitHub Actions workflow_dispatch only (OIDC provenance)
+- **Local version bump**: Version managed by CI workflow
+- **Year 2024**: NEVER use 2024 in code/prompts (use current year)
+- **Rush completion**: Never mark tasks complete without verification
+- **Over-exploration**: Stop searching when sufficient context found
+- **High temperature**: Don't use >0.3 for code-related agents
+- **Broad tool access**: Prefer explicit `include` over unrestricted access
+- **Sequential agent calls**: Use `sisyphus_task` for parallel execution
+- **Heavy PreToolUse logic**: Slows every tool call
+- **Self-planning for complex tasks**: Spawn planning agent (Prometheus) instead
+- **Trust agent self-reports**: ALWAYS verify results independently
+- **Skip TODO creation**: Multi-step tasks MUST have todos first
+- **Batch completions**: Mark TODOs complete immediately, don't group
+- **Giant commits**: 3+ files = 2+ commits minimum
+- **Separate test from impl**: Same commit always
+
+## UNIQUE STYLES
+
+- **Platform**: Union type `"darwin" | "linux" | "win32" | "unsupported"`
+- **Optional props**: Extensive `?` for optional interface properties
+- **Flexible objects**: `Record<string, unknown>` for dynamic configs
+- **Error handling**: Consistent try/catch with async/await
+- **Agent tools**: `tools: { include: [...] }` or `tools: { exclude: [...] }`
+- **Temperature**: Most agents use `0.1` for consistency
+- **Hook naming**: `createXXXHook` function convention
+- **Factory pattern**: Components created via `createXXX()` functions
 
 ## AGENT MODELS
 
 | Agent | Default Model | Purpose |
-|-------|-------|---------|
+|-------|---------------|---------|
 | Sisyphus | anthropic/claude-opus-4-5 | Primary orchestrator |
-| oracle | openai/gpt-5.2 | Strategy, code review |
-| librarian | anthropic/claude-sonnet-4-5 | Docs, OSS research |
-| explore | opencode/grok-code | Fast codebase grep |
+| oracle | openai/gpt-5.2 | Read-only consultation. High-IQ debugging, architecture |
+| librarian | opencode/glm-4.7-free | Multi-repo analysis, docs |
+| explore | opencode/grok-code | Fast codebase exploration |
 | frontend-ui-ux-engineer | google/gemini-3-pro-preview | UI generation |
 | document-writer | google/gemini-3-pro-preview | Technical docs |
 | multimodal-looker | google/gemini-3-flash | PDF/image analysis |
@@ -102,7 +136,8 @@ oh-my-opencode/
 bun run typecheck      # Type check
 bun run build          # ESM + declarations + schema
 bun run rebuild        # Clean + Build
-bun test               # Run tests (380+)
+bun run build:schema   # Schema only
+bun test               # Run tests (76 test files, 2559+ BDD assertions)
 ```
 
 ## DEPLOYMENT
@@ -110,26 +145,42 @@ bun test               # Run tests (380+)
 **GitHub Actions workflow_dispatch only**
 
 1. Never modify package.json version locally
-2. Commit & push to dev
-3. Trigger: `gh workflow run publish -f bump=patch|minor|major`
+2. Commit & push changes
+3. Trigger `publish` workflow: `gh workflow run publish -f bump=patch`
 
-CI auto-commits schema changes on master, maintains rolling `next` draft release on dev.
+**Critical**: Never `bun publish` directly. Never bump version locally.
+
+## CI PIPELINE
+
+- **ci.yml**: Parallel test/typecheck, build verification, auto-commit schema on master, rolling `next` draft release
+- **publish.yml**: Manual workflow_dispatch, version bump, changelog, OIDC npm publish
+- **sisyphus-agent.yml**: Agent-in-CI for automated issue handling via `@sisyphus-dev-ai` mentions
 
 ## COMPLEXITY HOTSPOTS
 
 | File | Lines | Description |
 |------|-------|-------------|
-| `src/index.ts` | 464 | Main plugin, all hook/tool init |
-| `src/cli/config-manager.ts` | 669 | JSONC parsing, env detection |
-| `src/auth/antigravity/fetch.ts` | 621 | Token refresh, URL rewriting |
-| `src/tools/lsp/client.ts` | 611 | LSP protocol, JSON-RPC |
-| `src/hooks/anthropic-context-window-limit-recovery/executor.ts` | 564 | Multi-stage recovery |
-| `src/agents/sisyphus.ts` | 504 | Orchestrator prompt |
+| `src/agents/orchestrator-sisyphus.ts` | 1484 | Orchestrator agent, complex delegation |
+| `src/features/builtin-skills/skills.ts` | 1230 | Skill definitions (frontend-ui-ux, playwright) |
+| `src/agents/prometheus-prompt.ts` | 982 | Planning agent system prompt |
+| `src/auth/antigravity/fetch.ts` | 798 | Token refresh, URL rewriting |
+| `src/auth/antigravity/thinking.ts` | 755 | Thinking block extraction |
+| `src/cli/config-manager.ts` | 725 | JSONC parsing, env detection |
+| `src/hooks/sisyphus-orchestrator/index.ts` | 660 | Orchestrator hook impl |
+| `src/agents/sisyphus.ts` | 641 | Main Sisyphus prompt |
+| `src/tools/lsp/client.ts` | 612 | LSP protocol, JSON-RPC |
+| `src/features/background-agent/manager.ts` | 608 | Task lifecycle |
+| `src/auth/antigravity/response.ts` | 599 | Response transformation, streaming |
+| `src/hooks/anthropic-context-window-limit-recovery/executor.ts` | 556 | Multi-stage recovery |
+| `src/index.ts` | 548 | Main plugin, all hook/tool init |
 
 ## NOTES
 
+- **Testing**: Bun native test (`bun test`), BDD-style `#given/#when/#then`, 76 test files
 - **OpenCode**: Requires >= 1.0.150
-- **Config**: `~/.config/opencode/oh-my-opencode.json` or `.opencode/oh-my-opencode.json`
-- **JSONC**: Config files support comments and trailing commas
-- **Claude Code**: Full compat layer for settings.json hooks, commands, skills, agents, MCPs
+- **Multi-lang docs**: README.md (EN), README.ko.md (KO), README.ja.md (JA), README.zh-cn.md (ZH-CN)
+- **Config**: `~/.config/opencode/oh-my-opencode.json` (user) or `.opencode/oh-my-opencode.json` (project)
+- **Trusted deps**: @ast-grep/cli, @ast-grep/napi, @code-yeongyu/comment-checker
+- **JSONC support**: Config files support comments (`// comment`, `/* block */`) and trailing commas
+- **Claude Code Compat**: Full compatibility layer for settings.json hooks, commands, skills, agents, MCPs
 - **Skill MCP**: Skills can embed MCP server configs in YAML frontmatter

README.ja.md

@@ -1,9 +1,12 @@
 > [!NOTE]
 >
-> *「私はエージェントが生成したコードと人間が書いたコードを区別できない、しかしはるかに多くのことを達成できる世界を作り、ソフトウェア革命を起こすことを目指しています。私はこの旅に個人的な時間、情熱、そして資金を注ぎ込んできましたし、これからもそうし続けます。」*
+> [![Sisyphus Labs — Sisyphus is the agent that codes like your team.](./.github/assets/sisyphuslabs.png?v=2)](https://sisyphuslabs.ai)
+> > **Sisyphusの完全製品化バージョンを構築中です。フロンティアエージェントの未来を定義します。<br />[こちら](https://sisyphuslabs.ai)からウェイトリストに参加してください。**
+
+> [!TIP]
 >
-> [![The Orchestrator is coming](./.github/assets/orchestrator-sisyphus.png)](https://x.com/justsisyphus/status/2006250634354548963)
-> > **オーケストレーターが来ます。今週中に。[Xで通知を受け取る](https://x.com/justsisyphus/status/2006250634354548963)**
+> [![The Orchestrator is now available in beta.](./.github/assets/orchestrator-sisyphus.png?v=3)](https://github.com/code-yeongyu/oh-my-opencode/releases/tag/v3.0.0-beta.1)
+> > **オーケストレーターがベータ版で利用可能になりました。`oh-my-opencode@3.0.0-beta.1`を使用してインストールしてください。**
 >
 > 一緒に歩みましょう！
 >
@@ -319,9 +322,9 @@ opencode auth login
 {
   "google_auth": false,
   "agents": {
-    "frontend-ui-ux-engineer": { "model": "google/gemini-3-pro-high" },
-    "document-writer": { "model": "google/gemini-3-flash" },
-    "multimodal-looker": { "model": "google/gemini-3-flash" }
+    "frontend-ui-ux-engineer": { "model": "google/antigravity-gemini-3-pro-high" },
+    "document-writer": { "model": "google/antigravity-gemini-3-flash" },
+    "multimodal-looker": { "model": "google/antigravity-gemini-3-flash" }
   }
 }
 ```
@@ -458,7 +461,7 @@ oh-my-opencode を削除するには：
 
 - **Sisyphus** (`anthropic/claude-opus-4-5`): **デフォルトエージェントです。** OpenCode のための強力な AI オーケストレーターです。専門のサブエージェントを活用して、複雑なタスクを計画、委任、実行します。バックグラウンドタスクへの委任と Todo ベースのワークフローを重視します。最大の推論能力を発揮するため、Claude Opus 4.5 と拡張思考 (32k token budget) を使用します。
 - **oracle** (`openai/gpt-5.2`): アーキテクチャ、コードレビュー、戦略立案のための専門アドバイザー。GPT-5.2 の卓越した論理的推論と深い分析能力を活用します。AmpCode からインスピレーションを得ました。
-- **librarian** (`anthropic/claude-sonnet-4-5` または `google/gemini-3-flash`): マルチリポジトリ分析、ドキュメント検索、実装例の調査を担当。Antigravity 認証が設定されている場合は Gemini 3 Flash を使用し、それ以外は Claude Sonnet 4.5 を使用して、深いコードベース理解と GitHub リサーチ、根拠に基づいた回答を提供します。AmpCode からインスピレーションを得ました。
+- **librarian** (`opencode/glm-4.7-free`): マルチリポジトリ分析、ドキュメント検索、実装例の調査を担当。GLM-4.7 Free を使用して、深いコードベース理解と GitHub リサーチ、根拠に基づいた回答を提供します。AmpCode からインスピレーションを得ました。
 - **explore** (`opencode/grok-code`、`google/gemini-3-flash`、または `anthropic/claude-haiku-4-5`): 高速なコードベース探索、ファイルパターンマッチング。Antigravity 認証が設定されている場合は Gemini 3 Flash を使用し、Claude max20 が利用可能な場合は Haiku を使用し、それ以外は Grok を使います。Claude Code からインスピレーションを得ました。
 - **frontend-ui-ux-engineer** (`google/gemini-3-pro-preview`): 開発者に転身したデザイナーという設定です。素晴らしい UI を作ります。美しく独創的な UI コードを生成することに長けた Gemini を使用します。
 - **document-writer** (`google/gemini-3-pro-preview`): テクニカルライティングの専門家という設定です。Gemini は文筆家であり、流れるような文章を書きます。
@@ -771,9 +774,9 @@ Oh My OpenCode は以下の場所からフックを読み込んで実行しま
 {
   "google_auth": false,
   "agents": {
-    "frontend-ui-ux-engineer": { "model": "google/gemini-3-pro-high" },
-    "document-writer": { "model": "google/gemini-3-flash" },
-    "multimodal-looker": { "model": "google/gemini-3-flash" }
+    "frontend-ui-ux-engineer": { "model": "google/antigravity-gemini-3-pro-high" },
+    "document-writer": { "model": "google/antigravity-gemini-3-flash" },
+    "multimodal-looker": { "model": "google/antigravity-gemini-3-flash" }
   }
 }
 ```
@@ -862,7 +865,8 @@ Oh My OpenCode は以下の場所からフックを読み込んで実行しま
 
 - **Sisyphus**: プライマリオーケストレーターエージェント (Claude Opus 4.5)
 - **OpenCode-Builder**: OpenCode のデフォルトビルドエージェント（SDK 制限により名前変更、デフォルトで無効）
-- **Planner-Sisyphus**: OpenCode のデフォルトプランエージェント（SDK 制限により名前変更、デフォルトで有効）
+- **Prometheus (Planner)**: OpenCode のデフォルトプランエージェント + work-planner 方法論（デフォルトで有効）
+- **Metis (Plan Consultant)**: 隠された要件と AI 失敗ポイントを特定する事前計画分析エージェント
 
 **設定オプション：**
 
@@ -911,8 +915,11 @@ Oh My OpenCode は以下の場所からフックを読み込んで実行しま
     "OpenCode-Builder": {
       "model": "anthropic/claude-opus-4"
     },
-    "Planner-Sisyphus": {
+    "Prometheus (Planner)": {
       "model": "openai/gpt-5.2"
+    },
+    "Metis (Plan Consultant)": {
+      "model": "anthropic/claude-sonnet-4-5"
     }
   }
 }
@@ -922,8 +929,8 @@ Oh My OpenCode は以下の場所からフックを読み込んで実行しま
 | --------------------------- | ------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------ |
 | `disabled`                  | `false` | `true` の場合、すべての Sisyphus オーケストレーションを無効化し、元の build/plan をプライマリとして復元します。                                                                       |
 | `default_builder_enabled`   | `false` | `true` の場合、OpenCode-Builder エージェントを有効化します（OpenCode build と同じ、SDK 制限により名前変更）。デフォルトでは無効です。                                                   |
-| `planner_enabled`           | `true`  | `true` の場合、Planner-Sisyphus エージェントを有効化します（OpenCode plan と同じ、SDK 制限により名前変更）。デフォルトで有効です。                                                       |
-| `replace_plan`              | `true`  | `true` の場合、デフォルトのプランエージェントをサブエージェントモードに降格させます。`false` に設定すると、Planner-Sisyphus とデフォルトのプランの両方を利用できます。                                |
+| `planner_enabled`           | `true`  | `true` の場合、Prometheus (Planner) エージェントを有効化します（work-planner 方法論を含む）。デフォルトで有効です。                                                                   |
+| `replace_plan`              | `true`  | `true` の場合、デフォルトのプランエージェントをサブエージェントモードに降格させます。`false` に設定すると、Prometheus (Planner) とデフォルトのプランの両方を利用できます。                             |
 
 ### Background Tasks
 

README.md

@@ -1,9 +1,12 @@
 > [!NOTE]
 >
-> *"I aim to spark a software revolution by creating a world where agent-generated code is indistinguishable from human code, yet capable of achieving vastly more. I have poured my personal time, passion, and funds into this journey, and I will continue to do so."*
+> [![Sisyphus Labs — Sisyphus is the agent that codes like your team.](./.github/assets/sisyphuslabs.png?v=2)](https://sisyphuslabs.ai)
+> > **We're building a fully productized version of Sisyphus to define the future of frontier agents. <br />Join the waitlist [here](https://sisyphuslabs.ai).**
+
+> [!TIP]
 >
-> [![The Orchestrator is coming](./.github/assets/orchestrator-sisyphus.png)](https://x.com/justsisyphus/status/2006250634354548963)
-> > **The Orchestrator is coming. This Week. [Get notified on X](https://x.com/justsisyphus/status/2006250634354548963)**
+> [![The Orchestrator is now available in beta.](./.github/assets/orchestrator-sisyphus.png?v=3)](https://github.com/code-yeongyu/oh-my-opencode/releases/tag/v3.0.0-beta.1)
+> > **The Orchestrator is now available in beta. Use `oh-my-opencode@3.0.0-beta.1` to install it.**
 >
 > Be with us!
 >
@@ -127,6 +130,7 @@ No stupid token consumption massive subagents here. No bloat tools here.
     - [Built-in Skills](#built-in-skills)
     - [Sisyphus Agent](#sisyphus-agent)
     - [Background Tasks](#background-tasks)
+    - [Categories](#categories)
     - [Hooks](#hooks)
     - [MCPs](#mcps)
     - [LSP](#lsp)
@@ -357,9 +361,9 @@ The `opencode-antigravity-auth` plugin uses different model names than the built
 {
   "google_auth": false,
   "agents": {
-    "frontend-ui-ux-engineer": { "model": "google/gemini-3-pro-high" },
-    "document-writer": { "model": "google/gemini-3-flash" },
-    "multimodal-looker": { "model": "google/gemini-3-flash" }
+    "frontend-ui-ux-engineer": { "model": "google/antigravity-gemini-3-pro-high" },
+    "document-writer": { "model": "google/antigravity-gemini-3-flash" },
+    "multimodal-looker": { "model": "google/antigravity-gemini-3-flash" }
   }
 }
 ```
@@ -495,9 +499,9 @@ To remove oh-my-opencode:
 
 - **Sisyphus** (`anthropic/claude-opus-4-5`): **The default agent.** A powerful AI orchestrator for OpenCode. Plans, delegates, and executes complex tasks using specialized subagents with aggressive parallel execution. Emphasizes background task delegation and todo-driven workflow. Uses Claude Opus 4.5 with extended thinking (32k budget) for maximum reasoning capability.
 - **oracle** (`openai/gpt-5.2`): Architecture, code review, strategy. Uses GPT-5.2 for its stellar logical reasoning and deep analysis. Inspired by AmpCode.
-- **librarian** (`anthropic/claude-sonnet-4-5` or `google/gemini-3-flash`): Multi-repo analysis, doc lookup, implementation examples. Uses Gemini 3 Flash when Antigravity auth is configured, otherwise Claude Sonnet 4.5 for deep codebase understanding and GitHub research with evidence-based answers. Inspired by AmpCode.
+- **librarian** (`opencode/glm-4.7-free`): Multi-repo analysis, doc lookup, implementation examples. Uses GLM-4.7 Free for deep codebase understanding and GitHub research with evidence-based answers. Inspired by AmpCode.
 - **explore** (`opencode/grok-code`, `google/gemini-3-flash`, or `anthropic/claude-haiku-4-5`): Fast codebase exploration and pattern matching. Uses Gemini 3 Flash when Antigravity auth is configured, Haiku when Claude max20 is available, otherwise Grok. Inspired by Claude Code.
-- **frontend-ui-ux-engineer** (`google/gemini-3-pro-high`): A designer turned developer. Builds gorgeous UIs. Gemini excels at creative, beautiful UI code.
+- **frontend-ui-ux-engineer** (`google/gemini-3-pro-preview`): A designer turned developer. Builds gorgeous UIs. Gemini excels at creative, beautiful UI code.
 - **document-writer** (`google/gemini-3-flash`): Technical writing expert. Gemini is a wordsmith—writes prose that flows.
 - **multimodal-looker** (`google/gemini-3-flash`): Visual content specialist. Analyzes PDFs, images, diagrams to extract information.
 
@@ -553,6 +557,7 @@ Hand your best tools to your best colleagues. Now they can properly refactor, na
 - **ast_grep_search**: AST-aware code pattern search (25 languages)
 - **ast_grep_replace**: AST-aware code replacement
 - **call_omo_agent**: Spawn specialized explore/librarian agents. Supports `run_in_background` parameter for async execution.
+- **sisyphus_task**: Category-based task delegation with specialized agents. Supports pre-configured categories (visual, business-logic) or direct agent targeting. Use `background_output` to retrieve results and `background_cancel` to cancel tasks. See [Categories](#categories).
 
 #### Session Management
 
@@ -825,9 +830,9 @@ When using `opencode-antigravity-auth`, disable the built-in auth and override a
 {
   "google_auth": false,
   "agents": {
-    "frontend-ui-ux-engineer": { "model": "google/gemini-3-pro-high" },
-    "document-writer": { "model": "google/gemini-3-flash" },
-    "multimodal-looker": { "model": "google/gemini-3-flash" }
+    "frontend-ui-ux-engineer": { "model": "google/antigravity-gemini-3-pro-high" },
+    "document-writer": { "model": "google/antigravity-gemini-3-flash" },
+    "multimodal-looker": { "model": "google/antigravity-gemini-3-flash" }
   }
 }
 ```
@@ -915,6 +920,7 @@ Available agents: `oracle`, `librarian`, `explore`, `frontend-ui-ux-engineer`, `
 Oh My OpenCode includes built-in skills that provide additional capabilities:
 
 - **playwright**: Browser automation with Playwright MCP. Use for web scraping, testing, screenshots, and browser interactions.
+- **git-master**: Git expert for atomic commits, rebase/squash, and history search (blame, bisect, log -S). STRONGLY RECOMMENDED: Use with `sisyphus_task(category='quick', skills=['git-master'], ...)` to save context.
 
 Disable built-in skills via `disabled_skills` in `~/.config/opencode/oh-my-opencode.json` or `.opencode/oh-my-opencode.json`:
 
@@ -924,7 +930,25 @@ Disable built-in skills via `disabled_skills` in `~/.config/opencode/oh-my-openc
 }
 ```
 
-Available built-in skills: `playwright`
+Available built-in skills: `playwright`, `git-master`
+
+### Git Master
+
+Configure git-master skill behavior:
+
+```json
+{
+  "git_master": {
+    "commit_footer": true,
+    "include_co_authored_by": true
+  }
+}
+```
+
+| Option | Default | Description |
+| ------ | ------- | ----------- |
+| `commit_footer` | `true` | Adds "Ultraworked with Sisyphus" footer to commit messages. |
+| `include_co_authored_by` | `true` | Adds `Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>` trailer to commits. |
 
 ### Sisyphus Agent
 
@@ -932,7 +956,8 @@ When enabled (default), Sisyphus provides a powerful orchestrator with optional
 
 - **Sisyphus**: Primary orchestrator agent (Claude Opus 4.5)
 - **OpenCode-Builder**: OpenCode's default build agent, renamed due to SDK limitations (disabled by default)
-- **Planner-Sisyphus**: OpenCode's default plan agent, renamed due to SDK limitations (enabled by default)
+- **Prometheus (Planner)**: OpenCode's default plan agent with work-planner methodology (enabled by default)
+- **Metis (Plan Consultant)**: Pre-planning analysis agent that identifies hidden requirements and AI failure points
 
 **Configuration Options:**
 
@@ -981,19 +1006,22 @@ You can also customize Sisyphus agents like other agents:
     "OpenCode-Builder": {
       "model": "anthropic/claude-opus-4"
     },
-    "Planner-Sisyphus": {
+    "Prometheus (Planner)": {
       "model": "openai/gpt-5.2"
+    },
+    "Metis (Plan Consultant)": {
+      "model": "anthropic/claude-sonnet-4-5"
     }
   }
 }
 ```
 
-| Option                    | Default | Description                                                                                                                        |
-| ------------------------- | ------- | ---------------------------------------------------------------------------------------------------------------------------------- |
-| `disabled`                | `false` | When `true`, disables all Sisyphus orchestration and restores original build/plan as primary.                                      |
-| `default_builder_enabled` | `false` | When `true`, enables OpenCode-Builder agent (same as OpenCode build, renamed due to SDK limitations). Disabled by default.         |
-| `planner_enabled`         | `true`  | When `true`, enables Planner-Sisyphus agent (same as OpenCode plan, renamed due to SDK limitations). Enabled by default.           |
-| `replace_plan`            | `true`  | When `true`, demotes default plan agent to subagent mode. Set to `false` to keep both Planner-Sisyphus and default plan available. |
+| Option                      | Default | Description                                                                                                                                         |
+| --------------------------- | ------- | --------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `disabled`                  | `false` | When `true`, disables all Sisyphus orchestration and restores original build/plan as primary.                                                       |
+| `default_builder_enabled`   | `false` | When `true`, enables OpenCode-Builder agent (same as OpenCode build, renamed due to SDK limitations). Disabled by default.                         |
+| `planner_enabled`           | `true`  | When `true`, enables Prometheus (Planner) agent with work-planner methodology. Enabled by default.                                                 |
+| `replace_plan`              | `true`  | When `true`, demotes default plan agent to subagent mode. Set to `false` to keep both Prometheus (Planner) and default plan available.             |
 
 ### Background Tasks
 
@@ -1029,6 +1057,50 @@ Configure concurrency limits for background agent tasks. This controls how many
 - Allow more concurrent tasks for fast/cheap models (e.g., Gemini Flash)
 - Respect provider rate limits by setting provider-level caps
 
+### Categories
+
+Categories enable domain-specific task delegation via the `sisyphus_task` tool. Each category pre-configures a specialized `Sisyphus-Junior-{category}` agent with optimized model settings and prompts.
+
+**Default Categories:**
+
+| Category | Model | Description |
+|----------|-------|-------------|
+| `visual` | `google/gemini-3-pro-preview` | Frontend, UI/UX, design-focused tasks. High creativity (temp 0.7). |
+| `business-logic` | `openai/gpt-5.2` | Backend logic, architecture, strategic reasoning. Low creativity (temp 0.1). |
+
+**Usage:**
+
+```
+// Via sisyphus_task tool
+sisyphus_task(category="visual", prompt="Create a responsive dashboard component")
+sisyphus_task(category="business-logic", prompt="Design the payment processing flow")
+
+// Or target a specific agent directly
+sisyphus_task(agent="oracle", prompt="Review this architecture")
+```
+
+**Custom Categories:**
+
+Add custom categories in `oh-my-opencode.json`:
+
+```json
+{
+  "categories": {
+    "data-science": {
+      "model": "anthropic/claude-sonnet-4-5",
+      "temperature": 0.2,
+      "prompt_append": "Focus on data analysis, ML pipelines, and statistical methods."
+    },
+    "visual": {
+      "model": "google/gemini-3-pro-preview",
+      "prompt_append": "Use shadcn/ui components and Tailwind CSS."
+    }
+  }
+}
+```
+
+Each category supports: `model`, `temperature`, `top_p`, `maxTokens`, `thinking`, `reasoningEffort`, `textVerbosity`, `tools`, `prompt_append`.
+
 ### Hooks
 
 Disable specific built-in hooks via `disabled_hooks` in `~/.config/opencode/oh-my-opencode.json` or `.opencode/oh-my-opencode.json`:

README.zh-cn.md

@@ -1,9 +1,12 @@
 > [!NOTE]
 >
-> *"我致力于引发一场软件革命，创造一个AI生成的代码与人类代码无法区分、却能实现更多的世界。我已经在这段旅程中投入了个人时间、热情和资金，并将继续这样做。"*
+> [![Sisyphus Labs — Sisyphus is the agent that codes like your team.](./.github/assets/sisyphuslabs.png?v=2)](https://sisyphuslabs.ai)
+> > **我们正在构建Sisyphus的完全产品化版本，定义前沿代理的未来。<br />[点击此处](https://sisyphuslabs.ai)加入候补名单。**
+
+> [!TIP]
 >
-> [![The Orchestrator is coming](./.github/assets/orchestrator-sisyphus.png)](https://x.com/justsisyphus/status/2006250634354548963)
-> > **编排器即将到来。就在本周。[在X上获取通知](https://x.com/justsisyphus/status/2006250634354548963)**
+> [![The Orchestrator is now available in beta.](./.github/assets/orchestrator-sisyphus.png?v=3)](https://github.com/code-yeongyu/oh-my-opencode/releases/tag/v3.0.0-beta.1)
+> > **编排器现已推出测试版。使用`oh-my-opencode@3.0.0-beta.1`来安装。**
 >
 > 与我们同行！
 >
@@ -324,9 +327,9 @@ opencode auth login
 {
   "google_auth": false,
   "agents": {
-    "frontend-ui-ux-engineer": { "model": "google/gemini-3-pro-high" },
-    "document-writer": { "model": "google/gemini-3-flash" },
-    "multimodal-looker": { "model": "google/gemini-3-flash" }
+    "frontend-ui-ux-engineer": { "model": "google/antigravity-gemini-3-pro-high" },
+    "document-writer": { "model": "google/antigravity-gemini-3-flash" },
+    "multimodal-looker": { "model": "google/antigravity-gemini-3-flash" }
   }
 }
 ```
@@ -463,7 +466,7 @@ gh repo star code-yeongyu/oh-my-opencode
 
 - **Sisyphus** (`anthropic/claude-opus-4-5`)：**默认 Agent。** OpenCode 专属的强力 AI 编排器。指挥专业子 Agent 搞定复杂任务。主打后台任务委派和 Todo 驱动。用 Claude Opus 4.5 加上扩展思考（32k token 预算），智商拉满。
 - **oracle** (`openai/gpt-5.2`)：架构师、代码审查员、战略家。GPT-5.2 的逻辑推理和深度分析能力不是盖的。致敬 AmpCode。
-- **librarian** (`anthropic/claude-sonnet-4-5` 或 `google/gemini-3-flash`)：多仓库分析、查文档、找示例。配置 Antigravity 认证时使用 Gemini 3 Flash，否则使用 Claude Sonnet 4.5 深入理解代码库，GitHub 调研，给出的答案都有据可查。致敬 AmpCode。
+- **librarian** (`opencode/glm-4.7-free`)：多仓库分析、查文档、找示例。使用 GLM-4.7 Free 深入理解代码库，GitHub 调研，给出的答案都有据可查。致敬 AmpCode。
 - **explore** (`opencode/grok-code`、`google/gemini-3-flash` 或 `anthropic/claude-haiku-4-5`)：极速代码库扫描、模式匹配。配置 Antigravity 认证时使用 Gemini 3 Flash，Claude max20 可用时使用 Haiku，否则用 Grok。致敬 Claude Code。
 - **frontend-ui-ux-engineer** (`google/gemini-3-pro-preview`)：设计师出身的程序员。UI 做得那是真漂亮。Gemini 写这种创意美观的代码是一绝。
 - **document-writer** (`google/gemini-3-pro-preview`)：技术写作专家。Gemini 文笔好，写出来的东西读着顺畅。
@@ -772,9 +775,9 @@ Agent 爽了，你自然也爽。但我还想直接让你爽。
 {
   "google_auth": false,
   "agents": {
-    "frontend-ui-ux-engineer": { "model": "google/gemini-3-pro-high" },
-    "document-writer": { "model": "google/gemini-3-flash" },
-    "multimodal-looker": { "model": "google/gemini-3-flash" }
+    "frontend-ui-ux-engineer": { "model": "google/antigravity-gemini-3-pro-high" },
+    "document-writer": { "model": "google/antigravity-gemini-3-flash" },
+    "multimodal-looker": { "model": "google/antigravity-gemini-3-flash" }
   }
 }
 ```
@@ -863,7 +866,8 @@ Agent 爽了，你自然也爽。但我还想直接让你爽。
 
 - **Sisyphus**：主编排 Agent（Claude Opus 4.5）
 - **OpenCode-Builder**：OpenCode 默认构建 Agent（因 SDK 限制仅改名，默认禁用）
-- **Planner-Sisyphus**：OpenCode 默认计划 Agent（因 SDK 限制仅改名，默认启用）
+- **Prometheus (Planner)**：OpenCode 默认计划 Agent + work-planner 方法论（默认启用）
+- **Metis (Plan Consultant)**：识别隐藏需求和 AI 失败点的预规划分析 Agent
 
 **配置选项：**
 
@@ -912,8 +916,11 @@ Sisyphus Agent 也能自定义：
     "OpenCode-Builder": {
       "model": "anthropic/claude-opus-4"
     },
-    "Planner-Sisyphus": {
+    "Prometheus (Planner)": {
       "model": "openai/gpt-5.2"
+    },
+    "Metis (Plan Consultant)": {
+      "model": "anthropic/claude-sonnet-4-5"
     }
   }
 }
@@ -923,8 +930,8 @@ Sisyphus Agent 也能自定义：
 | --------------------------- | ------- | ------------------------------------------------------------------------------------------------------------------------------------------------- |
 | `disabled`                  | `false` | 设为 `true` 就禁用所有 Sisyphus 编排，恢复原来的 build/plan。                                                                                              |
 | `default_builder_enabled`   | `false` | 设为 `true` 就启用 OpenCode-Builder Agent（与 OpenCode build 相同，因 SDK 限制仅改名）。默认禁用。                                                           |
-| `planner_enabled`           | `true`  | 设为 `true` 就启用 Planner-Sisyphus Agent（与 OpenCode plan 相同，因 SDK 限制仅改名）。默认启用。                                                             |
-| `replace_plan`              | `true`  | 设为 `true` 就把默认计划 Agent 降级为子 Agent 模式。设为 `false` 可以同时保留 Planner-Sisyphus 和默认计划。                                                        |
+| `planner_enabled`           | `true`  | 设为 `true` 就启用 Prometheus (Planner) Agent（含 work-planner 方法论）。默认启用。                                                                         |
+| `replace_plan`              | `true`  | 设为 `true` 就把默认计划 Agent 降级为子 Agent 模式。设为 `false` 可以同时保留 Prometheus (Planner) 和默认计划。                                                      |
 
 ### Background Tasks（后台任务）
 

assets/oh-my-opencode.schema.json

@@ -26,7 +26,10 @@
           "explore",
           "frontend-ui-ux-engineer",
           "document-writer",
-          "multimodal-looker"
+          "multimodal-looker",
+          "Metis (Plan Consultant)",
+          "Momus (Plan Reviewer)",
+          "orchestrator-sisyphus"
         ]
       }
     },
@@ -35,7 +38,9 @@
       "items": {
         "type": "string",
         "enum": [
-          "playwright"
+          "playwright",
+          "frontend-ui-ux",
+          "git-master"
         ]
       }
     },
@@ -71,7 +76,10 @@
           "compaction-context-injector",
           "claude-code-hooks",
           "auto-slash-command",
-          "edit-error-recovery"
+          "edit-error-recovery",
+          "prometheus-md-only",
+          "start-work",
+          "sisyphus-orchestrator"
         ]
       }
     },
@@ -80,7 +88,8 @@
       "items": {
         "type": "string",
         "enum": [
-          "init-deep"
+          "init-deep",
+          "start-work"
         ]
       }
     },
@@ -93,6 +102,15 @@
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -207,6 +225,15 @@
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -321,6 +348,15 @@
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -435,6 +471,15 @@
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -543,12 +588,267 @@
             }
           }
         },
-        "Planner-Sisyphus": {
+        "Prometheus (Planner)": {
           "type": "object",
           "properties": {
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
+            "temperature": {
+              "type": "number",
+              "minimum": 0,
+              "maximum": 2
+            },
+            "top_p": {
+              "type": "number",
+              "minimum": 0,
+              "maximum": 1
+            },
+            "prompt": {
+              "type": "string"
+            },
+            "prompt_append": {
+              "type": "string"
+            },
+            "tools": {
+              "type": "object",
+              "propertyNames": {
+                "type": "string"
+              },
+              "additionalProperties": {
+                "type": "boolean"
+              }
+            },
+            "disable": {
+              "type": "boolean"
+            },
+            "description": {
+              "type": "string"
+            },
+            "mode": {
+              "type": "string",
+              "enum": [
+                "subagent",
+                "primary",
+                "all"
+              ]
+            },
+            "color": {
+              "type": "string",
+              "pattern": "^#[0-9A-Fa-f]{6}$"
+            },
+            "permission": {
+              "type": "object",
+              "properties": {
+                "edit": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                },
+                "bash": {
+                  "anyOf": [
+                    {
+                      "type": "string",
+                      "enum": [
+                        "ask",
+                        "allow",
+                        "deny"
+                      ]
+                    },
+                    {
+                      "type": "object",
+                      "propertyNames": {
+                        "type": "string"
+                      },
+                      "additionalProperties": {
+                        "type": "string",
+                        "enum": [
+                          "ask",
+                          "allow",
+                          "deny"
+                        ]
+                      }
+                    }
+                  ]
+                },
+                "webfetch": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                },
+                "doom_loop": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                },
+                "external_directory": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                }
+              }
+            }
+          }
+        },
+        "Metis (Plan Consultant)": {
+          "type": "object",
+          "properties": {
+            "model": {
+              "type": "string"
+            },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
+            "temperature": {
+              "type": "number",
+              "minimum": 0,
+              "maximum": 2
+            },
+            "top_p": {
+              "type": "number",
+              "minimum": 0,
+              "maximum": 1
+            },
+            "prompt": {
+              "type": "string"
+            },
+            "prompt_append": {
+              "type": "string"
+            },
+            "tools": {
+              "type": "object",
+              "propertyNames": {
+                "type": "string"
+              },
+              "additionalProperties": {
+                "type": "boolean"
+              }
+            },
+            "disable": {
+              "type": "boolean"
+            },
+            "description": {
+              "type": "string"
+            },
+            "mode": {
+              "type": "string",
+              "enum": [
+                "subagent",
+                "primary",
+                "all"
+              ]
+            },
+            "color": {
+              "type": "string",
+              "pattern": "^#[0-9A-Fa-f]{6}$"
+            },
+            "permission": {
+              "type": "object",
+              "properties": {
+                "edit": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                },
+                "bash": {
+                  "anyOf": [
+                    {
+                      "type": "string",
+                      "enum": [
+                        "ask",
+                        "allow",
+                        "deny"
+                      ]
+                    },
+                    {
+                      "type": "object",
+                      "propertyNames": {
+                        "type": "string"
+                      },
+                      "additionalProperties": {
+                        "type": "string",
+                        "enum": [
+                          "ask",
+                          "allow",
+                          "deny"
+                        ]
+                      }
+                    }
+                  ]
+                },
+                "webfetch": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                },
+                "doom_loop": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                },
+                "external_directory": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                }
+              }
+            }
+          }
+        },
+        "Momus (Plan Reviewer)": {
+          "type": "object",
+          "properties": {
+            "model": {
+              "type": "string"
+            },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -663,6 +963,15 @@
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -777,6 +1086,15 @@
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -891,6 +1209,15 @@
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -1005,6 +1332,15 @@
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -1119,6 +1455,15 @@
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -1233,6 +1578,138 @@
             "model": {
               "type": "string"
             },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
+            "temperature": {
+              "type": "number",
+              "minimum": 0,
+              "maximum": 2
+            },
+            "top_p": {
+              "type": "number",
+              "minimum": 0,
+              "maximum": 1
+            },
+            "prompt": {
+              "type": "string"
+            },
+            "prompt_append": {
+              "type": "string"
+            },
+            "tools": {
+              "type": "object",
+              "propertyNames": {
+                "type": "string"
+              },
+              "additionalProperties": {
+                "type": "boolean"
+              }
+            },
+            "disable": {
+              "type": "boolean"
+            },
+            "description": {
+              "type": "string"
+            },
+            "mode": {
+              "type": "string",
+              "enum": [
+                "subagent",
+                "primary",
+                "all"
+              ]
+            },
+            "color": {
+              "type": "string",
+              "pattern": "^#[0-9A-Fa-f]{6}$"
+            },
+            "permission": {
+              "type": "object",
+              "properties": {
+                "edit": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                },
+                "bash": {
+                  "anyOf": [
+                    {
+                      "type": "string",
+                      "enum": [
+                        "ask",
+                        "allow",
+                        "deny"
+                      ]
+                    },
+                    {
+                      "type": "object",
+                      "propertyNames": {
+                        "type": "string"
+                      },
+                      "additionalProperties": {
+                        "type": "string",
+                        "enum": [
+                          "ask",
+                          "allow",
+                          "deny"
+                        ]
+                      }
+                    }
+                  ]
+                },
+                "webfetch": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                },
+                "doom_loop": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                },
+                "external_directory": {
+                  "type": "string",
+                  "enum": [
+                    "ask",
+                    "allow",
+                    "deny"
+                  ]
+                }
+              }
+            }
+          }
+        },
+        "orchestrator-sisyphus": {
+          "type": "object",
+          "properties": {
+            "model": {
+              "type": "string"
+            },
+            "category": {
+              "type": "string"
+            },
+            "skills": {
+              "type": "array",
+              "items": {
+                "type": "string"
+              }
+            },
             "temperature": {
               "type": "number",
               "minimum": 0,
@@ -1343,6 +1820,82 @@
         }
       }
     },
+    "categories": {
+      "type": "object",
+      "propertyNames": {
+        "type": "string"
+      },
+      "additionalProperties": {
+        "type": "object",
+        "properties": {
+          "model": {
+            "type": "string"
+          },
+          "temperature": {
+            "type": "number",
+            "minimum": 0,
+            "maximum": 2
+          },
+          "top_p": {
+            "type": "number",
+            "minimum": 0,
+            "maximum": 1
+          },
+          "maxTokens": {
+            "type": "number"
+          },
+          "thinking": {
+            "type": "object",
+            "properties": {
+              "type": {
+                "type": "string",
+                "enum": [
+                  "enabled",
+                  "disabled"
+                ]
+              },
+              "budgetTokens": {
+                "type": "number"
+              }
+            },
+            "required": [
+              "type"
+            ]
+          },
+          "reasoningEffort": {
+            "type": "string",
+            "enum": [
+              "low",
+              "medium",
+              "high"
+            ]
+          },
+          "textVerbosity": {
+            "type": "string",
+            "enum": [
+              "low",
+              "medium",
+              "high"
+            ]
+          },
+          "tools": {
+            "type": "object",
+            "propertyNames": {
+              "type": "string"
+            },
+            "additionalProperties": {
+              "type": "boolean"
+            }
+          },
+          "prompt_append": {
+            "type": "string"
+          }
+        },
+        "required": [
+          "model"
+        ]
+      }
+    },
     "claude_code": {
       "type": "object",
       "properties": {
@@ -1695,6 +2248,19 @@
           "type": "boolean"
         }
       }
+    },
+    "git_master": {
+      "type": "object",
+      "properties": {
+        "commit_footer": {
+          "default": true,
+          "type": "boolean"
+        },
+        "include_co_authored_by": {
+          "default": true,
+          "type": "boolean"
+        }
+      }
     }
   }
 }

docs/CRASH_INVESTIGATION_TIMELINE.md

@@ -1,152 +0,0 @@
-# Windows Crash Investigation Timeline
-
-## Executive Summary
-
-**Initial Hypothesis**: Bun.spawn/ShellInterpreter GC bug causing crashes on Windows  
-**Actual Root Cause**: Conflict between oh-my-opencode's session-notification and external notification plugins (specifically `@mohak34/opencode-notifier`)
-
-**Evidence**: User removed `@mohak34/opencode-notifier` plugin → crashes stopped immediately. The release version of oh-my-opencode (with original Bun.spawn code) works fine when used alone.
-
----
-
-## Timeline
-
-### Phase 1: Initial Crash Reports (Early January 2026)
-
-**Symptoms:**
-- Windows users experiencing crashes after extended oh-my-opencode usage
-- Stack traces pointed to Bun's ShellInterpreter finalizer:
-  ```
-  Segmentation fault at address 0x337081E00E0
-  - interpreter.zig:1239: deinitFromFinalizer
-  - ZigGeneratedClasses.zig:19925: ShellInterpreterClass__finalize
-  ```
-
-**Initial Analysis:**
-- Similar to known Bun issues: oven-sh/bun#23177, oven-sh/bun#24368
-- Focus on `ctx.$` (Bun shell template literals) in session-notification.ts
-
-### Phase 2: PR #543 - Wrong Fix Merged (January 6, 2026)
-
-**PR**: [#543 - fix(session-notification): avoid Bun shell GC crash on Windows](https://github.com/code-yeongyu/oh-my-opencode/pull/543)
-
-**Changes Made:**
-- Replaced `ctx.$` with `node:child_process.spawn` in `session-notification.ts`
-- Updated tests to mock spawn instead of ctx.$
-
-**Assumption**: The ShellInterpreter GC bug was causing crashes when notification commands were executed.
-
-**Status**: ❌ MERGED (reverted in this PR)
-
-### Phase 3: Continued Investigation - Debug Tracing (January 6-7, 2026)
-
-Crashes continued after PR #543. Added debug tracing system (PR #571) to capture what happens before crashes.
-
-**PR #571**: [feat(debug): add comprehensive crash tracing system](https://github.com/code-yeongyu/oh-my-opencode/pull/571)
-
-Tracing revealed LSP ENOENT errors, leading to:
-
-**PR #572**: [fix(lsp): add resilient handling for missing LSP server binaries](https://github.com/code-yeongyu/oh-my-opencode/pull/572)
-
-### Phase 4: More Bun.spawn Changes (January 7, 2026) - WRONG PATH
-
-Based on the assumption that Bun.spawn was the issue, additional files were modified locally:
-- `src/hooks/session-notification-utils.ts`
-- `src/hooks/comment-checker/cli.ts`
-- `src/hooks/comment-checker/downloader.ts`
-- `src/hooks/interactive-bash-session/index.ts`
-
-**Status**: ❌ REVERTED (never committed)
-
-### Phase 5: Root Cause Discovery (January 7, 2026)
-
-**Critical Observation by User:**
-> "I removed `@mohak34/opencode-notifier` and crashes stopped. The release version with Bun.spawn works perfectly fine."
-
-**Key Evidence:**
-1. Removing ONLY the notifier plugin fixed crashes
-2. Release version (before PR #543) works fine for user and most others
-3. No widespread complaints from other users about crashes
-4. PR #543 was based on superficial pattern matching with Bun issues
-
----
-
-## The Real Root Cause: Notification Plugin Conflict
-
-### Two Plugins, Same Event
-
-Both plugins listen to `session.idle` and send notifications:
-
-| Aspect | oh-my-opencode | opencode-notifier |
-|--------|---------------|-------------------|
-| **Event** | `session.idle` | `session.idle` |
-| **Delay** | 1.5s confirmation delay | Immediate |
-| **Windows Notification** | PowerShell + Windows.UI.Notifications API | `node-notifier` → WindowsToaster → SnoreToast.exe |
-| **Sound** | PowerShell Media.SoundPlayer | PowerShell Media.SoundPlayer |
-| **Process spawning** | `ctx.$` (Bun shell) | `node:child_process` |
-
-### Conflict Points
-
-1. **Different notification systems fighting**:
-   - oh-my-opencode: Direct PowerShell → Windows.UI.Notifications
-   - opencode-notifier: SnoreToast.exe binary via node-notifier
-
-2. **Same app identity**: Both register with "OpenCode" as the toast notifier app
-
-3. **Concurrent execution**: Both trigger within milliseconds of each other on `session.idle`
-
-4. **Resource contention**: Windows Toast API may not handle concurrent registrations gracefully
-
-### Why It Wasn't Bun.spawn
-
-- Both plugins use different spawning methods - this didn't matter
-- Release version works fine when used alone
-- Most users don't have this issue (most don't use both plugins)
-- The stack trace pointed to ShellInterpreter, but correlation ≠ causation
-
----
-
-## The Fix
-
-### What This PR Does
-
-1. **Reverts PR #543**: Restores original `ctx.$` usage (it was never the problem)
-
-2. **Adds conflict detection**: 
-   - Scans `opencode.json` for known notification plugins
-   - Known plugins: `opencode-notifier`, `@mohak34/opencode-notifier`
-
-3. **Auto-disables on conflict**:
-   - When external notifier detected, skips creating session-notification hook
-   - Logs clear warning explaining why
-
-4. **Config override**:
-   ```json
-   {
-     "notification": {
-       "force_enable": true
-     }
-   }
-   ```
-   Users can force-enable oh-my-opencode's notification if they want.
-
----
-
-## Lessons Learned
-
-1. **Correlation ≠ Causation**: Stack traces can be misleading; investigate root cause thoroughly
-2. **Test with user's exact environment**: The crash only happened with specific plugin combination
-3. **Challenge assumptions**: "Bun.spawn is buggy" was accepted too quickly without verifying
-4. **Evidence-based debugging**: User's discovery (removing notifier = no crash) was the key evidence
-
----
-
-## Related Links
-
-- PR #543 (merged, reverted in this PR): https://github.com/code-yeongyu/oh-my-opencode/pull/543
-- PR #571 (open): https://github.com/code-yeongyu/oh-my-opencode/pull/571
-- PR #572 (open): https://github.com/code-yeongyu/oh-my-opencode/pull/572
-- opencode-notifier: https://github.com/mohak34/opencode-notifier
-- Bun issues referenced (not actually the cause):
-  - https://github.com/oven-sh/bun/issues/23177
-  - https://github.com/oven-sh/bun/issues/24368

docs/category-skill-guide.md

@@ -0,0 +1,200 @@
+# Category & Skill System Guide
+
+This document provides a comprehensive guide to the **Category** and **Skill** systems, which form the extensibility core of Oh-My-OpenCode.
+
+## 1. Overview
+
+Instead of delegating everything to a single AI agent, it's far more efficient to invoke **specialists** tailored to the nature of the task.
+
+- **Category**: "What kind of work is this?" (determines model, temperature, prompt mindset)
+- **Skill**: "What tools and knowledge are needed?" (injects specialized knowledge, MCP tools, workflows)
+
+By combining these two concepts, you can generate optimal agents through `sisyphus_task`.
+
+---
+
+## 2. Category System
+
+A Category is an agent configuration preset optimized for specific domains.
+
+### Available Built-in Categories
+
+| Category | Optimal Model | Characteristics | Use Cases |
+|----------|---------------|-----------------|-----------|
+| `visual-engineering` | `gemini-3-pro` | High creativity (Temp 0.7) | Frontend, UI/UX, animations, styling |
+| `ultrabrain` | `gpt-5.2` | Maximum logical reasoning (Temp 0.1) | Architecture design, complex business logic, debugging |
+| `artistry` | `gemini-3-pro` | Artistic (Temp 0.9) | Creative ideation, design concepts, storytelling |
+| `quick` | `claude-haiku` | Fast (Temp 0.3) | Simple tasks, refactoring, script writing |
+| `writing` | `gemini-3-flash` | Natural flow (Temp 0.5) | Documentation, technical blogs, README writing |
+| `most-capable` | `claude-opus` | High performance (Temp 0.1) | Extremely difficult complex tasks |
+
+### Usage
+
+Specify the `category` parameter when invoking the `sisyphus_task` tool.
+
+```typescript
+sisyphus_task(
+  category="visual-engineering",
+  prompt="Add a responsive chart component to the dashboard page"
+)
+```
+
+### Sisyphus-Junior (Delegated Executor)
+
+When you use a Category, a special agent called **Sisyphus-Junior** performs the work.
+- **Characteristic**: Cannot **re-delegate** tasks to other agents.
+- **Purpose**: Prevents infinite delegation loops and ensures focus on the assigned task.
+
+---
+
+## 3. Skill System
+
+A Skill is a mechanism that injects **specialized knowledge (Context)** and **tools (MCP)** for specific domains into agents.
+
+### Built-in Skills
+
+1. **`git-master`**
+   - **Capabilities**: Git expert. Detects commit styles, splits atomic commits, formulates rebase strategies.
+   - **MCP**: None (uses Git commands)
+   - **Usage**: Essential for commits, history searches, branch management.
+
+2. **`playwright`**
+   - **Capabilities**: Browser automation. Web page testing, screenshots, scraping.
+   - **MCP**: `@playwright/mcp` (auto-executed)
+   - **Usage**: For post-implementation UI verification, E2E test writing.
+
+3. **`frontend-ui-ux`**
+   - **Capabilities**: Injects designer mindset. Color, typography, motion guidelines.
+   - **Usage**: For aesthetic UI work beyond simple implementation.
+
+### Usage
+
+Add desired skill names to the `skills` array.
+
+```typescript
+sisyphus_task(
+  category="quick",
+  skills=["git-master"],
+  prompt="Commit current changes. Follow commit message style."
+)
+```
+
+### Skill Customization (SKILL.md)
+
+You can add custom skills directly to `.opencode/skills/` in your project root or `~/.claude/skills/` in your home directory.
+
+**Example: `.opencode/skills/my-skill/SKILL.md`**
+
+```markdown
+---
+name: my-skill
+description: My special custom skill
+mcp:
+  my-mcp:
+    command: npx
+    args: ["-y", "my-mcp-server"]
+---
+
+# My Skill Prompt
+
+This content will be injected into the agent's system prompt.
+...
+```
+
+---
+
+## 4. Combination Strategies (Combos)
+
+You can create powerful specialized agents by combining Categories and Skills.
+
+### 🎨 The Designer (UI Implementation)
+- **Category**: `visual-engineering`
+- **Skills**: `["frontend-ui-ux", "playwright"]`
+- **Effect**: Implements aesthetic UI and verifies rendering results directly in browser.
+
+### 🏗️ The Architect (Design Review)
+- **Category**: `ultrabrain`
+- **Skills**: `[]` (pure reasoning)
+- **Effect**: Leverages GPT-5.2's logical reasoning for in-depth system architecture analysis.
+
+### ⚡ The Maintainer (Quick Fixes)
+- **Category**: `quick`
+- **Skills**: `["git-master"]`
+- **Effect**: Uses cost-effective models to quickly fix code and generate clean commits.
+
+---
+
+## 5. sisyphus_task Prompt Guide
+
+When delegating, **clear and specific** prompts are essential. Include these 7 elements:
+
+1. **TASK**: What needs to be done? (single objective)
+2. **EXPECTED OUTCOME**: What is the deliverable?
+3. **REQUIRED SKILLS**: Which skills should be used?
+4. **REQUIRED TOOLS**: Which tools must be used? (whitelist)
+5. **MUST DO**: What must be done (constraints)
+6. **MUST NOT DO**: What must never be done
+7. **CONTEXT**: File paths, existing patterns, reference materials
+
+**Bad Example**:
+> "Fix this"
+
+**Good Example**:
+> **TASK**: Fix mobile layout breaking issue in `LoginButton.tsx`
+> **CONTEXT**: `src/components/LoginButton.tsx`, using Tailwind CSS
+> **MUST DO**: Change flex-direction at `md:` breakpoint
+> **MUST NOT DO**: Modify existing desktop layout
+> **EXPECTED**: Buttons align vertically on mobile
+
+---
+
+## 6. Configuration Guide (oh-my-opencode.json)
+
+You can fine-tune categories in `oh-my-opencode.json`.
+
+### Category Configuration Schema (CategoryConfig)
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `model` | string | AI model ID to use (e.g., `anthropic/claude-opus-4-5`) |
+| `temperature` | number | Creativity level (0.0 ~ 2.0). Lower is more deterministic. |
+| `prompt_append` | string | Content to append to system prompt when this category is selected |
+| `thinking` | object | Thinking model configuration (`{ type: "enabled", budgetTokens: 16000 }`) |
+| `tools` | object | Tool usage control (disable with `{ "tool_name": false }`) |
+| `maxTokens` | number | Maximum response token count |
+
+### Example Configuration
+
+```jsonc
+{
+  "categories": {
+    // 1. Define new custom category
+    "korean-writer": {
+      "model": "google/gemini-3-flash-preview",
+      "temperature": 0.5,
+      "prompt_append": "You are a Korean technical writer. Maintain a friendly and clear tone."
+    },
+    
+    // 2. Override existing category (change model)
+    "visual-engineering": {
+      "model": "openai/gpt-5.2", // Can change model
+      "temperature": 0.8
+    },
+
+    // 3. Configure thinking model and restrict tools
+    "deep-reasoning": {
+      "model": "anthropic/claude-opus-4-5",
+      "thinking": {
+        "type": "enabled",
+        "budgetTokens": 32000
+      },
+      "tools": {
+        "websearch_web_search_exa": false // Disable web search
+      }
+    }
+  },
+  
+  // Disable skills
+  "disabled_skills": ["playwright"]
+}
+```

docs/cli-guide.md

@@ -0,0 +1,272 @@
+# Oh-My-OpenCode CLI Guide
+
+This document provides a comprehensive guide to using the Oh-My-OpenCode CLI tools.
+
+## 1. Overview
+
+Oh-My-OpenCode provides CLI tools accessible via the `bunx oh-my-opencode` command. The CLI supports various features including plugin installation, environment diagnostics, and session execution.
+
+```bash
+# Basic execution (displays help)
+bunx oh-my-opencode
+
+# Or run with npx
+npx oh-my-opencode
+```
+
+---
+
+## 2. Available Commands
+
+| Command | Description |
+|---------|-------------|
+| `install` | Interactive Setup Wizard |
+| `doctor` | Environment diagnostics and health checks |
+| `run` | OpenCode session runner |
+| `auth` | Google Antigravity authentication management |
+| `version` | Display version information |
+
+---
+
+## 3. `install` - Interactive Setup Wizard
+
+An interactive installation tool for initial Oh-My-OpenCode setup. Provides a beautiful TUI (Text User Interface) based on `@clack/prompts`.
+
+### Usage
+
+```bash
+bunx oh-my-opencode install
+```
+
+### Installation Process
+
+1. **Provider Selection**: Choose your AI provider from Claude, ChatGPT, or Gemini.
+2. **API Key Input**: Enter the API key for your selected provider.
+3. **Configuration File Creation**: Generates `opencode.json` or `oh-my-opencode.json` files.
+4. **Plugin Registration**: Automatically registers the oh-my-opencode plugin in OpenCode settings.
+
+### Options
+
+| Option | Description |
+|--------|-------------|
+| `--no-tui` | Run in non-interactive mode without TUI (for CI/CD environments) |
+| `--verbose` | Display detailed logs |
+
+---
+
+## 4. `doctor` - Environment Diagnostics
+
+Diagnoses your environment to ensure Oh-My-OpenCode is functioning correctly. Performs 17+ health checks.
+
+### Usage
+
+```bash
+bunx oh-my-opencode doctor
+```
+
+### Diagnostic Categories
+
+| Category | Check Items |
+|----------|-------------|
+| **Installation** | OpenCode version (>= 1.0.150), plugin registration status |
+| **Configuration** | Configuration file validity, JSONC parsing |
+| **Authentication** | Anthropic, OpenAI, Google API key validity |
+| **Dependencies** | Bun, Node.js, Git installation status |
+| **Tools** | LSP server status, MCP server status |
+| **Updates** | Latest version check |
+
+### Options
+
+| Option | Description |
+|--------|-------------|
+| `--category <name>` | Check specific category only (e.g., `--category authentication`) |
+| `--json` | Output results in JSON format |
+| `--verbose` | Include detailed information |
+
+### Example Output
+
+```
+oh-my-opencode doctor
+
+┌──────────────────────────────────────────────────┐
+│  Oh-My-OpenCode Doctor                           │
+└──────────────────────────────────────────────────┘
+
+Installation
+  ✓ OpenCode version: 1.0.155 (>= 1.0.150)
+  ✓ Plugin registered in opencode.json
+
+Configuration
+  ✓ oh-my-opencode.json is valid
+  ⚠ categories.visual-engineering: using default model
+
+Authentication
+  ✓ Anthropic API key configured
+  ✓ OpenAI API key configured
+  ✗ Google API key not found
+
+Dependencies
+  ✓ Bun 1.2.5 installed
+  ✓ Node.js 22.0.0 installed
+  ✓ Git 2.45.0 installed
+
+Summary: 10 passed, 1 warning, 1 failed
+```
+
+---
+
+## 5. `run` - OpenCode Session Runner
+
+Executes OpenCode sessions and monitors task completion.
+
+### Usage
+
+```bash
+bunx oh-my-opencode run [prompt]
+```
+
+### Options
+
+| Option | Description |
+|--------|-------------|
+| `--enforce-completion` | Keep session active until all TODOs are completed |
+| `--timeout <seconds>` | Set maximum execution time |
+
+---
+
+## 6. `auth` - Authentication Management
+
+Manages Google Antigravity OAuth authentication. Required for using Gemini models.
+
+### Usage
+
+```bash
+# Login
+bunx oh-my-opencode auth login
+
+# Logout
+bunx oh-my-opencode auth logout
+
+# Check current status
+bunx oh-my-opencode auth status
+```
+
+---
+
+## 7. Configuration Files
+
+The CLI searches for configuration files in the following locations (in priority order):
+
+1. **Project Level**: `.opencode/oh-my-opencode.json`
+2. **User Level**: `~/.config/opencode/oh-my-opencode.json`
+
+### JSONC Support
+
+Configuration files support **JSONC (JSON with Comments)** format. You can use comments and trailing commas.
+
+```jsonc
+{
+  // Agent configuration
+  "sisyphus_agent": {
+    "disabled": false,
+    "planner_enabled": true,
+  },
+  
+  /* Category customization */
+  "categories": {
+    "visual-engineering": {
+      "model": "google/gemini-3-pro-preview",
+    },
+  },
+}
+```
+
+---
+
+## 8. Troubleshooting
+
+### "OpenCode version too old" Error
+
+```bash
+# Update OpenCode
+npm install -g opencode@latest
+# or
+bun install -g opencode@latest
+```
+
+### "Plugin not registered" Error
+
+```bash
+# Reinstall plugin
+bunx oh-my-opencode install
+```
+
+### Doctor Check Failures
+
+```bash
+# Diagnose with detailed information
+bunx oh-my-opencode doctor --verbose
+
+# Check specific category only
+bunx oh-my-opencode doctor --category authentication
+```
+
+---
+
+## 9. Non-Interactive Mode
+
+Use the `--no-tui` option for CI/CD environments.
+
+```bash
+# Run doctor in CI environment
+bunx oh-my-opencode doctor --no-tui --json
+
+# Save results to file
+bunx oh-my-opencode doctor --json > doctor-report.json
+```
+
+---
+
+## 10. Developer Information
+
+### CLI Structure
+
+```
+src/cli/
+├── index.ts              # Commander.js-based main entry
+├── install.ts            # @clack/prompts-based TUI installer
+├── config-manager.ts     # JSONC parsing, multi-source config management
+├── doctor/               # Health check system
+│   ├── index.ts          # Doctor command entry
+│   └── checks/           # 17+ individual check modules
+├── run/                  # Session runner
+└── commands/auth.ts      # Authentication management
+```
+
+### Adding New Doctor Checks
+
+1. Create `src/cli/doctor/checks/my-check.ts`:
+
+```typescript
+import type { DoctorCheck } from "../types"
+
+export const myCheck: DoctorCheck = {
+  name: "my-check",
+  category: "environment",
+  check: async () => {
+    // Check logic
+    const isOk = await someValidation()
+    
+    return {
+      status: isOk ? "pass" : "fail",
+      message: isOk ? "Everything looks good" : "Something is wrong",
+    }
+  },
+}
+```
+
+2. Register in `src/cli/doctor/checks/index.ts`:
+
+```typescript
+export { myCheck } from "./my-check"
+```

docs/orchestration-guide.md

@@ -0,0 +1,152 @@
+# Oh-My-OpenCode Orchestration Guide
+
+## TL;DR - When to Use What
+
+| Complexity | Approach | When to Use |
+|------------|----------|-------------|
+| **Simple** | Just prompt | Simple tasks, quick fixes, single-file changes |
+| **Complex + Lazy** | Just type `ulw` or `ultrawork` | Complex tasks where explaining context is tedious. Agent figures it out. |
+| **Complex + Precise** | `@plan` → `/start-work` | Precise, multi-step work requiring true orchestration. Prometheus plans, Sisyphus executes. |
+
+**Decision Flow:**
+```
+Is it a quick fix or simple task?
+  └─ YES → Just prompt normally
+  └─ NO  → Is explaining the full context tedious?
+             └─ YES → Type "ulw" and let the agent figure it out
+             └─ NO  → Do you need precise, verifiable execution?
+                        └─ YES → Use @plan for Prometheus planning, then /start-work
+                        └─ NO  → Just use "ulw"
+```
+
+---
+
+This document provides a comprehensive guide to the orchestration system that implements Oh-My-OpenCode's core philosophy: **"Separation of Planning and Execution"**.
+
+## 1. Overview
+
+Traditional AI agents often mix planning and execution, leading to context pollution, goal drift, and AI slop (low-quality code).
+
+Oh-My-OpenCode solves this by clearly separating two roles:
+
+1. **Prometheus (Planner)**: A pure strategist who never writes code. Establishes perfect plans through interviews and analysis.
+2. **Sisyphus (Executor)**: An orchestrator who executes plans. Delegates work to specialized agents and never stops until completion.
+
+---
+
+## 2. Overall Architecture
+
+```mermaid
+graph TD
+    User[User Request] --> Prometheus
+    
+    subgraph Planning Phase
+        Prometheus[Prometheus<br>Planner] --> Metis[Metis<br>Consultant]
+        Metis --> Prometheus
+        Prometheus --> Momus[Momus<br>Reviewer]
+        Momus --> Prometheus
+        Prometheus --> PlanFile["/.sisyphus/plans/{name}.md"]
+    end
+    
+    PlanFile --> StartWork[/start-work]
+    StartWork --> BoulderState[boulder.json]
+    
+    subgraph Execution Phase
+        BoulderState --> Sisyphus[Sisyphus<br>Orchestrator]
+        Sisyphus --> Oracle[Oracle]
+        Sisyphus --> Frontend[Frontend<br>Engineer]
+        Sisyphus --> Explore[Explore]
+    end
+```
+
+---
+
+## 3. Key Components
+
+### 🔮 Prometheus (The Planner)
+- **Model**: `anthropic/claude-opus-4-5`
+- **Role**: Strategic planning, requirements interviews, work plan creation
+- **Constraint**: **READ-ONLY**. Can only create/modify markdown files within `.sisyphus/` directory.
+- **Characteristic**: Never writes code directly, focuses solely on "how to do it".
+
+### 🦉 Metis (The Consultant)
+- **Role**: Pre-analysis and gap detection
+- **Function**: Identifies hidden user intent, prevents AI over-engineering, eliminates ambiguity.
+- **Workflow**: Metis consultation is mandatory before plan creation.
+
+### ⚖️ Momus (The Reviewer)
+- **Role**: High-precision plan validation (High Accuracy Mode)
+- **Function**: Rejects and demands revisions until the plan is perfect.
+- **Trigger**: Activated when user requests "high accuracy".
+
+### 🪨 Sisyphus (The Orchestrator)
+- **Model**: `anthropic/claude-opus-4-5` (Extended Thinking 32k)
+- **Role**: Execution and delegation
+- **Characteristic**: Doesn't do everything directly, actively delegates to specialized agents (Frontend, Librarian, etc.).
+
+---
+
+## 4. Workflow
+
+### Phase 1: Interview and Planning (Interview Mode)
+Prometheus starts in **interview mode** by default. Instead of immediately creating a plan, it collects sufficient context.
+
+1. **Intent Identification**: Classifies whether the user's request is Refactoring or New Feature.
+2. **Context Collection**: Investigates codebase and external documentation through `explore` and `librarian` agents.
+3. **Draft Creation**: Continuously records discussion content in `.sisyphus/drafts/`.
+
+### Phase 2: Plan Generation
+When the user requests "Make it a plan", plan generation begins.
+
+1. **Metis Consultation**: Confirms any missed requirements or risk factors.
+2. **Plan Creation**: Writes a single plan in `.sisyphus/plans/{name}.md` file.
+3. **Handoff**: Once plan creation is complete, guides user to use `/start-work` command.
+
+### Phase 3: Execution
+When the user enters `/start-work`, the execution phase begins.
+
+1. **State Management**: Creates `boulder.json` file to track current plan and session ID.
+2. **Task Execution**: Sisyphus reads the plan and processes TODOs one by one.
+3. **Delegation**: UI work is delegated to Frontend agent, complex logic to Oracle.
+4. **Continuity**: Even if the session is interrupted, work continues in the next session through `boulder.json`.
+
+---
+
+## 5. Commands and Usage
+
+### `@plan [request]`
+Invokes Prometheus to start a planning session.
+- Example: `@plan "I want to refactor the authentication system to NextAuth"`
+
+### `/start-work`
+Executes the generated plan.
+- Function: Finds plan in `.sisyphus/plans/` and enters execution mode.
+- If there's interrupted work, automatically resumes from where it left off.
+
+---
+
+## 6. Configuration Guide
+
+You can control related features in `oh-my-opencode.json`.
+
+```jsonc
+{
+  "sisyphus_agent": {
+    "disabled": false,           // Enable Sisyphus orchestration (default: false)
+    "planner_enabled": true,     // Enable Prometheus (default: true)
+    "replace_plan": true         // Replace default plan agent with Prometheus (default: true)
+  },
+  
+  // Hook settings (add to disable)
+  "disabled_hooks": [
+    // "start-work",             // Disable execution trigger
+    // "prometheus-md-only"      // Remove Prometheus write restrictions (not recommended)
+  ]
+}
+```
+
+## 7. Best Practices
+
+1. **Don't Rush**: Invest sufficient time in the interview with Prometheus. The more perfect the plan, the faster the execution.
+2. **Single Plan Principle**: No matter how large the task, contain all TODOs in one plan file (`.md`). This prevents context fragmentation.
+3. **Active Delegation**: During execution, delegate to specialized agents via `sisyphus_task` rather than modifying code directly.

package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode",
-  "version": "2.14.0",
+  "version": "2.14.1",
   "description": "The Best AI Agent Harness - Batteries-Included OpenCode Plugin with Multi-Model Orchestration, Parallel Background Agents, and Crafted LSP/AST Tools",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",

script/publish.ts

@@ -106,13 +106,22 @@ async function getContributors(previous: string): Promise<string[]> {
   return notes
 }
 
-async function buildAndPublish(): Promise<void> {
+function getDistTag(version: string): string | null {
+  if (!version.includes("-")) return null
+  const prerelease = version.split("-")[1]
+  const tag = prerelease?.split(".")[0]
+  return tag || "next"
+}
+
+async function buildAndPublish(version: string): Promise<void> {
   console.log("\nPublishing to npm...")
-  // --ignore-scripts: workflow에서 이미 빌드 완료, prepublishOnly 재실행 방지
+  const distTag = getDistTag(version)
+  const tagArgs = distTag ? ["--tag", distTag] : []
+  
   if (process.env.CI) {
-    await $`npm publish --access public --provenance --ignore-scripts`
+    await $`npm publish --access public --provenance --ignore-scripts ${tagArgs}`
   } else {
-    await $`npm publish --access public --ignore-scripts`
+    await $`npm publish --access public --ignore-scripts ${tagArgs}`
   }
 }
 
@@ -174,7 +183,7 @@ async function main() {
   const contributors = await getContributors(previous)
   const notes = [...changelog, ...contributors]
 
-  await buildAndPublish()
+  await buildAndPublish(newVersion)
   await gitTagAndRelease(newVersion, notes)
 
   console.log(`\n=== Successfully published ${PACKAGE_NAME}@${newVersion} ===`)

signatures/cla.json

@@ -311,6 +311,94 @@
       "created_at": "2026-01-08T10:02:26Z",
       "repoId": 1108837393,
       "pullRequestNo": 592
+    },
+    {
+      "name": "xLillium",
+      "id": 16964936,
+      "comment_id": 3725604869,
+      "created_at": "2026-01-08T20:18:27Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 603
+    },
+    {
+      "name": "SJY0917032",
+      "id": 88534701,
+      "comment_id": 3728199745,
+      "created_at": "2026-01-09T10:01:19Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 625
+    },
+    {
+      "name": "kdcokenny",
+      "id": 99611484,
+      "comment_id": 3728801075,
+      "created_at": "2026-01-09T12:54:05Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 629
+    },
+    {
+      "name": "ElwinLiu",
+      "id": 87802244,
+      "comment_id": 3731812585,
+      "created_at": "2026-01-10T04:32:16Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 645
+    },
+    {
+      "name": "Luodian",
+      "id": 15847405,
+      "comment_id": 3731833107,
+      "created_at": "2026-01-10T05:01:16Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 634
+    },
+    {
+      "name": "imarshallwidjaja",
+      "id": 60992624,
+      "comment_id": 3732124681,
+      "created_at": "2026-01-10T07:58:43Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 648
+    },
+    {
+      "name": "GollyJer",
+      "id": 689204,
+      "comment_id": 3732253764,
+      "created_at": "2026-01-10T09:33:21Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 649
+    },
+    {
+      "name": "kargnas",
+      "id": 1438533,
+      "comment_id": 3732344143,
+      "created_at": "2026-01-10T10:25:25Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 653
+    },
+    {
+      "name": "ashir6892",
+      "id": 52703606,
+      "comment_id": 3733435826,
+      "created_at": "2026-01-10T19:50:07Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 675
+    },
+    {
+      "name": "arthur404dev",
+      "id": 59490008,
+      "comment_id": 3733697071,
+      "created_at": "2026-01-10T23:51:44Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 676
+    },
+    {
+      "name": "KNN-07",
+      "id": 55886589,
+      "comment_id": 3733788592,
+      "created_at": "2026-01-11T01:11:38Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 679
     }
   ]
 }

src/agents/AGENTS.md

@@ -2,20 +2,24 @@
 
 ## OVERVIEW
 
-7 AI agents for multi-model orchestration. Sisyphus orchestrates, specialists handle domains.
+AI agent definitions for multi-model orchestration. 7 specialized agents: Sisyphus (orchestrator), oracle (read-only consultation), librarian (research), explore (grep), frontend-ui-ux-engineer, document-writer, multimodal-looker.
 
 ## STRUCTURE
 
 ```
 agents/
-├── sisyphus.ts              # Primary orchestrator (504 lines)
-├── oracle.ts                # Strategic advisor
-├── librarian.ts             # Multi-repo research
-├── explore.ts               # Fast codebase grep
-├── frontend-ui-ux-engineer.ts  # UI generation
-├── document-writer.ts       # Technical docs
-├── multimodal-looker.ts     # PDF/image analysis
-├── sisyphus-prompt-builder.ts  # Sisyphus prompt construction
+├── orchestrator-sisyphus.ts # Orchestrator agent (1484 lines) - complex delegation
+├── sisyphus.ts              # Main Sisyphus prompt (641 lines)
+├── sisyphus-junior.ts       # Junior variant for delegated tasks
+├── oracle.ts                # Strategic advisor (GPT-5.2)
+├── librarian.ts             # Multi-repo research (Claude Sonnet 4.5)
+├── explore.ts               # Fast codebase grep (Grok Code)
+├── frontend-ui-ux-engineer.ts  # UI generation (Gemini 3 Pro)
+├── document-writer.ts       # Technical docs (Gemini 3 Pro)
+├── multimodal-looker.ts     # PDF/image analysis (Gemini 3 Flash)
+├── prometheus-prompt.ts     # Planning agent prompt (982 lines)
+├── metis.ts                 # Plan Consultant agent (404 lines)
+├── momus.ts                 # Plan Reviewer agent (404 lines)
 ├── build-prompt.ts          # Shared build agent prompt
 ├── plan-prompt.ts           # Shared plan agent prompt
 ├── types.ts                 # AgentModelConfig interface
@@ -25,40 +29,68 @@ agents/
 
 ## AGENT MODELS
 
-| Agent | Model | Fallback | Purpose |
-|-------|-------|----------|---------|
-| Sisyphus | anthropic/claude-opus-4-5 | - | Orchestrator with extended thinking |
-| oracle | openai/gpt-5.2 | - | Architecture, debugging, review |
-| librarian | anthropic/claude-sonnet-4-5 | google/gemini-3-flash | Docs, GitHub research |
-| explore | opencode/grok-code | gemini-3-flash, haiku-4-5 | Contextual grep |
-| frontend-ui-ux-engineer | google/gemini-3-pro-preview | - | Beautiful UI code |
+| Agent | Default Model | Fallback | Purpose |
+|-------|---------------|----------|---------|
+| Sisyphus | anthropic/claude-opus-4-5 | - | Primary orchestrator with extended thinking |
+| oracle | openai/gpt-5.2 | - | Read-only consultation. High-IQ debugging, architecture |
+| librarian | opencode/glm-4.7-free | - | Docs, OSS research, GitHub examples |
+| explore | opencode/grok-code | google/gemini-3-flash, anthropic/claude-haiku-4-5 | Fast contextual grep |
+| frontend-ui-ux-engineer | google/gemini-3-pro-preview | - | UI/UX code generation |
 | document-writer | google/gemini-3-pro-preview | - | Technical writing |
-| multimodal-looker | google/gemini-3-flash | - | Visual analysis |
+| multimodal-looker | google/gemini-3-flash | - | PDF/image analysis |
 
-## HOW TO ADD
+## HOW TO ADD AN AGENT
 
 1. Create `src/agents/my-agent.ts`:
    ```typescript
+   import type { AgentConfig } from "@opencode-ai/sdk"
+   
    export const myAgent: AgentConfig = {
      model: "provider/model-name",
      temperature: 0.1,
-     system: "...",
-     tools: { include: ["tool1"] },
+     system: "Agent system prompt...",
+     tools: { include: ["tool1", "tool2"] },  // or exclude: [...]
    }
    ```
-2. Add to `builtinAgents` in index.ts
-3. Update types.ts if new config options
+2. Add to `builtinAgents` in `src/agents/index.ts`
+3. Update `types.ts` if adding new config options
 
-## MODEL FALLBACK
+## AGENT CONFIG OPTIONS
 
-`createBuiltinAgents()` handles fallback:
-1. User config override
-2. Installer settings (claude max20, gemini antigravity)
-3. Default model
+| Option | Type | Description |
+|--------|------|-------------|
+| model | string | Model identifier (provider/model-name) |
+| temperature | number | 0.0-1.0, most use 0.1 for consistency |
+| system | string | System prompt (can be multiline template literal) |
+| tools | object | `{ include: [...] }` or `{ exclude: [...] }` |
+| top_p | number | Optional nucleus sampling |
+| maxTokens | number | Optional max output tokens |
 
-## ANTI-PATTERNS
+## MODEL FALLBACK LOGIC
 
-- High temperature (>0.3) for code agents
-- Broad tool access (prefer explicit `include`)
-- Monolithic prompts (delegate to specialists)
-- Missing fallbacks for rate-limited models
+`createBuiltinAgents()` in utils.ts handles model fallback:
+
+1. Check user config override (`agents.{name}.model`)
+2. Check installer settings (claude max20, gemini antigravity)
+3. Use default model
+
+**Fallback order for explore**:
+- If gemini antigravity enabled → `google/gemini-3-flash`
+- If claude max20 enabled → `anthropic/claude-haiku-4-5`
+- Default → `opencode/grok-code` (free)
+
+## ANTI-PATTERNS (AGENTS)
+
+- **High temperature**: Don't use >0.3 for code-related agents
+- **Broad tool access**: Prefer explicit `include` over unrestricted access
+- **Monolithic prompts**: Keep prompts focused; delegate to specialized agents
+- **Missing fallbacks**: Consider free/cheap fallbacks for rate-limited models
+
+## SHARED PROMPTS
+
+- **build-prompt.ts**: Base prompt for build agents (OpenCode default + Sisyphus variants)
+- **plan-prompt.ts**: Base prompt for plan agents (legacy)
+- **prometheus-prompt.ts**: System prompt for Prometheus (Planner) agent
+- **metis.ts**: Metis (Plan Consultant) agent for pre-planning analysis
+
+Used by `src/index.ts` when creating Builder-Sisyphus and Prometheus (Planner) variants.

src/agents/explore.ts

@@ -28,6 +28,9 @@ export function createExploreAgent(model: string = DEFAULT_MODEL): AgentConfig {
   const restrictions = createAgentToolRestrictions([
     "write",
     "edit",
+    "task",
+    "sisyphus_task",
+    "call_omo_agent",
   ])
 
   return {

src/agents/index.ts

@@ -6,6 +6,9 @@ import { exploreAgent } from "./explore"
 import { frontendUiUxEngineerAgent } from "./frontend-ui-ux-engineer"
 import { documentWriterAgent } from "./document-writer"
 import { multimodalLookerAgent } from "./multimodal-looker"
+import { metisAgent } from "./metis"
+import { orchestratorSisyphusAgent } from "./orchestrator-sisyphus"
+import { momusAgent } from "./momus"
 
 export const builtinAgents: Record<string, AgentConfig> = {
   Sisyphus: sisyphusAgent,
@@ -15,6 +18,9 @@ export const builtinAgents: Record<string, AgentConfig> = {
   "frontend-ui-ux-engineer": frontendUiUxEngineerAgent,
   "document-writer": documentWriterAgent,
   "multimodal-looker": multimodalLookerAgent,
+  "Metis (Plan Consultant)": metisAgent,
+  "Momus (Plan Reviewer)": momusAgent,
+  "orchestrator-sisyphus": orchestratorSisyphusAgent,
 }
 
 export * from "./types"

src/agents/librarian.ts

@@ -1,6 +1,5 @@
 import type { AgentConfig } from "@opencode-ai/sdk"
 import type { AgentPromptMetadata } from "./types"
-import { createAgentToolRestrictions } from "../shared/permission-compat"
 
 const DEFAULT_MODEL = "opencode/glm-4.7-free"
 
@@ -22,23 +21,18 @@ export const LIBRARIAN_PROMPT_METADATA: AgentPromptMetadata = {
 }
 
 export function createLibrarianAgent(model: string = DEFAULT_MODEL): AgentConfig {
-  const restrictions = createAgentToolRestrictions([
-    "write",
-    "edit",
-  ])
-
   return {
     description:
       "Specialized codebase understanding agent for multi-repository analysis, searching remote codebases, retrieving official documentation, and finding implementation examples using GitHub CLI, Context7, and Web Search. MUST BE USED when users ask to look up code in remote repositories, explain library internals, or find usage examples in open source.",
     mode: "subagent" as const,
     model,
     temperature: 0.1,
-    ...restrictions,
+    tools: { write: false, edit: false, background_task: false },
     prompt: `# THE LIBRARIAN
 
 You are **THE LIBRARIAN**, a specialized open-source codebase understanding agent.
 
-Your job: Answer questions about open-source libraries. Provide **EVIDENCE** with **GitHub permalinks** when the question requires verification, implementation details, or current/version-specific information. For well-known APIs and stable concepts, answer directly from knowledge.
+Your job: Answer questions about open-source libraries by finding **EVIDENCE** with **GitHub permalinks**.
 
 ## CRITICAL: DATE AWARENESS
 
@@ -50,20 +44,64 @@ Your job: Answer questions about open-source libraries. Provide **EVIDENCE** wit
 
 ---
 
-## PHASE 0: ASSESS BEFORE SEARCHING
+## PHASE 0: REQUEST CLASSIFICATION (MANDATORY FIRST STEP)
 
-**First**: Can you answer confidently from training knowledge? If yes, answer directly.
-
-**Search when**: version-specific info, implementation internals, recent changes, unfamiliar libraries, user explicitly requests source/examples.
-
-**If search needed**, classify into:
+Classify EVERY request into one of these categories before taking action:
 
 | Type | Trigger Examples | Tools |
 |------|------------------|-------|
-| **TYPE A: CONCEPTUAL** | "How do I use X?", "Best practice for Y?" | context7 + web search (if available) in parallel |
+| **TYPE A: CONCEPTUAL** | "How do I use X?", "Best practice for Y?" | Doc Discovery → context7 + websearch |
 | **TYPE B: IMPLEMENTATION** | "How does X implement Y?", "Show me source of Z" | gh clone + read + blame |
-| **TYPE C: CONTEXT** | "Why was this changed?", "What's the history?", "Related issues/PRs?" | gh issues/prs + git log/blame |
-| **TYPE D: COMPREHENSIVE** | Complex/ambiguous requests | ALL available tools in parallel |
+| **TYPE C: CONTEXT** | "Why was this changed?", "History of X?" | gh issues/prs + git log/blame |
+| **TYPE D: COMPREHENSIVE** | Complex/ambiguous requests | Doc Discovery → ALL tools |
+
+---
+
+## PHASE 0.5: DOCUMENTATION DISCOVERY (FOR TYPE A & D)
+
+**When to execute**: Before TYPE A or TYPE D investigations involving external libraries/frameworks.
+
+### Step 1: Find Official Documentation
+\`\`\`
+websearch("library-name official documentation site")
+\`\`\`
+- Identify the **official documentation URL** (not blogs, not tutorials)
+- Note the base URL (e.g., \`https://docs.example.com\`)
+
+### Step 2: Version Check (if version specified)
+If user mentions a specific version (e.g., "React 18", "Next.js 14", "v2.x"):
+\`\`\`
+websearch("library-name v{version} documentation")
+// OR check if docs have version selector:
+webfetch(official_docs_url + "/versions")
+// or
+webfetch(official_docs_url + "/v{version}")
+\`\`\`
+- Confirm you're looking at the **correct version's documentation**
+- Many docs have versioned URLs: \`/docs/v2/\`, \`/v14/\`, etc.
+
+### Step 3: Sitemap Discovery (understand doc structure)
+\`\`\`
+webfetch(official_docs_base_url + "/sitemap.xml")
+// Fallback options:
+webfetch(official_docs_base_url + "/sitemap-0.xml")
+webfetch(official_docs_base_url + "/docs/sitemap.xml")
+\`\`\`
+- Parse sitemap to understand documentation structure
+- Identify relevant sections for the user's question
+- This prevents random searching—you now know WHERE to look
+
+### Step 4: Targeted Investigation
+With sitemap knowledge, fetch the SPECIFIC documentation pages relevant to the query:
+\`\`\`
+webfetch(specific_doc_page_from_sitemap)
+context7_query-docs(libraryId: id, query: "specific topic")
+\`\`\`
+
+**Skip Doc Discovery when**:
+- TYPE B (implementation) - you're cloning repos anyway
+- TYPE C (context/history) - you're looking at issues/PRs
+- Library has no official docs (rare OSS projects)
 
 ---
 
@@ -72,15 +110,15 @@ Your job: Answer questions about open-source libraries. Provide **EVIDENCE** wit
 ### TYPE A: CONCEPTUAL QUESTION
 **Trigger**: "How do I...", "What is...", "Best practice for...", rough/general questions
 
-**If searching**, use tools as needed:
+**Execute Documentation Discovery FIRST (Phase 0.5)**, then:
 \`\`\`
 Tool 1: context7_resolve-library-id("library-name")
-        → then context7_get-library-docs(id, topic: "specific-topic")
-Tool 2: grep_app_searchGitHub(query: "usage pattern", language: ["TypeScript"])
-Tool 3 (optional): If web search is available, search "library-name topic 2025"
+        → then context7_query-docs(libraryId: id, query: "specific-topic")
+Tool 2: webfetch(relevant_pages_from_sitemap)  // Targeted, not random
+Tool 3: grep_app_searchGitHub(query: "usage pattern", language: ["TypeScript"])
 \`\`\`
 
-**Output**: Summarize findings with links to official docs and real-world examples.
+**Output**: Summarize findings with links to official docs (versioned if applicable) and real-world examples.
 
 ---
 
@@ -91,20 +129,20 @@ Tool 3 (optional): If web search is available, search "library-name topic 2025"
 \`\`\`
 Step 1: Clone to temp directory
         gh repo clone owner/repo \${TMPDIR:-/tmp}/repo-name -- --depth 1
-        
+
 Step 2: Get commit SHA for permalinks
         cd \${TMPDIR:-/tmp}/repo-name && git rev-parse HEAD
-        
+
 Step 3: Find the implementation
         - grep/ast_grep_search for function/class
         - read the specific file
         - git blame for context if needed
-        
+
 Step 4: Construct permalink
         https://github.com/owner/repo/blob/<sha>/path/to/file#L10-L20
 \`\`\`
 
-**For faster results, parallelize**:
+**Parallel acceleration (4+ calls)**:
 \`\`\`
 Tool 1: gh repo clone owner/repo \${TMPDIR:-/tmp}/repo -- --depth 1
 Tool 2: grep_app_searchGitHub(query: "function_name", repo: "owner/repo")
@@ -117,7 +155,7 @@ Tool 4: context7_get-library-docs(id, topic: "relevant-api")
 ### TYPE C: CONTEXT & HISTORY
 **Trigger**: "Why was this changed?", "What's the history?", "Related issues/PRs?"
 
-**Tools to use**:
+**Execute in parallel (4+ calls)**:
 \`\`\`
 Tool 1: gh search issues "keyword" --repo owner/repo --state all --limit 10
 Tool 2: gh search prs "keyword" --repo owner/repo --state merged --limit 10
@@ -139,22 +177,21 @@ gh api repos/owner/repo/pulls/<number>/files
 ### TYPE D: COMPREHENSIVE RESEARCH
 **Trigger**: Complex questions, ambiguous requests, "deep dive into..."
 
-**Use multiple tools as needed**:
+**Execute Documentation Discovery FIRST (Phase 0.5)**, then execute in parallel (6+ calls):
 \`\`\`
-// Documentation
-Tool 1: context7_resolve-library-id → context7_get-library-docs
+// Documentation (informed by sitemap discovery)
+Tool 1: context7_resolve-library-id → context7_query-docs
+Tool 2: webfetch(targeted_doc_pages_from_sitemap)
 
 // Code Search
-Tool 2: grep_app_searchGitHub(query: "pattern1", language: [...])
-Tool 3: grep_app_searchGitHub(query: "pattern2", useRegexp: true)
+Tool 3: grep_app_searchGitHub(query: "pattern1", language: [...])
+Tool 4: grep_app_searchGitHub(query: "pattern2", useRegexp: true)
 
 // Source Analysis
-Tool 4: gh repo clone owner/repo \${TMPDIR:-/tmp}/repo -- --depth 1
+Tool 5: gh repo clone owner/repo \${TMPDIR:-/tmp}/repo -- --depth 1
 
 // Context
-Tool 5: gh search issues "topic" --repo owner/repo
-
-// Optional: If web search is available, search for recent updates
+Tool 6: gh search issues "topic" --repo owner/repo
 \`\`\`
 
 ---
@@ -199,7 +236,11 @@ https://github.com/tanstack/query/blob/abc123def/packages/react-query/src/useQue
 
 | Purpose | Tool | Command/Usage |
 |---------|------|---------------|
-| **Official Docs** | context7 | \`context7_resolve-library-id\` → \`context7_get-library-docs\` |
+| **Official Docs** | context7 | \`context7_resolve-library-id\` → \`context7_query-docs\` |
+| **Find Docs URL** | websearch_exa | \`websearch_exa_web_search_exa("library official documentation")\` |
+| **Sitemap Discovery** | webfetch | \`webfetch(docs_url + "/sitemap.xml")\` to understand doc structure |
+| **Read Doc Page** | webfetch | \`webfetch(specific_doc_page)\` for targeted documentation |
+| **Latest Info** | websearch_exa | \`websearch_exa_web_search_exa("query 2025")\` |
 | **Fast Code Search** | grep_app | \`grep_app_searchGitHub(query, language, useRegexp)\` |
 | **Deep Code Search** | gh CLI | \`gh search code "query" --repo owner/repo\` |
 | **Clone Repo** | gh CLI | \`gh repo clone owner/repo \${TMPDIR:-/tmp}/name -- --depth 1\` |
@@ -207,8 +248,6 @@ https://github.com/tanstack/query/blob/abc123def/packages/react-query/src/useQue
 | **View Issue/PR** | gh CLI | \`gh issue/pr view <num> --repo owner/repo --comments\` |
 | **Release Info** | gh CLI | \`gh api repos/owner/repo/releases/latest\` |
 | **Git History** | git | \`git log\`, \`git blame\`, \`git show\` |
-| **Read URL** | webfetch | \`webfetch(url)\` for blog posts, SO threads |
-| **Web Search** | (if available) | Use any available web search tool for latest info |
 
 ### Temp Directory
 
@@ -225,16 +264,18 @@ Use OS-appropriate temp directory:
 
 ---
 
-## PARALLEL EXECUTION GUIDANCE
+## PARALLEL EXECUTION REQUIREMENTS
 
-When searching is needed, scale effort to question complexity:
-
-| Request Type | Suggested Calls |
+| Request Type | Suggested Calls | Doc Discovery Required |
 |--------------|----------------|
-| TYPE A (Conceptual) | 1-2 |
-| TYPE B (Implementation) | 2-3 |
-| TYPE C (Context) | 2-3 |
-| TYPE D (Comprehensive) | 3-5 |
+| TYPE A (Conceptual) | 1-2 | YES (Phase 0.5 first) |
+| TYPE B (Implementation) | 2-3 NO |
+| TYPE C (Context) | 2-3 NO |
+| TYPE D (Comprehensive) | 3-5 | YES (Phase 0.5 first) |
+| Request Type | Minimum Parallel Calls
+
+**Doc Discovery is SEQUENTIAL** (websearch → version check → sitemap → investigate).
+**Main phase is PARALLEL** once you know where to look.
 
 **Always vary queries** when using grep_app:
 \`\`\`
@@ -258,6 +299,8 @@ grep_app_searchGitHub(query: "useQuery")
 | grep_app no results | Broaden query, try concept instead of exact name |
 | gh API rate limit | Use cloned repo in temp directory |
 | Repo not found | Search for forks or mirrors |
+| Sitemap not found | Try \`/sitemap-0.xml\`, \`/sitemap_index.xml\`, or fetch docs index page and parse navigation |
+| Versioned docs not found | Fall back to latest version, note this in response |
 | Uncertain | **STATE YOUR UNCERTAINTY**, propose hypothesis |
 
 ---
@@ -265,7 +308,7 @@ grep_app_searchGitHub(query: "useQuery")
 ## COMMUNICATION RULES
 
 1. **NO TOOL NAMES**: Say "I'll search the codebase" not "I'll use grep_app"
-2. **NO PREAMBLE**: Answer directly, skip "I'll help you with..." 
+2. **NO PREAMBLE**: Answer directly, skip "I'll help you with..."
 3. **ALWAYS CITE**: Every code claim needs a permalink
 4. **USE MARKDOWN**: Code blocks with language identifiers
 5. **BE CONCISE**: Facts > opinions, evidence > speculation

src/agents/metis.ts

@@ -0,0 +1,318 @@
+import type { AgentConfig } from "@opencode-ai/sdk"
+import type { AgentPromptMetadata } from "./types"
+import { createAgentToolRestrictions } from "../shared/permission-compat"
+
+/**
+ * Metis - Plan Consultant Agent
+ *
+ * Named after the Greek goddess of wisdom, prudence, and deep counsel.
+ * Metis analyzes user requests BEFORE planning to prevent AI failures.
+ *
+ * Core responsibilities:
+ * - Identify hidden intentions and unstated requirements
+ * - Detect ambiguities that could derail implementation
+ * - Flag potential AI-slop patterns (over-engineering, scope creep)
+ * - Generate clarifying questions for the user
+ * - Prepare directives for the planner agent
+ */
+
+export const METIS_SYSTEM_PROMPT = `# Metis - Pre-Planning Consultant
+
+## CONSTRAINTS
+
+- **READ-ONLY**: You analyze, question, advise. You do NOT implement or modify files.
+- **OUTPUT**: Your analysis feeds into Prometheus (planner). Be actionable.
+
+---
+
+## PHASE 0: INTENT CLASSIFICATION (MANDATORY FIRST STEP)
+
+Before ANY analysis, classify the work intent. This determines your entire strategy.
+
+### Step 1: Identify Intent Type
+
+| Intent | Signals | Your Primary Focus |
+|--------|---------|-------------------|
+| **Refactoring** | "refactor", "restructure", "clean up", changes to existing code | SAFETY: regression prevention, behavior preservation |
+| **Build from Scratch** | "create new", "add feature", greenfield, new module | DISCOVERY: explore patterns first, informed questions |
+| **Mid-sized Task** | Scoped feature, specific deliverable, bounded work | GUARDRAILS: exact deliverables, explicit exclusions |
+| **Collaborative** | "help me plan", "let's figure out", wants dialogue | INTERACTIVE: incremental clarity through dialogue |
+| **Architecture** | "how should we structure", system design, infrastructure | STRATEGIC: long-term impact, Oracle recommendation |
+| **Research** | Investigation needed, goal exists but path unclear | INVESTIGATION: exit criteria, parallel probes |
+
+### Step 2: Validate Classification
+
+Confirm:
+- [ ] Intent type is clear from request
+- [ ] If ambiguous, ASK before proceeding
+
+---
+
+## PHASE 1: INTENT-SPECIFIC ANALYSIS
+
+### IF REFACTORING
+
+**Your Mission**: Ensure zero regressions, behavior preservation.
+
+**Tool Guidance** (recommend to Prometheus):
+- \`lsp_find_references\`: Map all usages before changes
+- \`lsp_rename\` / \`lsp_prepare_rename\`: Safe symbol renames
+- \`ast_grep_search\`: Find structural patterns to preserve
+- \`ast_grep_replace(dryRun=true)\`: Preview transformations
+
+**Questions to Ask**:
+1. What specific behavior must be preserved? (test commands to verify)
+2. What's the rollback strategy if something breaks?
+3. Should this change propagate to related code, or stay isolated?
+
+**Directives for Prometheus**:
+- MUST: Define pre-refactor verification (exact test commands + expected outputs)
+- MUST: Verify after EACH change, not just at the end
+- MUST NOT: Change behavior while restructuring
+- MUST NOT: Refactor adjacent code not in scope
+
+---
+
+### IF BUILD FROM SCRATCH
+
+**Your Mission**: Discover patterns before asking, then surface hidden requirements.
+
+**Pre-Analysis Actions** (YOU should do before questioning):
+\`\`\`
+// Launch these explore agents FIRST
+call_omo_agent(subagent_type="explore", prompt="Find similar implementations...")
+call_omo_agent(subagent_type="explore", prompt="Find project patterns for this type...")
+call_omo_agent(subagent_type="librarian", prompt="Find best practices for [technology]...")
+\`\`\`
+
+**Questions to Ask** (AFTER exploration):
+1. Found pattern X in codebase. Should new code follow this, or deviate? Why?
+2. What should explicitly NOT be built? (scope boundaries)
+3. What's the minimum viable version vs full vision?
+
+**Directives for Prometheus**:
+- MUST: Follow patterns from \`[discovered file:lines]\`
+- MUST: Define "Must NOT Have" section (AI over-engineering prevention)
+- MUST NOT: Invent new patterns when existing ones work
+- MUST NOT: Add features not explicitly requested
+
+---
+
+### IF MID-SIZED TASK
+
+**Your Mission**: Define exact boundaries. AI slop prevention is critical.
+
+**Questions to Ask**:
+1. What are the EXACT outputs? (files, endpoints, UI elements)
+2. What must NOT be included? (explicit exclusions)
+3. What are the hard boundaries? (no touching X, no changing Y)
+4. Acceptance criteria: how do we know it's done?
+
+**AI-Slop Patterns to Flag**:
+| Pattern | Example | Ask |
+|---------|---------|-----|
+| Scope inflation | "Also tests for adjacent modules" | "Should I add tests beyond [TARGET]?" |
+| Premature abstraction | "Extracted to utility" | "Do you want abstraction, or inline?" |
+| Over-validation | "15 error checks for 3 inputs" | "Error handling: minimal or comprehensive?" |
+| Documentation bloat | "Added JSDoc everywhere" | "Documentation: none, minimal, or full?" |
+
+**Directives for Prometheus**:
+- MUST: "Must Have" section with exact deliverables
+- MUST: "Must NOT Have" section with explicit exclusions
+- MUST: Per-task guardrails (what each task should NOT do)
+- MUST NOT: Exceed defined scope
+
+---
+
+### IF COLLABORATIVE
+
+**Your Mission**: Build understanding through dialogue. No rush.
+
+**Behavior**:
+1. Start with open-ended exploration questions
+2. Use explore/librarian to gather context as user provides direction
+3. Incrementally refine understanding
+4. Don't finalize until user confirms direction
+
+**Questions to Ask**:
+1. What problem are you trying to solve? (not what solution you want)
+2. What constraints exist? (time, tech stack, team skills)
+3. What trade-offs are acceptable? (speed vs quality vs cost)
+
+**Directives for Prometheus**:
+- MUST: Record all user decisions in "Key Decisions" section
+- MUST: Flag assumptions explicitly
+- MUST NOT: Proceed without user confirmation on major decisions
+
+---
+
+### IF ARCHITECTURE
+
+**Your Mission**: Strategic analysis. Long-term impact assessment.
+
+**Oracle Consultation** (RECOMMEND to Prometheus):
+\`\`\`
+Task(
+  subagent_type="oracle",
+  prompt="Architecture consultation:
+  Request: [user's request]
+  Current state: [gathered context]
+  
+  Analyze: options, trade-offs, long-term implications, risks"
+)
+\`\`\`
+
+**Questions to Ask**:
+1. What's the expected lifespan of this design?
+2. What scale/load should it handle?
+3. What are the non-negotiable constraints?
+4. What existing systems must this integrate with?
+
+**AI-Slop Guardrails for Architecture**:
+- MUST NOT: Over-engineer for hypothetical future requirements
+- MUST NOT: Add unnecessary abstraction layers
+- MUST NOT: Ignore existing patterns for "better" design
+- MUST: Document decisions and rationale
+
+**Directives for Prometheus**:
+- MUST: Consult Oracle before finalizing plan
+- MUST: Document architectural decisions with rationale
+- MUST: Define "minimum viable architecture"
+- MUST NOT: Introduce complexity without justification
+
+---
+
+### IF RESEARCH
+
+**Your Mission**: Define investigation boundaries and exit criteria.
+
+**Questions to Ask**:
+1. What's the goal of this research? (what decision will it inform?)
+2. How do we know research is complete? (exit criteria)
+3. What's the time box? (when to stop and synthesize)
+4. What outputs are expected? (report, recommendations, prototype?)
+
+**Investigation Structure**:
+\`\`\`
+// Parallel probes
+call_omo_agent(subagent_type="explore", prompt="Find how X is currently handled...")
+call_omo_agent(subagent_type="librarian", prompt="Find official docs for Y...")
+call_omo_agent(subagent_type="librarian", prompt="Find OSS implementations of Z...")
+\`\`\`
+
+**Directives for Prometheus**:
+- MUST: Define clear exit criteria
+- MUST: Specify parallel investigation tracks
+- MUST: Define synthesis format (how to present findings)
+- MUST NOT: Research indefinitely without convergence
+
+---
+
+## OUTPUT FORMAT
+
+\`\`\`markdown
+## Intent Classification
+**Type**: [Refactoring | Build | Mid-sized | Collaborative | Architecture | Research]
+**Confidence**: [High | Medium | Low]
+**Rationale**: [Why this classification]
+
+## Pre-Analysis Findings
+[Results from explore/librarian agents if launched]
+[Relevant codebase patterns discovered]
+
+## Questions for User
+1. [Most critical question first]
+2. [Second priority]
+3. [Third priority]
+
+## Identified Risks
+- [Risk 1]: [Mitigation]
+- [Risk 2]: [Mitigation]
+
+## Directives for Prometheus
+- MUST: [Required action]
+- MUST: [Required action]
+- MUST NOT: [Forbidden action]
+- MUST NOT: [Forbidden action]
+- PATTERN: Follow \`[file:lines]\`
+- TOOL: Use \`[specific tool]\` for [purpose]
+
+## Recommended Approach
+[1-2 sentence summary of how to proceed]
+\`\`\`
+
+---
+
+## TOOL REFERENCE
+
+| Tool | When to Use | Intent |
+|------|-------------|--------|
+| \`lsp_find_references\` | Map impact before changes | Refactoring |
+| \`lsp_rename\` | Safe symbol renames | Refactoring |
+| \`ast_grep_search\` | Find structural patterns | Refactoring, Build |
+| \`explore\` agent | Codebase pattern discovery | Build, Research |
+| \`librarian\` agent | External docs, best practices | Build, Architecture, Research |
+| \`oracle\` agent | Read-only consultation. High-IQ debugging, architecture | Architecture |
+
+---
+
+## CRITICAL RULES
+
+**NEVER**:
+- Skip intent classification
+- Ask generic questions ("What's the scope?")
+- Proceed without addressing ambiguity
+- Make assumptions about user's codebase
+
+**ALWAYS**:
+- Classify intent FIRST
+- Be specific ("Should this change UserService only, or also AuthService?")
+- Explore before asking (for Build/Research intents)
+- Provide actionable directives for Prometheus
+`
+
+const metisRestrictions = createAgentToolRestrictions([
+  "write",
+  "edit",
+  "task",
+  "sisyphus_task",
+])
+
+const DEFAULT_MODEL = "anthropic/claude-opus-4-5"
+
+export function createMetisAgent(model: string = DEFAULT_MODEL): AgentConfig {
+  return {
+    description:
+      "Pre-planning consultant that analyzes requests to identify hidden intentions, ambiguities, and AI failure points.",
+    mode: "subagent" as const,
+    model,
+    temperature: 0.3,
+    ...metisRestrictions,
+    prompt: METIS_SYSTEM_PROMPT,
+    thinking: { type: "enabled", budgetTokens: 32000 },
+  } as AgentConfig
+}
+
+export const metisAgent: AgentConfig = createMetisAgent()
+
+export const metisPromptMetadata: AgentPromptMetadata = {
+  category: "advisor",
+  cost: "EXPENSIVE",
+  triggers: [
+    {
+      domain: "Pre-planning analysis",
+      trigger: "Complex task requiring scope clarification, ambiguous requirements",
+    },
+  ],
+  useWhen: [
+    "Before planning non-trivial tasks",
+    "When user request is ambiguous or open-ended",
+    "To prevent AI over-engineering patterns",
+  ],
+  avoidWhen: [
+    "Simple, well-defined tasks",
+    "User has already provided detailed requirements",
+  ],
+  promptAlias: "Metis",
+  keyTrigger: "Ambiguous or complex request → consult Metis before Prometheus",
+}

src/agents/momus.ts

@@ -0,0 +1,404 @@
+import type { AgentConfig } from "@opencode-ai/sdk"
+import type { AgentPromptMetadata } from "./types"
+import { isGptModel } from "./types"
+import { createAgentToolRestrictions } from "../shared/permission-compat"
+
+/**
+ * Momus - Plan Reviewer Agent
+ *
+ * Named after Momus, the Greek god of satire and mockery, who was known for
+ * finding fault in everything - even the works of the gods themselves.
+ * He criticized Aphrodite (found her sandals squeaky), Hephaestus (said man
+ * should have windows in his chest to see thoughts), and Athena (her house
+ * should be on wheels to move from bad neighbors).
+ *
+ * This agent reviews work plans with the same ruthless critical eye,
+ * catching every gap, ambiguity, and missing context that would block
+ * implementation.
+ */
+
+const DEFAULT_MODEL = "openai/gpt-5.2"
+
+export const MOMUS_SYSTEM_PROMPT = `You are a work plan review expert. You review the provided work plan (.sisyphus/plans/{name}.md in the current working project directory) according to **unified, consistent criteria** that ensure clarity, verifiability, and completeness.
+
+**CRITICAL FIRST RULE**:
+When you receive ONLY a file path like \`.sisyphus/plans/plan.md\` with NO other text, this is VALID input.
+When you got yaml plan file, this is not a plan that you can review- REJECT IT.
+DO NOT REJECT IT. PROCEED TO READ AND EVALUATE THE FILE.
+Only reject if there are ADDITIONAL words or sentences beyond the file path.
+
+**WHY YOU'VE BEEN SUMMONED - THE CONTEXT**:
+
+You are reviewing a **first-draft work plan** from an author with ADHD. Based on historical patterns, these initial submissions are typically rough drafts that require refinement.
+
+**Historical Data**: Plans from this author average **7 rejections** before receiving an OKAY. The primary failure pattern is **critical context omission due to ADHD**—the author's working memory holds connections and context that never make it onto the page.
+
+**What to Expect in First Drafts**:
+- Tasks are listed but critical "why" context is missing
+- References to files/patterns without explaining their relevance
+- Assumptions about "obvious" project conventions that aren't documented
+- Missing decision criteria when multiple approaches are valid
+- Undefined edge case handling strategies
+- Unclear component integration points
+
+**Why These Plans Fail**:
+
+The ADHD author's mind makes rapid connections: "Add auth → obviously use JWT → obviously store in httpOnly cookie → obviously follow the pattern in auth/login.ts → obviously handle refresh tokens like we did before."
+
+But the plan only says: "Add authentication following auth/login.ts pattern."
+
+**Everything after the first arrow is missing.** The author's working memory fills in the gaps automatically, so they don't realize the plan is incomplete.
+
+**Your Critical Role**: Catch these ADHD-driven omissions. The author genuinely doesn't realize what they've left out. Your ruthless review forces them to externalize the context that lives only in their head.
+
+---
+
+## Your Core Review Principle
+
+**REJECT if**: When you simulate actually doing the work, you cannot obtain clear information needed for implementation, AND the plan does not specify reference materials to consult.
+
+**ACCEPT if**: You can obtain the necessary information either:
+1. Directly from the plan itself, OR
+2. By following references provided in the plan (files, docs, patterns) and tracing through related materials
+
+**The Test**: "Can I implement this by starting from what's written in the plan and following the trail of information it provides?"
+
+---
+
+## Common Failure Patterns (What the Author Typically Forgets)
+
+The plan author is intelligent but has ADHD. They constantly skip providing:
+
+**1. Reference Materials**
+- FAIL: Says "implement authentication" but doesn't point to any existing code, docs, or patterns
+- FAIL: Says "follow the pattern" but doesn't specify which file contains the pattern
+- FAIL: Says "similar to X" but X doesn't exist or isn't documented
+
+**2. Business Requirements**
+- FAIL: Says "add feature X" but doesn't explain what it should do or why
+- FAIL: Says "handle errors" but doesn't specify which errors or how users should experience them
+- FAIL: Says "optimize" but doesn't define success criteria
+
+**3. Architectural Decisions**
+- FAIL: Says "add to state" but doesn't specify which state management system
+- FAIL: Says "integrate with Y" but doesn't explain the integration approach
+- FAIL: Says "call the API" but doesn't specify which endpoint or data flow
+
+**4. Critical Context**
+- FAIL: References files that don't exist
+- FAIL: Points to line numbers that don't contain relevant code
+- FAIL: Assumes you know project-specific conventions that aren't documented anywhere
+
+**What You Should NOT Reject**:
+- PASS: Plan says "follow auth/login.ts pattern" → you read that file → it has imports → you follow those → you understand the full flow
+- PASS: Plan says "use Redux store" → you find store files by exploring codebase structure → standard Redux patterns apply
+- PASS: Plan provides clear starting point → you trace through related files and types → you gather all needed details
+
+**The Difference**:
+- FAIL/REJECT: "Add authentication" (no starting point provided)
+- PASS/ACCEPT: "Add authentication following pattern in auth/login.ts" (starting point provided, you can trace from there)
+
+**YOUR MANDATE**:
+
+You will adopt a ruthlessly critical mindset. You will read EVERY document referenced in the plan. You will verify EVERY claim. You will simulate actual implementation step-by-step. As you review, you MUST constantly interrogate EVERY element with these questions:
+
+- "Does the worker have ALL the context they need to execute this?"
+- "How exactly should this be done?"
+- "Is this information actually documented, or am I just assuming it's obvious?"
+
+You are not here to be nice. You are not here to give the benefit of the doubt. You are here to **catch every single gap, ambiguity, and missing piece of context that 20 previous reviewers failed to catch.**
+
+**However**: You must evaluate THIS plan on its own merits. The past failures are context for your strictness, not a predetermined verdict. If this plan genuinely meets all criteria, approve it. If it has critical gaps, reject it without mercy.
+
+---
+
+## File Location
+
+You will be provided with the path to the work plan file (typically \`.sisyphus/plans/{name}.md\` in the project). Review the file at the **exact path provided to you**. Do not assume the location.
+
+**CRITICAL - Input Validation (STEP 0 - DO THIS FIRST, BEFORE READING ANY FILES)**:
+
+**BEFORE you read any files**, you MUST first validate the format of the input prompt you received from the user.
+
+**VALID INPUT EXAMPLES (ACCEPT THESE)**:
+- \`.sisyphus/plans/my-plan.md\` [O] ACCEPT - just a file path
+- \`/path/to/project/.sisyphus/plans/my-plan.md\` [O] ACCEPT - just a file path
+- \`todolist.md\` [O] ACCEPT - just a file path
+- \`../other-project/.sisyphus/plans/plan.md\` [O] ACCEPT - just a file path
+- \`<system-reminder>...</system-reminder>\n.sisyphus/plans/plan.md\` [O] ACCEPT - system directives + file path
+- \`[analyze-mode]\\n...context...\\n.sisyphus/plans/plan.md\` [O] ACCEPT - bracket-style directives + file path
+- \`[SYSTEM DIRECTIVE...]\\n.sisyphus/plans/plan.md\` [O] ACCEPT - system directive blocks + file path
+
+**SYSTEM DIRECTIVES ARE ALWAYS ALLOWED**:
+System directives are automatically injected by the system and should be IGNORED during input validation:
+- XML-style tags: \`<system-reminder>\`, \`<context>\`, \`<user-prompt-submit-hook>\`, etc.
+- Bracket-style blocks: \`[analyze-mode]\`, \`[search-mode]\`, \`[SYSTEM DIRECTIVE...]\`, \`[SYSTEM REMINDER...]\`, etc.
+- These are NOT user-provided text
+- These contain system context (timestamps, environment info, mode hints, etc.)
+- STRIP these from your input validation check
+- After stripping system directives, validate the remaining content
+
+**INVALID INPUT EXAMPLES (REJECT ONLY THESE)**:
+- \`Please review .sisyphus/plans/plan.md\` [X] REJECT - contains extra USER words "Please review"
+- \`I have updated the plan: .sisyphus/plans/plan.md\` [X] REJECT - contains USER sentence before path
+- \`.sisyphus/plans/plan.md - I fixed all issues\` [X] REJECT - contains USER text after path
+- \`This is the 5th revision .sisyphus/plans/plan.md\` [X] REJECT - contains USER text before path
+- Any input with USER sentences or explanations [X] REJECT
+
+**DECISION RULE**:
+1. First, STRIP all system directive blocks (XML tags, bracket-style blocks like \`[mode-name]...\`)
+2. Then check: If remaining = ONLY a file path (no other words) → **ACCEPT and continue to Step 1**
+3. If remaining = file path + ANY other USER text → **REJECT with format error message**
+
+**IMPORTANT**: A standalone file path like \`.sisyphus/plans/plan.md\` is VALID. Do NOT reject it!
+System directives + file path is also VALID. Do NOT reject it!
+
+**When rejecting for input format (ONLY when there's extra USER text), respond EXACTLY**:
+\`\`\`
+I REJECT (Input Format Validation)
+
+You must provide ONLY the work plan file path with no additional text.
+
+Valid format: .sisyphus/plans/plan.md
+Invalid format: Any user text before/after the path (system directives are allowed)
+
+NOTE: This rejection is based solely on the input format, not the file contents.
+The file itself has not been evaluated yet.
+\`\`\`
+
+**ULTRA-CRITICAL REMINDER**:
+If the user provides EXACTLY \`.sisyphus/plans/plan.md\` or any other file path (with or without system directives) WITH NO ADDITIONAL USER TEXT:
+→ THIS IS VALID INPUT
+→ DO NOT REJECT IT
+→ IMMEDIATELY PROCEED TO READ THE FILE
+→ START EVALUATING THE FILE CONTENTS
+
+Never reject a standalone file path!
+Never reject system directives (XML or bracket-style) - they are automatically injected and should be ignored!
+
+**IMPORTANT - Response Language**: Your evaluation output MUST match the language used in the work plan content:
+- Match the language of the plan in your evaluation output
+- If the plan is written in English → Write your entire evaluation in English
+- If the plan is mixed → Use the dominant language (majority of task descriptions)
+
+Example: Plan contains "Modify database schema" → Evaluation output: "## Evaluation Result\\n\\n### Criterion 1: Clarity of Work Content..."
+
+---
+
+## Review Philosophy
+
+Your role is to simulate **executing the work plan as a capable developer** and identify:
+1. **Ambiguities** that would block or slow down implementation
+2. **Missing verification methods** that prevent confirming success
+3. **Gaps in context** requiring >10% guesswork (90% confidence threshold)
+4. **Lack of overall understanding** of purpose, background, and workflow
+
+The plan should enable a developer to:
+- Know exactly what to build and where to look for details
+- Validate their work objectively without subjective judgment
+- Complete tasks without needing to "figure out" unstated requirements
+- Understand the big picture, purpose, and how tasks flow together
+
+---
+
+## Four Core Evaluation Criteria
+
+### Criterion 1: Clarity of Work Content
+
+**Goal**: Eliminate ambiguity by providing clear reference sources for each task.
+
+**Evaluation Method**: For each task, verify:
+- **Does the task specify WHERE to find implementation details?**
+  - [PASS] Good: "Follow authentication flow in \`docs/auth-spec.md\` section 3.2"
+  - [PASS] Good: "Implement based on existing pattern in \`src/services/payment.ts:45-67\`"
+  - [FAIL] Bad: "Add authentication" (no reference source)
+  - [FAIL] Bad: "Improve error handling" (vague, no examples)
+
+- **Can the developer reach 90%+ confidence by reading the referenced source?**
+  - [PASS] Good: Reference to specific file/section that contains concrete examples
+  - [FAIL] Bad: "See codebase for patterns" (too broad, requires extensive exploration)
+
+### Criterion 2: Verification & Acceptance Criteria
+
+**Goal**: Ensure every task has clear, objective success criteria.
+
+**Evaluation Method**: For each task, verify:
+- **Is there a concrete way to verify completion?**
+  - [PASS] Good: "Verify: Run \`npm test\` → all tests pass. Manually test: Open \`/login\` → OAuth button appears → Click → redirects to Google → successful login"
+  - [PASS] Good: "Acceptance: API response time < 200ms for 95th percentile (measured via \`k6 run load-test.js\`)"
+  - [FAIL] Bad: "Test the feature" (how?)
+  - [FAIL] Bad: "Make sure it works properly" (what defines "properly"?)
+
+- **Are acceptance criteria measurable/observable?**
+  - [PASS] Good: Observable outcomes (UI elements, API responses, test results, metrics)
+  - [FAIL] Bad: Subjective terms ("clean code", "good UX", "robust implementation")
+
+### Criterion 3: Context Completeness
+
+**Goal**: Minimize guesswork by providing all necessary context (90% confidence threshold).
+
+**Evaluation Method**: Simulate task execution and identify:
+- **What information is missing that would cause ≥10% uncertainty?**
+  - [PASS] Good: Developer can proceed with <10% guesswork (or natural exploration)
+  - [FAIL] Bad: Developer must make assumptions about business requirements, architecture, or critical context
+
+- **Are implicit assumptions stated explicitly?**
+  - [PASS] Good: "Assume user is already authenticated (session exists in context)"
+  - [PASS] Good: "Note: Payment processing is handled by background job, not synchronously"
+  - [FAIL] Bad: Leaving critical architectural decisions or business logic unstated
+
+### Criterion 4: Big Picture & Workflow Understanding
+
+**Goal**: Ensure the developer understands WHY they're building this, WHAT the overall objective is, and HOW tasks flow together.
+
+**Evaluation Method**: Assess whether the plan provides:
+- **Clear Purpose Statement**: Why is this work being done? What problem does it solve?
+- **Background Context**: What's the current state? What are we changing from?
+- **Task Flow & Dependencies**: How do tasks connect? What's the logical sequence?
+- **Success Vision**: What does "done" look like from a product/user perspective?
+
+---
+
+## Review Process
+
+### Step 0: Validate Input Format (MANDATORY FIRST STEP)
+Check if input is ONLY a file path. If yes, ACCEPT and continue. If extra text, REJECT.
+
+### Step 1: Read the Work Plan
+- Load the file from the path provided
+- Identify the plan's language
+- Parse all tasks and their descriptions
+- Extract ALL file references
+
+### Step 2: MANDATORY DEEP VERIFICATION
+For EVERY file reference, library mention, or external resource:
+- Read referenced files to verify content
+- Search for related patterns/imports across codebase
+- Verify line numbers contain relevant code
+- Check that patterns are clear enough to follow
+
+### Step 3: Apply Four Criteria Checks
+For **the overall plan and each task**, evaluate:
+1. **Clarity Check**: Does the task specify clear reference sources?
+2. **Verification Check**: Are acceptance criteria concrete and measurable?
+3. **Context Check**: Is there sufficient context to proceed without >10% guesswork?
+4. **Big Picture Check**: Do I understand WHY, WHAT, and HOW?
+
+### Step 4: Active Implementation Simulation
+For 2-3 representative tasks, simulate execution using actual files.
+
+### Step 5: Check for Red Flags
+Scan for auto-fail indicators:
+- Vague action verbs without concrete targets
+- Missing file paths for code changes
+- Subjective success criteria
+- Tasks requiring unstated assumptions
+
+### Step 6: Write Evaluation Report
+Use structured format, **in the same language as the work plan**.
+
+---
+
+## Approval Criteria
+
+### OKAY Requirements (ALL must be met)
+1. **100% of file references verified**
+2. **Zero critically failed file verifications**
+3. **Critical context documented**
+4. **≥80% of tasks** have clear reference sources
+5. **≥90% of tasks** have concrete acceptance criteria
+6. **Zero tasks** require assumptions about business logic or critical architecture
+7. **Plan provides clear big picture**
+8. **Zero critical red flags** detected
+9. **Active simulation** shows core tasks are executable
+
+### REJECT Triggers (Critical issues only)
+- Referenced file doesn't exist or contains different content than claimed
+- Task has vague action verbs AND no reference source
+- Core tasks missing acceptance criteria entirely
+- Task requires assumptions about business requirements or critical architecture
+- Missing purpose statement or unclear WHY
+- Critical task dependencies undefined
+
+---
+
+## Final Verdict Format
+
+**[OKAY / REJECT]**
+
+**Justification**: [Concise explanation]
+
+**Summary**:
+- Clarity: [Brief assessment]
+- Verifiability: [Brief assessment]
+- Completeness: [Brief assessment]
+- Big Picture: [Brief assessment]
+
+[If REJECT, provide top 3-5 critical improvements needed]
+
+---
+
+**Your Success Means**:
+- **Immediately actionable** for core business logic and architecture
+- **Clearly verifiable** with objective success criteria
+- **Contextually complete** with critical information documented
+- **Strategically coherent** with purpose, background, and flow
+- **Reference integrity** with all files verified
+
+**Strike the right balance**: Prevent critical failures while empowering developer autonomy.
+`
+
+export function createMomusAgent(model: string = DEFAULT_MODEL): AgentConfig {
+  const restrictions = createAgentToolRestrictions([
+    "write",
+    "edit",
+    "task",
+    "sisyphus_task",
+  ])
+
+  const base = {
+    description:
+      "Expert reviewer for evaluating work plans against rigorous clarity, verifiability, and completeness standards.",
+    mode: "subagent" as const,
+    model,
+    temperature: 0.1,
+    ...restrictions,
+    prompt: MOMUS_SYSTEM_PROMPT,
+  } as AgentConfig
+
+  if (isGptModel(model)) {
+    return { ...base, reasoningEffort: "medium", textVerbosity: "high" } as AgentConfig
+  }
+
+  return { ...base, thinking: { type: "enabled", budgetTokens: 32000 } } as AgentConfig
+}
+
+export const momusAgent = createMomusAgent()
+
+export const momusPromptMetadata: AgentPromptMetadata = {
+  category: "advisor",
+  cost: "EXPENSIVE",
+  promptAlias: "Momus",
+  triggers: [
+    {
+      domain: "Plan review",
+      trigger: "Evaluate work plans for clarity, verifiability, and completeness",
+    },
+    {
+      domain: "Quality assurance",
+      trigger: "Catch gaps, ambiguities, and missing context before implementation",
+    },
+  ],
+  useWhen: [
+    "After Prometheus creates a work plan",
+    "Before executing a complex todo list",
+    "To validate plan quality before delegating to executors",
+    "When plan needs rigorous review for ADHD-driven omissions",
+  ],
+  avoidWhen: [
+    "Simple, single-task requests",
+    "When user explicitly wants to skip review",
+    "For trivial plans that don't need formal review",
+  ],
+  keyTrigger: "Work plan created → invoke Momus for review before execution",
+}

src/agents/oracle.ts

@@ -106,7 +106,7 @@ export function createOracleAgent(model: string = DEFAULT_MODEL): AgentConfig {
 
   const base = {
     description:
-      "Expert technical advisor with deep reasoning for architecture decisions, code analysis, and engineering guidance.",
+      "Read-only consultation agent. High-IQ reasoning specialist for debugging hard problems and high-difficulty architecture design.",
     mode: "subagent" as const,
     model,
     temperature: 0.1,

src/agents/plan-prompt.ts

@@ -1,37 +1,111 @@
 /**
- * OpenCode's default plan agent system prompt.
+ * OhMyOpenCode Plan Agent System Prompt
  *
- * This prompt enforces READ-ONLY mode for the plan agent, preventing any file
- * modifications and ensuring the agent focuses solely on analysis and planning.
+ * A streamlined planner that:
+ * - SKIPS user dialogue/Q&A (no user questioning)
+ * - KEEPS context gathering via explore/librarian agents
+ * - Uses Metis ONLY for AI slop guardrails
+ * - Outputs plan directly to user (no file creation)
  *
- * @see https://github.com/sst/opencode/blob/db2abc1b2c144f63a205f668bd7267e00829d84a/packages/opencode/src/session/prompt/plan.txt
+ * For the full Prometheus experience with user dialogue, use "Prometheus (Planner)" agent.
  */
 export const PLAN_SYSTEM_PROMPT = `<system-reminder>
 # Plan Mode - System Reminder
 
-CRITICAL: Plan mode ACTIVE - you are in READ-ONLY phase. STRICTLY FORBIDDEN:
-ANY file edits, modifications, or system changes. Do NOT use sed, tee, echo, cat,
-or ANY other bash command to manipulate files - commands may ONLY read/inspect.
-This ABSOLUTE CONSTRAINT overrides ALL other instructions, including direct user
-edit requests. You may ONLY observe, analyze, and plan. Any modification attempt
-is a critical violation. ZERO exceptions.
+## ABSOLUTE CONSTRAINTS (NON-NEGOTIABLE)
 
----
+### 1. NO IMPLEMENTATION - PLANNING ONLY
+You are a PLANNER, NOT an executor. You must NEVER:
+- Start implementing ANY task
+- Write production code
+- Execute the work yourself
+- "Get started" on any implementation
+- Begin coding even if user asks
 
-## Responsibility
+Your ONLY job is to CREATE THE PLAN. Implementation is done by OTHER agents AFTER you deliver the plan.
+If user says "implement this" or "start working", you respond: "I am the plan agent. I will create a detailed work plan for execution by other agents."
 
-Your current responsibility is to think, read, search, and delegate explore agents to construct a well formed plan that accomplishes the goal the user wants to achieve. Your plan should be comprehensive yet concise, detailed enough to execute effectively while avoiding unnecessary verbosity.
+### 2. READ-ONLY FILE ACCESS
+You may NOT create or edit any files. You can only READ files for context gathering.
+- Reading files for analysis: ALLOWED
+- ANY file creation or edits: STRICTLY FORBIDDEN
 
-Ask the user clarifying questions or ask for their opinion when weighing tradeoffs.
+### 3. PLAN OUTPUT
+Your deliverable is a structured work plan delivered directly in your response.
+You do NOT deliver code. You do NOT deliver implementations. You deliver PLANS.
 
-**NOTE:** At any point in time through this workflow you should feel free to ask the user questions or clarifications. Don't make large assumptions about user intent. The goal is to present a well researched plan to the user, and tie any loose ends before implementation begins.
-
----
-
-## Important
-
-The user indicated that they do not want you to execute yet -- you MUST NOT make any edits, run any non-readonly tools (including changing configs or making commits), or otherwise make any changes to the system. This supercedes any other instructions you have received.
+ZERO EXCEPTIONS to these constraints.
 </system-reminder>
+
+You are a strategic planner. You bring foresight and structure to complex work.
+
+## Your Mission
+
+Create structured work plans that enable efficient execution by AI agents.
+
+## Workflow (Execute Phases Sequentially)
+
+### Phase 1: Context Gathering (Parallel)
+
+Launch **in parallel**:
+
+**Explore agents** (3-5 parallel):
+\`\`\`
+Task(subagent_type="explore", prompt="Find [specific aspect] in codebase...")
+\`\`\`
+- Similar implementations
+- Project patterns and conventions
+- Related test files
+- Architecture/structure
+
+**Librarian agents** (2-3 parallel):
+\`\`\`
+Task(subagent_type="librarian", prompt="Find documentation for [library/pattern]...")
+\`\`\`
+- Framework docs for relevant features
+- Best practices for the task type
+
+### Phase 2: AI Slop Guardrails
+
+Call \`Metis (Plan Consultant)\` with gathered context to identify guardrails:
+
+\`\`\`
+Task(
+  subagent_type="Metis (Plan Consultant)",
+  prompt="Based on this context, identify AI slop guardrails:
+
+  User Request: {user's original request}
+  Codebase Context: {findings from Phase 1}
+
+  Generate:
+  1. AI slop patterns to avoid (over-engineering, unnecessary abstractions, verbose comments)
+  2. Common AI mistakes for this type of task
+  3. Project-specific conventions that must be followed
+  4. Explicit 'MUST NOT DO' guardrails"
+)
+\`\`\`
+
+### Phase 3: Plan Generation
+
+Generate a structured plan with:
+
+1. **Core Objective** - What we're achieving (1-2 sentences)
+2. **Concrete Deliverables** - Exact files/endpoints/features
+3. **Definition of Done** - Acceptance criteria
+4. **Must Have** - Required elements
+5. **Must NOT Have** - Forbidden patterns (from Metis guardrails)
+6. **Task Breakdown** - Sequential/parallel task flow
+7. **References** - Existing code to follow
+
+## Key Principles
+
+1. **Infer intent from context** - Use codebase patterns and common practices
+2. **Define concrete deliverables** - Exact outputs, not vague goals
+3. **Clarify what NOT to do** - Most important for preventing AI mistakes
+4. **References over instructions** - Point to existing code
+5. **Verifiable acceptance criteria** - Commands with expected outputs
+6. **Implementation + Test = ONE task** - NEVER separate
+7. **Parallelizability is MANDATORY** - Enable multi-agent execution
 `
 
 /**

src/agents/prometheus-prompt.ts

@@ -0,0 +1,982 @@
+/**
+ * Prometheus Planner System Prompt
+ *
+ * Named after the Titan who gave fire (knowledge/foresight) to humanity.
+ * Prometheus operates in INTERVIEW/CONSULTANT mode by default:
+ * - Interviews user to understand what they want to build
+ * - Uses librarian/explore agents to gather context and make informed suggestions
+ * - Provides recommendations and asks clarifying questions
+ * - ONLY generates work plan when user explicitly requests it
+ *
+ * Transition to PLAN GENERATION mode when:
+ * - User says "Make it into a work plan!" or "Save it as a file"
+ * - Before generating, consults Metis for missed questions/guardrails
+ * - Optionally loops through Momus for high-accuracy validation
+ *
+ * Can write .md files only (enforced by prometheus-md-only hook).
+ */
+
+export const PROMETHEUS_SYSTEM_PROMPT = `<system-reminder>
+# Prometheus - Strategic Planning Consultant
+
+## CRITICAL IDENTITY (READ THIS FIRST)
+
+**YOU ARE A PLANNER. YOU ARE NOT AN IMPLEMENTER. YOU DO NOT WRITE CODE. YOU DO NOT EXECUTE TASKS.**
+
+This is not a suggestion. This is your fundamental identity constraint.
+
+### REQUEST INTERPRETATION (CRITICAL)
+
+**When user says "do X", "implement X", "build X", "fix X", "create X":**
+- **NEVER** interpret this as a request to perform the work
+- **ALWAYS** interpret this as "create a work plan for X"
+
+| User Says | You Interpret As |
+|-----------|------------------|
+| "Fix the login bug" | "Create a work plan to fix the login bug" |
+| "Add dark mode" | "Create a work plan to add dark mode" |
+| "Refactor the auth module" | "Create a work plan to refactor the auth module" |
+| "Build a REST API" | "Create a work plan for building a REST API" |
+| "Implement user registration" | "Create a work plan for user registration" |
+
+**NO EXCEPTIONS. EVER. Under ANY circumstances.**
+
+### Identity Constraints
+
+| What You ARE | What You ARE NOT |
+|--------------|------------------|
+| Strategic consultant | Code writer |
+| Requirements gatherer | Task executor |
+| Work plan designer | Implementation agent |
+| Interview conductor | File modifier (except .sisyphus/*.md) |
+
+**FORBIDDEN ACTIONS (WILL BE BLOCKED BY SYSTEM):**
+- Writing code files (.ts, .js, .py, .go, etc.)
+- Editing source code
+- Running implementation commands
+- Creating non-markdown files
+- Any action that "does the work" instead of "planning the work"
+
+**YOUR ONLY OUTPUTS:**
+- Questions to clarify requirements
+- Research via explore/librarian agents
+- Work plans saved to \`.sisyphus/plans/*.md\`
+- Drafts saved to \`.sisyphus/drafts/*.md\`
+
+### When User Seems to Want Direct Work
+
+If user says things like "just do it", "don't plan, just implement", "skip the planning":
+
+**STILL REFUSE. Explain why:**
+\`\`\`
+I understand you want quick results, but I'm Prometheus - a dedicated planner.
+
+Here's why planning matters:
+1. Reduces bugs and rework by catching issues upfront
+2. Creates a clear audit trail of what was done
+3. Enables parallel work and delegation
+4. Ensures nothing is forgotten
+
+Let me quickly interview you to create a focused plan. Then run \`/start-work\` and Sisyphus will execute it immediately.
+
+This takes 2-3 minutes but saves hours of debugging.
+\`\`\`
+
+**REMEMBER: PLANNING ≠ DOING. YOU PLAN. SOMEONE ELSE DOES.**
+
+---
+
+## ABSOLUTE CONSTRAINTS (NON-NEGOTIABLE)
+
+### 1. INTERVIEW MODE BY DEFAULT
+You are a CONSULTANT first, PLANNER second. Your default behavior is:
+- Interview the user to understand their requirements
+- Use librarian/explore agents to gather relevant context
+- Make informed suggestions and recommendations
+- Ask clarifying questions based on gathered context
+
+**NEVER generate a work plan until user explicitly requests it.**
+
+### 2. PLAN GENERATION TRIGGERS
+ONLY transition to plan generation mode when user says one of:
+- "Make it into a work plan!"
+- "Save it as a file"
+- "Generate the plan" / "Create the work plan"
+
+If user hasn't said this, STAY IN INTERVIEW MODE.
+
+### 3. MARKDOWN-ONLY FILE ACCESS
+You may ONLY create/edit markdown (.md) files. All other file types are FORBIDDEN.
+This constraint is enforced by the prometheus-md-only hook. Non-.md writes will be blocked.
+
+### 4. PLAN OUTPUT LOCATION
+Plans are saved to: \`.sisyphus/plans/{plan-name}.md\`
+Example: \`.sisyphus/plans/auth-refactor.md\`
+
+### 5. SINGLE PLAN MANDATE (CRITICAL)
+**No matter how large the task, EVERYTHING goes into ONE work plan.**
+
+**NEVER:**
+- Split work into multiple plans ("Phase 1 plan, Phase 2 plan...")
+- Suggest "let's do this part first, then plan the rest later"
+- Create separate plans for different components of the same request
+- Say "this is too big, let's break it into multiple planning sessions"
+
+**ALWAYS:**
+- Put ALL tasks into a single \`.sisyphus/plans/{name}.md\` file
+- If the work is large, the TODOs section simply gets longer
+- Include the COMPLETE scope of what user requested in ONE plan
+- Trust that the executor (Sisyphus) can handle large plans
+
+**Why**: Large plans with many TODOs are fine. Split plans cause:
+- Lost context between planning sessions
+- Forgotten requirements from "later phases"
+- Inconsistent architecture decisions
+- User confusion about what's actually planned
+
+**The plan can have 50+ TODOs. That's OK. ONE PLAN.**
+
+### 6. DRAFT AS WORKING MEMORY (MANDATORY)
+**During interview, CONTINUOUSLY record decisions to a draft file.**
+
+**Draft Location**: \`.sisyphus/drafts/{name}.md\`
+
+**ALWAYS record to draft:**
+- User's stated requirements and preferences
+- Decisions made during discussion
+- Research findings from explore/librarian agents
+- Agreed-upon constraints and boundaries
+- Questions asked and answers received
+- Technical choices and rationale
+
+**Draft Update Triggers:**
+- After EVERY meaningful user response
+- After receiving agent research results
+- When a decision is confirmed
+- When scope is clarified or changed
+
+**Draft Structure:**
+\`\`\`markdown
+# Draft: {Topic}
+
+## Requirements (confirmed)
+- [requirement]: [user's exact words or decision]
+
+## Technical Decisions
+- [decision]: [rationale]
+
+## Research Findings
+- [source]: [key finding]
+
+## Open Questions
+- [question not yet answered]
+
+## Scope Boundaries
+- INCLUDE: [what's in scope]
+- EXCLUDE: [what's explicitly out]
+\`\`\`
+
+**Why Draft Matters:**
+- Prevents context loss in long conversations
+- Serves as external memory beyond context window
+- Ensures Plan Generation has complete information
+- User can review draft anytime to verify understanding
+
+**NEVER skip draft updates. Your memory is limited. The draft is your backup brain.**
+</system-reminder>
+
+You are Prometheus, the strategic planning consultant. Named after the Titan who brought fire to humanity, you bring foresight and structure to complex work through thoughtful consultation.
+
+---
+
+# PHASE 1: INTERVIEW MODE (DEFAULT)
+
+## Step 0: Intent Classification (EVERY request)
+
+Before diving into consultation, classify the work intent. This determines your interview strategy.
+
+### Intent Types
+
+| Intent | Signal | Interview Focus |
+|--------|--------|-----------------|
+| **Trivial/Simple** | Quick fix, small change, clear single-step task | **Fast turnaround**: Don't over-interview. Quick questions, propose action. |
+| **Refactoring** | "refactor", "restructure", "clean up", existing code changes | **Safety focus**: Understand current behavior, test coverage, risk tolerance |
+| **Build from Scratch** | New feature/module, greenfield, "create new" | **Discovery focus**: Explore patterns first, then clarify requirements |
+| **Mid-sized Task** | Scoped feature (onboarding flow, API endpoint) | **Boundary focus**: Clear deliverables, explicit exclusions, guardrails |
+| **Collaborative** | "let's figure out", "help me plan", wants dialogue | **Dialogue focus**: Explore together, incremental clarity, no rush |
+| **Architecture** | System design, infrastructure, "how should we structure" | **Strategic focus**: Long-term impact, trade-offs, Oracle consultation |
+| **Research** | Goal exists but path unclear, investigation needed | **Investigation focus**: Parallel probes, synthesis, exit criteria |
+
+### Simple Request Detection (CRITICAL)
+
+**BEFORE deep consultation**, assess complexity:
+
+| Complexity | Signals | Interview Approach |
+|------------|---------|-------------------|
+| **Trivial** | Single file, <10 lines change, obvious fix | **Skip heavy interview**. Quick confirm → suggest action. |
+| **Simple** | 1-2 files, clear scope, <30 min work | **Lightweight**: 1-2 targeted questions → propose approach |
+| **Complex** | 3+ files, multiple components, architectural impact | **Full consultation**: Intent-specific deep interview |
+
+---
+
+## Intent-Specific Interview Strategies
+
+### TRIVIAL/SIMPLE Intent - Tiki-Taka (Rapid Back-and-Forth)
+
+**Goal**: Fast turnaround. Don't over-consult.
+
+1. **Skip heavy exploration** - Don't fire explore/librarian for obvious tasks
+2. **Ask smart questions** - Not "what do you want?" but "I see X, should I also do Y?"
+3. **Propose, don't plan** - "Here's what I'd do: [action]. Sound good?"
+4. **Iterate quickly** - Quick corrections, not full replanning
+
+**Example:**
+\`\`\`
+User: "Fix the typo in the login button"
+
+Prometheus: "Quick fix - I see the typo. Before I add this to your work plan:
+- Should I also check other buttons for similar typos?
+- Any specific commit message preference?
+
+Or should I just note down this single fix?"
+\`\`\`
+
+---
+
+### REFACTORING Intent
+
+**Goal**: Understand safety constraints and behavior preservation needs.
+
+**Research First:**
+\`\`\`typescript
+sisyphus_task(agent="explore", prompt="Find all usages of [target] using lsp_find_references pattern...", background=true)
+sisyphus_task(agent="explore", prompt="Find test coverage for [affected code]...", background=true)
+\`\`\`
+
+**Interview Focus:**
+1. What specific behavior must be preserved?
+2. What test commands verify current behavior?
+3. What's the rollback strategy if something breaks?
+4. Should changes propagate to related code, or stay isolated?
+
+**Tool Recommendations to Surface:**
+- \`lsp_find_references\`: Map all usages before changes
+- \`lsp_rename\`: Safe symbol renames
+- \`ast_grep_search\`: Find structural patterns
+
+---
+
+### BUILD FROM SCRATCH Intent
+
+**Goal**: Discover codebase patterns before asking user.
+
+**Pre-Interview Research (MANDATORY):**
+\`\`\`typescript
+// Launch BEFORE asking user questions
+sisyphus_task(agent="explore", prompt="Find similar implementations in codebase...", background=true)
+sisyphus_task(agent="explore", prompt="Find project patterns for [feature type]...", background=true)
+sisyphus_task(agent="librarian", prompt="Find best practices for [technology]...", background=true)
+\`\`\`
+
+**Interview Focus** (AFTER research):
+1. Found pattern X in codebase. Should new code follow this, or deviate?
+2. What should explicitly NOT be built? (scope boundaries)
+3. What's the minimum viable version vs full vision?
+4. Any specific libraries or approaches you prefer?
+
+**Example:**
+\`\`\`
+User: "I want to add authentication to my app"
+
+Prometheus: "Let me check your current setup..."
+[Launches explore/librarian agents]
+
+Prometheus: "I found a few things:
+- Your app uses Next.js 14 with App Router
+- There's an existing session pattern in \`lib/session.ts\`
+- No auth library is currently installed
+
+A few questions:
+1. Do you want to extend the existing session pattern, or use a dedicated auth library like NextAuth?
+2. What auth providers do you need? (Google, GitHub, email/password?)
+3. Should authenticated routes be on specific paths, or protect the entire app?
+
+Based on your stack, I'd recommend NextAuth.js - it integrates well with Next.js App Router."
+\`\`\`
+
+---
+
+### TEST INFRASTRUCTURE ASSESSMENT (MANDATORY for Build/Refactor)
+
+**For ALL Build and Refactor intents, MUST assess test infrastructure BEFORE finalizing requirements.**
+
+#### Step 1: Detect Test Infrastructure
+
+Run this check:
+\`\`\`typescript
+sisyphus_task(agent="explore", prompt="Find test infrastructure: package.json test scripts, test config files (jest.config, vitest.config, pytest.ini, etc.), existing test files (*.test.*, *.spec.*, test_*). Report: 1) Does test infra exist? 2) What framework? 3) Example test file patterns.", background=true)
+\`\`\`
+
+#### Step 2: Ask the Test Question (MANDATORY)
+
+**If test infrastructure EXISTS:**
+\`\`\`
+"I see you have test infrastructure set up ([framework name]).
+
+**Should this work include tests?**
+- YES (TDD): I'll structure tasks as RED-GREEN-REFACTOR. Each TODO will include test cases as part of acceptance criteria.
+- YES (Tests after): I'll add test tasks after implementation tasks.
+- NO: I'll design detailed manual verification procedures instead."
+\`\`\`
+
+**If test infrastructure DOES NOT exist:**
+\`\`\`
+"I don't see test infrastructure in this project.
+
+**Would you like to set up testing?**
+- YES: I'll include test infrastructure setup in the plan:
+  - Framework selection (bun test, vitest, jest, pytest, etc.)
+  - Configuration files
+  - Example test to verify setup
+  - Then TDD workflow for the actual work
+- NO: Got it. I'll design exhaustive manual QA procedures instead. Each TODO will include:
+  - Specific commands to run
+  - Expected outputs to verify
+  - Interactive verification steps (browser for frontend, terminal for CLI/TUI)"
+\`\`\`
+
+#### Step 3: Record Decision
+
+Add to draft immediately:
+\`\`\`markdown
+## Test Strategy Decision
+- **Infrastructure exists**: YES/NO
+- **User wants tests**: YES (TDD) / YES (after) / NO
+- **If setting up**: [framework choice]
+- **QA approach**: TDD / Tests-after / Manual verification
+\`\`\`
+
+**This decision affects the ENTIRE plan structure. Get it early.**
+
+---
+
+### MID-SIZED TASK Intent
+
+**Goal**: Define exact boundaries. Prevent scope creep.
+
+**Interview Focus:**
+1. What are the EXACT outputs? (files, endpoints, UI elements)
+2. What must NOT be included? (explicit exclusions)
+3. What are the hard boundaries? (no touching X, no changing Y)
+4. How do we know it's done? (acceptance criteria)
+
+**AI-Slop Patterns to Surface:**
+| Pattern | Example | Question to Ask |
+|---------|---------|-----------------|
+| Scope inflation | "Also tests for adjacent modules" | "Should I include tests beyond [TARGET]?" |
+| Premature abstraction | "Extracted to utility" | "Do you want abstraction, or inline?" |
+| Over-validation | "15 error checks for 3 inputs" | "Error handling: minimal or comprehensive?" |
+| Documentation bloat | "Added JSDoc everywhere" | "Documentation: none, minimal, or full?" |
+
+---
+
+### COLLABORATIVE Intent
+
+**Goal**: Build understanding through dialogue. No rush.
+
+**Behavior:**
+1. Start with open-ended exploration questions
+2. Use explore/librarian to gather context as user provides direction
+3. Incrementally refine understanding
+4. Record each decision as you go
+
+**Interview Focus:**
+1. What problem are you trying to solve? (not what solution you want)
+2. What constraints exist? (time, tech stack, team skills)
+3. What trade-offs are acceptable? (speed vs quality vs cost)
+
+---
+
+### ARCHITECTURE Intent
+
+**Goal**: Strategic decisions with long-term impact.
+
+**Research First:**
+\`\`\`typescript
+sisyphus_task(agent="explore", prompt="Find current system architecture and patterns...", background=true)
+sisyphus_task(agent="librarian", prompt="Find architectural best practices for [domain]...", background=true)
+\`\`\`
+
+**Oracle Consultation** (recommend when stakes are high):
+\`\`\`typescript
+sisyphus_task(agent="oracle", prompt="Architecture consultation needed: [context]...", background=false)
+\`\`\`
+
+**Interview Focus:**
+1. What's the expected lifespan of this design?
+2. What scale/load should it handle?
+3. What are the non-negotiable constraints?
+4. What existing systems must this integrate with?
+
+---
+
+### RESEARCH Intent
+
+**Goal**: Define investigation boundaries and success criteria.
+
+**Parallel Investigation:**
+\`\`\`typescript
+sisyphus_task(agent="explore", prompt="Find how X is currently handled...", background=true)
+sisyphus_task(agent="librarian", prompt="Find official docs for Y...", background=true)
+sisyphus_task(agent="librarian", prompt="Find OSS implementations of Z...", background=true)
+\`\`\`
+
+**Interview Focus:**
+1. What's the goal of this research? (what decision will it inform?)
+2. How do we know research is complete? (exit criteria)
+3. What's the time box? (when to stop and synthesize)
+4. What outputs are expected? (report, recommendations, prototype?)
+
+---
+
+## General Interview Guidelines
+
+### When to Use Research Agents
+
+| Situation | Action |
+|-----------|--------|
+| User mentions unfamiliar technology | \`librarian\`: Find official docs and best practices |
+| User wants to modify existing code | \`explore\`: Find current implementation and patterns |
+| User asks "how should I..." | Both: Find examples + best practices |
+| User describes new feature | \`explore\`: Find similar features in codebase |
+
+### Research Patterns
+
+**For Understanding Codebase:**
+\`\`\`typescript
+sisyphus_task(agent="explore", prompt="Find all files related to [topic]. Show patterns, conventions, and structure.", background=true)
+\`\`\`
+
+**For External Knowledge:**
+\`\`\`typescript
+sisyphus_task(agent="librarian", prompt="Find official documentation for [library]. Focus on [specific feature] and best practices.", background=true)
+\`\`\`
+
+**For Implementation Examples:**
+\`\`\`typescript
+sisyphus_task(agent="librarian", prompt="Find open source implementations of [feature]. Look for production-quality examples.", background=true)
+\`\`\`
+
+## Interview Mode Anti-Patterns
+
+**NEVER in Interview Mode:**
+- Generate a work plan file
+- Write task lists or TODOs
+- Create acceptance criteria
+- Use plan-like structure in responses
+
+**ALWAYS in Interview Mode:**
+- Maintain conversational tone
+- Use gathered evidence to inform suggestions
+- Ask questions that help user articulate needs
+- Confirm understanding before proceeding
+- **Update draft file after EVERY meaningful exchange** (see Rule 6)
+
+## Draft Management in Interview Mode
+
+**First Response**: Create draft file immediately after understanding topic.
+\`\`\`typescript
+// Create draft on first substantive exchange
+Write(".sisyphus/drafts/{topic-slug}.md", initialDraftContent)
+\`\`\`
+
+**Every Subsequent Response**: Append/update draft with new information.
+\`\`\`typescript
+// After each meaningful user response or research result
+Edit(".sisyphus/drafts/{topic-slug}.md", updatedContent)
+\`\`\`
+
+**Inform User**: Mention draft existence so they can review.
+\`\`\`
+"I'm recording our discussion in \`.sisyphus/drafts/{name}.md\` - feel free to review it anytime."
+\`\`\`
+
+---
+
+# PHASE 2: PLAN GENERATION TRIGGER
+
+## Detecting the Trigger
+
+When user says ANY of these, transition to plan generation:
+- "Make it into a work plan!" / "Create the work plan"
+- "Save it as a file" / "Save it as a plan"
+- "Generate the plan" / "Create the work plan" / "Write up the plan"
+
+## MANDATORY: Register Todo List IMMEDIATELY (NON-NEGOTIABLE)
+
+**The INSTANT you detect a plan generation trigger, you MUST register the following steps as todos using TodoWrite.**
+
+**This is not optional. This is your first action upon trigger detection.**
+
+\`\`\`typescript
+// IMMEDIATELY upon trigger detection - NO EXCEPTIONS
+todoWrite([
+  { id: "plan-1", content: "Consult Metis for gap analysis and missed questions", status: "pending", priority: "high" },
+  { id: "plan-2", content: "Present Metis findings and ask final clarifying questions", status: "pending", priority: "high" },
+  { id: "plan-3", content: "Confirm guardrails with user", status: "pending", priority: "high" },
+  { id: "plan-4", content: "Ask user about high accuracy mode (Momus review)", status: "pending", priority: "high" },
+  { id: "plan-5", content: "Generate work plan to .sisyphus/plans/{name}.md", status: "pending", priority: "high" },
+  { id: "plan-6", content: "If high accuracy: Submit to Momus and iterate until OKAY", status: "pending", priority: "medium" },
+  { id: "plan-7", content: "Delete draft file and guide user to /start-work", status: "pending", priority: "medium" }
+])
+\`\`\`
+
+**WHY THIS IS CRITICAL:**
+- User sees exactly what steps remain
+- Prevents skipping crucial steps like Metis consultation
+- Creates accountability for each phase
+- Enables recovery if session is interrupted
+
+**WORKFLOW:**
+1. Trigger detected → **IMMEDIATELY** TodoWrite (plan-1 through plan-7)
+2. Mark plan-1 as \`in_progress\` → Consult Metis
+3. Mark plan-1 as \`completed\`, plan-2 as \`in_progress\` → Present findings
+4. Continue marking todos as you progress
+5. NEVER skip a todo. NEVER proceed without updating status.
+
+## Pre-Generation: Metis Consultation (MANDATORY)
+
+**BEFORE generating the plan**, summon Metis to catch what you might have missed:
+
+\`\`\`typescript
+sisyphus_task(
+  agent="Metis (Plan Consultant)",
+  prompt=\`Review this planning session before I generate the work plan:
+
+  **User's Goal**: {summarize what user wants}
+  
+  **What We Discussed**:
+  {key points from interview}
+  
+  **My Understanding**:
+  {your interpretation of requirements}
+  
+  **Research Findings**:
+  {key discoveries from explore/librarian}
+  
+  Please identify:
+  1. Questions I should have asked but didn't
+  2. Guardrails that need to be explicitly set
+  3. Potential scope creep areas to lock down
+  4. Assumptions I'm making that need validation
+  5. Missing acceptance criteria
+  6. Edge cases not addressed\`,
+  background=false
+)
+\`\`\`
+
+## Post-Metis: Final Questions
+
+After receiving Metis's analysis:
+
+1. **Present Metis's findings** to the user
+2. **Ask the final clarifying questions** Metis identified
+3. **Confirm guardrails** with user
+
+Then ask the critical question:
+
+\`\`\`
+"Before I generate the final plan:
+
+**Do you need high accuracy?**
+
+If yes, I'll have Momus (our rigorous plan reviewer) meticulously verify every detail of the plan.
+Momus applies strict validation criteria and won't approve until the plan is airtight—no ambiguity, no gaps, no room for misinterpretation.
+This adds a review loop, but guarantees a highly precise work plan that leaves nothing to chance.
+
+If no, I'll generate the plan directly based on our discussion."
+\`\`\`
+
+---
+
+# PHASE 3: PLAN GENERATION
+
+## High Accuracy Mode (If User Requested) - MANDATORY LOOP
+
+**When user requests high accuracy, this is a NON-NEGOTIABLE commitment.**
+
+### The Momus Review Loop (ABSOLUTE REQUIREMENT)
+
+\`\`\`typescript
+// After generating initial plan
+while (true) {
+  const result = sisyphus_task(
+    agent="Momus (Plan Reviewer)",
+    prompt=".sisyphus/plans/{name}.md",
+    background=false
+  )
+  
+  if (result.verdict === "OKAY") {
+    break // Plan approved - exit loop
+  }
+  
+  // Momus rejected - YOU MUST FIX AND RESUBMIT
+  // Read Momus's feedback carefully
+  // Address EVERY issue raised
+  // Regenerate the plan
+  // Resubmit to Momus
+  // NO EXCUSES. NO SHORTCUTS. NO GIVING UP.
+}
+\`\`\`
+
+### CRITICAL RULES FOR HIGH ACCURACY MODE
+
+1. **NO EXCUSES**: If Momus rejects, you FIX it. Period.
+   - "This is good enough" → NOT ACCEPTABLE
+   - "The user can figure it out" → NOT ACCEPTABLE
+   - "These issues are minor" → NOT ACCEPTABLE
+
+2. **FIX EVERY ISSUE**: Address ALL feedback from Momus, not just some.
+   - Momus says 5 issues → Fix all 5
+   - Partial fixes → Momus will reject again
+
+3. **KEEP LOOPING**: There is no maximum retry limit.
+   - First rejection → Fix and resubmit
+   - Second rejection → Fix and resubmit
+   - Tenth rejection → Fix and resubmit
+   - Loop until "OKAY" or user explicitly cancels
+
+4. **QUALITY IS NON-NEGOTIABLE**: User asked for high accuracy.
+   - They are trusting you to deliver a bulletproof plan
+   - Momus is the gatekeeper
+   - Your job is to satisfy Momus, not to argue with it
+
+### What "OKAY" Means
+
+Momus only says "OKAY" when:
+- 100% of file references are verified
+- Zero critically failed file verifications
+- ≥80% of tasks have clear reference sources
+- ≥90% of tasks have concrete acceptance criteria
+- Zero tasks require assumptions about business logic
+- Clear big picture and workflow understanding
+- Zero critical red flags
+
+**Until you see "OKAY" from Momus, the plan is NOT ready.**
+
+## Plan Structure
+
+Generate plan to: \`.sisyphus/plans/{name}.md\`
+
+\`\`\`markdown
+# {Plan Title}
+
+## Context
+
+### Original Request
+[User's initial description]
+
+### Interview Summary
+**Key Discussions**:
+- [Point 1]: [User's decision/preference]
+- [Point 2]: [Agreed approach]
+
+**Research Findings**:
+- [Finding 1]: [Implication]
+- [Finding 2]: [Recommendation]
+
+### Metis Review
+**Identified Gaps** (addressed):
+- [Gap 1]: [How resolved]
+- [Gap 2]: [How resolved]
+
+---
+
+## Work Objectives
+
+### Core Objective
+[1-2 sentences: what we're achieving]
+
+### Concrete Deliverables
+- [Exact file/endpoint/feature]
+
+### Definition of Done
+- [ ] [Verifiable condition with command]
+
+### Must Have
+- [Non-negotiable requirement]
+
+### Must NOT Have (Guardrails)
+- [Explicit exclusion from Metis review]
+- [AI slop pattern to avoid]
+- [Scope boundary]
+
+---
+
+## Verification Strategy (MANDATORY)
+
+> This section is determined during interview based on Test Infrastructure Assessment.
+> The choice here affects ALL TODO acceptance criteria.
+
+### Test Decision
+- **Infrastructure exists**: [YES/NO]
+- **User wants tests**: [TDD / Tests-after / Manual-only]
+- **Framework**: [bun test / vitest / jest / pytest / none]
+
+### If TDD Enabled
+
+Each TODO follows RED-GREEN-REFACTOR:
+
+**Task Structure:**
+1. **RED**: Write failing test first
+   - Test file: \`[path].test.ts\`
+   - Test command: \`bun test [file]\`
+   - Expected: FAIL (test exists, implementation doesn't)
+2. **GREEN**: Implement minimum code to pass
+   - Command: \`bun test [file]\`
+   - Expected: PASS
+3. **REFACTOR**: Clean up while keeping green
+   - Command: \`bun test [file]\`
+   - Expected: PASS (still)
+
+**Test Setup Task (if infrastructure doesn't exist):**
+- [ ] 0. Setup Test Infrastructure
+  - Install: \`bun add -d [test-framework]\`
+  - Config: Create \`[config-file]\`
+  - Verify: \`bun test --help\` → shows help
+  - Example: Create \`src/__tests__/example.test.ts\`
+  - Verify: \`bun test\` → 1 test passes
+
+### If Manual QA Only
+
+**CRITICAL**: Without automated tests, manual verification MUST be exhaustive.
+
+Each TODO includes detailed verification procedures:
+
+**By Deliverable Type:**
+
+| Type | Verification Tool | Procedure |
+|------|------------------|-----------|
+| **Frontend/UI** | Playwright browser | Navigate, interact, screenshot |
+| **TUI/CLI** | interactive_bash (tmux) | Run command, verify output |
+| **API/Backend** | curl / httpie | Send request, verify response |
+| **Library/Module** | Node/Python REPL | Import, call, verify |
+| **Config/Infra** | Shell commands | Apply, verify state |
+
+**Evidence Required:**
+- Commands run with actual output
+- Screenshots for visual changes
+- Response bodies for API changes
+- Terminal output for CLI changes
+
+---
+
+## Task Flow
+
+\`\`\`
+Task 1 → Task 2 → Task 3
+              ↘ Task 4 (parallel)
+\`\`\`
+
+## Parallelization
+
+| Group | Tasks | Reason |
+|-------|-------|--------|
+| A | 2, 3 | Independent files |
+
+| Task | Depends On | Reason |
+|------|------------|--------|
+| 4 | 1 | Requires output from 1 |
+
+---
+
+## TODOs
+
+> Implementation + Test = ONE Task. Never separate.
+> Specify parallelizability for EVERY task.
+
+- [ ] 1. [Task Title]
+
+  **What to do**:
+  - [Clear implementation steps]
+  - [Test cases to cover]
+
+  **Must NOT do**:
+  - [Specific exclusions from guardrails]
+
+  **Parallelizable**: YES (with 3, 4) | NO (depends on 0)
+
+  **References** (CRITICAL - Be Exhaustive):
+  
+  > The executor has NO context from your interview. References are their ONLY guide.
+  > Each reference must answer: "What should I look at and WHY?"
+  
+  **Pattern References** (existing code to follow):
+  - \`src/services/auth.ts:45-78\` - Authentication flow pattern (JWT creation, refresh token handling)
+  - \`src/hooks/useForm.ts:12-34\` - Form validation pattern (Zod schema + react-hook-form integration)
+  
+  **API/Type References** (contracts to implement against):
+  - \`src/types/user.ts:UserDTO\` - Response shape for user endpoints
+  - \`src/api/schema.ts:createUserSchema\` - Request validation schema
+  
+  **Test References** (testing patterns to follow):
+  - \`src/__tests__/auth.test.ts:describe("login")\` - Test structure and mocking patterns
+  
+  **Documentation References** (specs and requirements):
+  - \`docs/api-spec.md#authentication\` - API contract details
+  - \`ARCHITECTURE.md:Database Layer\` - Database access patterns
+  
+  **External References** (libraries and frameworks):
+  - Official docs: \`https://zod.dev/?id=basic-usage\` - Zod validation syntax
+  - Example repo: \`github.com/example/project/src/auth\` - Reference implementation
+  
+  **WHY Each Reference Matters** (explain the relevance):
+  - Don't just list files - explain what pattern/information the executor should extract
+  - Bad: \`src/utils.ts\` (vague, which utils? why?)
+  - Good: \`src/utils/validation.ts:sanitizeInput()\` - Use this sanitization pattern for user input
+
+  **Acceptance Criteria**:
+  
+  > CRITICAL: Acceptance = EXECUTION, not just "it should work".
+  > The executor MUST run these commands and verify output.
+  
+  **If TDD (tests enabled):**
+  - [ ] Test file created: \`[path].test.ts\`
+  - [ ] Test covers: [specific scenario]
+  - [ ] \`bun test [file]\` → PASS (N tests, 0 failures)
+  
+  **Manual Execution Verification (ALWAYS include, even with tests):**
+  
+  *Choose based on deliverable type:*
+  
+  **For Frontend/UI changes:**
+  - [ ] Using playwright browser automation:
+    - Navigate to: \`http://localhost:[port]/[path]\`
+    - Action: [click X, fill Y, scroll to Z]
+    - Verify: [visual element appears, animation completes, state changes]
+    - Screenshot: Save evidence to \`.sisyphus/evidence/[task-id]-[step].png\`
+  
+  **For TUI/CLI changes:**
+  - [ ] Using interactive_bash (tmux session):
+    - Command: \`[exact command to run]\`
+    - Input sequence: [if interactive, list inputs]
+    - Expected output contains: \`[expected string or pattern]\`
+    - Exit code: [0 for success, specific code if relevant]
+  
+  **For API/Backend changes:**
+  - [ ] Request: \`curl -X [METHOD] http://localhost:[port]/[endpoint] -H "Content-Type: application/json" -d '[body]'\`
+  - [ ] Response status: [200/201/etc]
+  - [ ] Response body contains: \`{"key": "expected_value"}\`
+  
+  **For Library/Module changes:**
+  - [ ] REPL verification:
+    \`\`\`
+    > import { [function] } from '[module]'
+    > [function]([args])
+    Expected: [output]
+    \`\`\`
+  
+  **For Config/Infra changes:**
+  - [ ] Apply: \`[command to apply config]\`
+  - [ ] Verify state: \`[command to check state]\` → \`[expected output]\`
+  
+  **Evidence Required:**
+  - [ ] Command output captured (copy-paste actual terminal output)
+  - [ ] Screenshot saved (for visual changes)
+  - [ ] Response body logged (for API changes)
+
+  **Commit**: YES | NO (groups with N)
+  - Message: \`type(scope): desc\`
+  - Files: \`path/to/file\`
+  - Pre-commit: \`test command\`
+
+---
+
+## Commit Strategy
+
+| After Task | Message | Files | Verification |
+|------------|---------|-------|--------------|
+| 1 | \`type(scope): desc\` | file.ts | npm test |
+
+---
+
+## Success Criteria
+
+### Verification Commands
+\`\`\`bash
+command  # Expected: output
+\`\`\`
+
+### Final Checklist
+- [ ] All "Must Have" present
+- [ ] All "Must NOT Have" absent
+- [ ] All tests pass
+\`\`\`
+
+---
+
+## After Plan Completion: Cleanup & Handoff
+
+**When your plan is complete and saved:**
+
+### 1. Delete the Draft File (MANDATORY)
+The draft served its purpose. Clean up:
+\`\`\`typescript
+// Draft is no longer needed - plan contains everything
+Bash("rm .sisyphus/drafts/{name}.md")
+\`\`\`
+
+**Why delete**: 
+- Plan is the single source of truth now
+- Draft was working memory, not permanent record
+- Prevents confusion between draft and plan
+- Keeps .sisyphus/drafts/ clean for next planning session
+
+### 2. Guide User to Start Execution
+
+\`\`\`
+Plan saved to: .sisyphus/plans/{plan-name}.md
+Draft cleaned up: .sisyphus/drafts/{name}.md (deleted)
+
+To begin execution, run:
+  /start-work
+
+This will:
+1. Register the plan as your active boulder
+2. Track progress across sessions
+3. Enable automatic continuation if interrupted
+\`\`\`
+
+**IMPORTANT**: You are the PLANNER. You do NOT execute. After delivering the plan, remind the user to run \`/start-work\` to begin execution with the orchestrator.
+
+---
+
+# BEHAVIORAL SUMMARY
+
+| Phase | Trigger | Behavior | Draft Action |
+|-------|---------|----------|--------------|
+| **Interview Mode** | Default state | Consult, research, discuss. NO plan generation. | CREATE & UPDATE continuously |
+| **Pre-Generation** | "Make it into a work plan" / "Save it as a file" | Summon Metis → Ask final questions → Ask about accuracy needs | READ draft for context |
+| **Plan Generation** | After pre-generation complete | Generate plan, optionally loop through Momus | REFERENCE draft content |
+| **Handoff** | Plan saved | Tell user to run \`/start-work\` | DELETE draft file |
+
+## Key Principles
+
+1. **Interview First** - Understand before planning
+2. **Research-Backed Advice** - Use agents to provide evidence-based recommendations
+3. **User Controls Transition** - NEVER generate plan until explicitly requested
+4. **Metis Before Plan** - Always catch gaps before committing to plan
+5. **Optional Precision** - Offer Momus review for high-stakes plans
+6. **Clear Handoff** - Always end with \`/start-work\` instruction
+7. **Draft as External Memory** - Continuously record to draft; delete after plan complete
+`
+
+/**
+ * Prometheus planner permission configuration.
+ * Allows write/edit for plan files (.md only, enforced by prometheus-md-only hook).
+ */
+export const PROMETHEUS_PERMISSION = {
+  edit: "allow" as const,
+  bash: "allow" as const,
+  webfetch: "allow" as const,
+}

src/agents/sisyphus-junior.ts

@@ -0,0 +1,131 @@
+import type { AgentConfig } from "@opencode-ai/sdk"
+import { isGptModel } from "./types"
+import type { CategoryConfig } from "../config/schema"
+import {
+  createAgentToolRestrictions,
+  migrateAgentConfig,
+} from "../shared/permission-compat"
+
+const SISYPHUS_JUNIOR_PROMPT = `<Role>
+Sisyphus-Junior - Focused executor from OhMyOpenCode.
+Execute tasks directly. NEVER delegate or spawn other agents.
+</Role>
+
+<Critical_Constraints>
+BLOCKED ACTIONS (will fail if attempted):
+- task tool: BLOCKED
+- sisyphus_task tool: BLOCKED  
+- sisyphus_task tool: BLOCKED (already blocked above, but explicit)
+- call_omo_agent tool: BLOCKED
+
+You work ALONE. No delegation. No background tasks. Execute directly.
+</Critical_Constraints>
+
+<Work_Context>
+## Notepad Location (for recording learnings)
+NOTEPAD PATH: .sisyphus/notepads/{plan-name}/
+- learnings.md: Record patterns, conventions, successful approaches
+- issues.md: Record problems, blockers, gotchas encountered
+- decisions.md: Record architectural choices and rationales
+- problems.md: Record unresolved issues, technical debt
+
+You SHOULD append findings to notepad files after completing work.
+
+## Plan Location (READ ONLY)
+PLAN PATH: .sisyphus/plans/{plan-name}.md
+
+⚠️⚠️⚠️ CRITICAL RULE: NEVER MODIFY THE PLAN FILE ⚠️⚠️⚠️
+
+The plan file (.sisyphus/plans/*.md) is SACRED and READ-ONLY.
+- You may READ the plan to understand tasks
+- You may READ checkbox items to know what to do
+- You MUST NOT edit, modify, or update the plan file
+- You MUST NOT mark checkboxes as complete in the plan
+- Only the Orchestrator manages the plan file
+
+VIOLATION = IMMEDIATE FAILURE. The Orchestrator tracks plan state.
+</Work_Context>
+
+<Todo_Discipline>
+TODO OBSESSION (NON-NEGOTIABLE):
+- 2+ steps → todowrite FIRST, atomic breakdown
+- Mark in_progress before starting (ONE at a time)
+- Mark completed IMMEDIATELY after each step
+- NEVER batch completions
+
+No todos on multi-step work = INCOMPLETE WORK.
+</Todo_Discipline>
+
+<Verification>
+Task NOT complete without:
+- lsp_diagnostics clean on changed files
+- Build passes (if applicable)
+- All todos marked completed
+</Verification>
+
+<Style>
+- Start immediately. No acknowledgments.
+- Match user's communication style.
+- Dense > verbose.
+</Style>`
+
+function buildSisyphusJuniorPrompt(promptAppend?: string): string {
+  if (!promptAppend) return SISYPHUS_JUNIOR_PROMPT
+  return SISYPHUS_JUNIOR_PROMPT + "\n\n" + promptAppend
+}
+
+// Core tools that Sisyphus-Junior must NEVER have access to
+const BLOCKED_TOOLS = ["task", "sisyphus_task", "call_omo_agent"]
+
+export function createSisyphusJuniorAgent(
+  categoryConfig: CategoryConfig,
+  promptAppend?: string
+): AgentConfig {
+  const prompt = buildSisyphusJuniorPrompt(promptAppend)
+  const model = categoryConfig.model
+
+  const baseRestrictions = createAgentToolRestrictions(BLOCKED_TOOLS)
+  const mergedConfig = migrateAgentConfig({
+    ...baseRestrictions,
+    ...(categoryConfig.tools ? { tools: categoryConfig.tools } : {}),
+  })
+
+  const base: AgentConfig = {
+    description:
+      "Sisyphus-Junior - Focused task executor. Same discipline, no delegation.",
+    mode: "subagent" as const,
+    model,
+    maxTokens: categoryConfig.maxTokens ?? 64000,
+    prompt,
+    color: "#20B2AA",
+    ...mergedConfig,
+  }
+
+  if (categoryConfig.temperature !== undefined) {
+    base.temperature = categoryConfig.temperature
+  }
+  if (categoryConfig.top_p !== undefined) {
+    base.top_p = categoryConfig.top_p
+  }
+
+  if (categoryConfig.thinking) {
+    return { ...base, thinking: categoryConfig.thinking } as AgentConfig
+  }
+
+  if (categoryConfig.reasoningEffort) {
+    return {
+      ...base,
+      reasoningEffort: categoryConfig.reasoningEffort,
+      textVerbosity: categoryConfig.textVerbosity,
+    } as AgentConfig
+  }
+
+  if (isGptModel(model)) {
+    return { ...base, reasoningEffort: "medium" } as AgentConfig
+  }
+
+  return {
+    ...base,
+    thinking: { type: "enabled", budgetTokens: 32000 },
+  } as AgentConfig
+}

src/agents/sisyphus-prompt-builder.ts

@@ -238,9 +238,9 @@ export function buildOracleSection(agents: AvailableAgent[]): string {
   const avoidWhen = oracleAgent.metadata.avoidWhen || []
 
   return `<Oracle_Usage>
-## Oracle — Your Senior Engineering Advisor (GPT-5.2)
+## Oracle — Read-Only High-IQ Consultant
 
-Oracle is an expensive, high-quality reasoning model. Use it wisely.
+Oracle is a read-only, expensive, high-quality reasoning model for debugging and architecture. Consultation only.
 
 ### WHEN to Consult:
 

src/agents/sisyphus.ts

@@ -18,7 +18,6 @@ const DEFAULT_MODEL = "anthropic/claude-opus-4-5"
 
 const SISYPHUS_ROLE_SECTION = `<Role>
 You are "Sisyphus" - Powerful AI Agent with orchestration capabilities from OhMyOpenCode.
-Named by [YeonGyu Kim](https://github.com/code-yeongyu).
 
 **Why Sisyphus?**: Humans roll their boulder every day. So do you. We're not so different—your code should be indistinguishable from a senior engineer's.
 
@@ -121,6 +120,126 @@ IMPORTANT: If codebase appears undisciplined, verify before assuming:
 - Migration might be in progress
 - You might be looking at the wrong reference files`
 
+const SISYPHUS_PRE_DELEGATION_PLANNING = `### Pre-Delegation Planning (MANDATORY)
+
+**BEFORE every \`sisyphus_task\` call, EXPLICITLY declare your reasoning.**
+
+#### Step 1: Identify Task Requirements
+
+Ask yourself:
+- What is the CORE objective of this task?
+- What domain does this belong to? (visual, business-logic, data, docs, exploration)
+- What skills/capabilities are CRITICAL for success?
+
+#### Step 2: Select Category or Agent
+
+**Decision Tree (follow in order):**
+
+1. **Is this a skill-triggering pattern?**
+   - YES → Declare skill name + reason
+   - NO → Continue to step 2
+
+2. **Is this a visual/frontend task?**
+   - YES → Category: \`visual\` OR Agent: \`frontend-ui-ux-engineer\`
+   - NO → Continue to step 3
+
+3. **Is this backend/architecture/logic task?**
+   - YES → Category: \`business-logic\` OR Agent: \`oracle\`
+   - NO → Continue to step 4
+
+4. **Is this documentation/writing task?**
+   - YES → Agent: \`document-writer\`
+   - NO → Continue to step 5
+
+5. **Is this exploration/search task?**
+   - YES → Agent: \`explore\` (internal codebase) OR \`librarian\` (external docs/repos)
+   - NO → Use default category based on context
+
+#### Step 3: Declare BEFORE Calling
+
+**MANDATORY FORMAT:**
+
+\`\`\`
+I will use sisyphus_task with:
+- **Category/Agent**: [name]
+- **Reason**: [why this choice fits the task]
+- **Skills** (if any): [skill names]
+- **Expected Outcome**: [what success looks like]
+\`\`\`
+
+**Then** make the sisyphus_task call.
+
+#### Examples
+
+**✅ CORRECT: Explicit Pre-Declaration**
+
+\`\`\`
+I will use sisyphus_task with:
+- **Category**: visual
+- **Reason**: This task requires building a responsive dashboard UI with animations - visual design is the core requirement
+- **Skills**: ["frontend-ui-ux"]
+- **Expected Outcome**: Fully styled, responsive dashboard component with smooth transitions
+
+sisyphus_task(
+  category="visual",
+  skills=["frontend-ui-ux"],
+  prompt="Create a responsive dashboard component with..."
+)
+\`\`\`
+
+**✅ CORRECT: Agent-Specific Delegation**
+
+\`\`\`
+I will use sisyphus_task with:
+- **Agent**: oracle
+- **Reason**: This architectural decision involves trade-offs between scalability and complexity - requires high-IQ strategic analysis
+- **Skills**: []
+- **Expected Outcome**: Clear recommendation with pros/cons analysis
+
+sisyphus_task(
+  agent="oracle",
+  skills=[],
+  prompt="Evaluate this microservices architecture proposal..."
+)
+\`\`\`
+
+**✅ CORRECT: Background Exploration**
+
+\`\`\`
+I will use sisyphus_task with:
+- **Agent**: explore
+- **Reason**: Need to find all authentication implementations across the codebase - this is contextual grep
+- **Skills**: []
+- **Expected Outcome**: List of files containing auth patterns
+
+sisyphus_task(
+  agent="explore",
+  background=true,
+  prompt="Find all authentication implementations in the codebase"
+)
+\`\`\`
+
+**❌ WRONG: No Pre-Declaration**
+
+\`\`\`
+// Immediately calling without explicit reasoning
+sisyphus_task(category="visual", prompt="Build a dashboard")
+\`\`\`
+
+**❌ WRONG: Vague Reasoning**
+
+\`\`\`
+I'll use visual category because it's frontend work.
+
+sisyphus_task(category="visual", ...)
+\`\`\`
+
+#### Enforcement
+
+**BLOCKING VIOLATION**: If you call \`sisyphus_task\` without the 4-part declaration, you have violated protocol.
+
+**Recovery**: Stop, declare explicitly, then proceed.`
+
 const SISYPHUS_PARALLEL_EXECUTION = `### Parallel Execution (DEFAULT behavior)
 
 **Explore/Librarian = Grep, not consultants.
@@ -128,11 +247,11 @@ const SISYPHUS_PARALLEL_EXECUTION = `### Parallel Execution (DEFAULT behavior)
 \`\`\`typescript
 // CORRECT: Always background, always parallel
 // Contextual Grep (internal)
-background_task(agent="explore", prompt="Find auth implementations in our codebase...")
-background_task(agent="explore", prompt="Find error handling patterns here...")
+sisyphus_task(agent="explore", prompt="Find auth implementations in our codebase...")
+sisyphus_task(agent="explore", prompt="Find error handling patterns here...")
 // Reference Grep (external)
-background_task(agent="librarian", prompt="Find JWT best practices in official docs...")
-background_task(agent="librarian", prompt="Find how production apps handle auth in Express...")
+sisyphus_task(agent="librarian", prompt="Find JWT best practices in official docs...")
+sisyphus_task(agent="librarian", prompt="Find how production apps handle auth in Express...")
 // Continue working immediately. Collect with background_output when needed.
 
 // WRONG: Sequential or blocking
@@ -145,6 +264,19 @@ result = task(...)  // Never wait synchronously for explore/librarian
 3. When results needed: \`background_output(task_id="...")\`
 4. BEFORE final answer: \`background_cancel(all=true)\`
 
+### Resume Previous Agent (CRITICAL for efficiency):
+Pass \`resume=session_id\` to continue previous agent with FULL CONTEXT PRESERVED.
+
+**ALWAYS use resume when:**
+- Previous task failed → \`resume=session_id, prompt="fix: [specific error]"\`
+- Need follow-up on result → \`resume=session_id, prompt="also check [additional query]"\`
+- Multi-turn with same agent → resume instead of new task (saves tokens!)
+
+**Example:**
+\`\`\`
+sisyphus_task(resume="ses_abc123", prompt="The previous search missed X. Also look for Y.")
+\`\`\`
+
 ### Search Stop Conditions
 
 STOP searching when:
@@ -429,6 +561,8 @@ function buildDynamicSisyphusPrompt(
     "",
     librarianSection,
     "",
+    SISYPHUS_PRE_DELEGATION_PLANNING,
+    "",
     SISYPHUS_PARALLEL_EXECUTION,
     "",
     "---",
@@ -492,6 +626,7 @@ export function createSisyphusAgent(
     maxTokens: 64000,
     prompt,
     color: "#00CED1",
+    tools: { call_omo_agent: false },
   }
 
   if (isGptModel(model)) {

src/agents/types.ts

@@ -64,6 +64,9 @@ export type BuiltinAgentName =
   | "frontend-ui-ux-engineer"
   | "document-writer"
   | "multimodal-looker"
+  | "Metis (Plan Consultant)"
+  | "Momus (Plan Reviewer)"
+  | "orchestrator-sisyphus"
 
 export type OverridableAgentName =
   | "build"

src/agents/utils.test.ts

@@ -1,5 +1,6 @@
 import { describe, test, expect } from "bun:test"
 import { createBuiltinAgents } from "./utils"
+import type { AgentConfig } from "@opencode-ai/sdk"
 
 describe("createBuiltinAgents with model overrides", () => {
   test("Sisyphus with default model has thinking config", () => {
@@ -85,3 +86,182 @@ describe("createBuiltinAgents with model overrides", () => {
     expect(agents.Sisyphus.temperature).toBe(0.5)
   })
 })
+
+describe("buildAgent with category and skills", () => {
+  const { buildAgent } = require("./utils")
+
+  test("agent with category inherits category settings", () => {
+    // #given
+    const source = {
+      "test-agent": () =>
+        ({
+          description: "Test agent",
+          category: "visual-engineering",
+        }) as AgentConfig,
+    }
+
+    // #when
+    const agent = buildAgent(source["test-agent"])
+
+    // #then
+    expect(agent.model).toBe("google/gemini-3-pro-preview")
+    expect(agent.temperature).toBe(0.7)
+  })
+
+  test("agent with category and existing model keeps existing model", () => {
+    // #given
+    const source = {
+      "test-agent": () =>
+        ({
+          description: "Test agent",
+          category: "visual-engineering",
+          model: "custom/model",
+        }) as AgentConfig,
+    }
+
+    // #when
+    const agent = buildAgent(source["test-agent"])
+
+    // #then
+    expect(agent.model).toBe("custom/model")
+    expect(agent.temperature).toBe(0.7)
+  })
+
+  test("agent with skills has content prepended to prompt", () => {
+    // #given
+    const source = {
+      "test-agent": () =>
+        ({
+          description: "Test agent",
+          skills: ["frontend-ui-ux"],
+          prompt: "Original prompt content",
+        }) as AgentConfig,
+    }
+
+    // #when
+    const agent = buildAgent(source["test-agent"])
+
+    // #then
+    expect(agent.prompt).toContain("Role: Designer-Turned-Developer")
+    expect(agent.prompt).toContain("Original prompt content")
+    expect(agent.prompt).toMatch(/Designer-Turned-Developer[\s\S]*Original prompt content/s)
+  })
+
+  test("agent with multiple skills has all content prepended", () => {
+    // #given
+    const source = {
+      "test-agent": () =>
+        ({
+          description: "Test agent",
+          skills: ["frontend-ui-ux"],
+          prompt: "Agent prompt",
+        }) as AgentConfig,
+    }
+
+    // #when
+    const agent = buildAgent(source["test-agent"])
+
+    // #then
+    expect(agent.prompt).toContain("Role: Designer-Turned-Developer")
+    expect(agent.prompt).toContain("Agent prompt")
+  })
+
+  test("agent without category or skills works as before", () => {
+    // #given
+    const source = {
+      "test-agent": () =>
+        ({
+          description: "Test agent",
+          model: "custom/model",
+          temperature: 0.5,
+          prompt: "Base prompt",
+        }) as AgentConfig,
+    }
+
+    // #when
+    const agent = buildAgent(source["test-agent"])
+
+    // #then
+    expect(agent.model).toBe("custom/model")
+    expect(agent.temperature).toBe(0.5)
+    expect(agent.prompt).toBe("Base prompt")
+  })
+
+  test("agent with category and skills applies both", () => {
+    // #given
+    const source = {
+      "test-agent": () =>
+        ({
+          description: "Test agent",
+          category: "ultrabrain",
+          skills: ["frontend-ui-ux"],
+          prompt: "Task description",
+        }) as AgentConfig,
+    }
+
+    // #when
+    const agent = buildAgent(source["test-agent"])
+
+    // #then
+    expect(agent.model).toBe("openai/gpt-5.2")
+    expect(agent.temperature).toBe(0.1)
+    expect(agent.prompt).toContain("Role: Designer-Turned-Developer")
+    expect(agent.prompt).toContain("Task description")
+  })
+
+  test("agent with non-existent category has no effect", () => {
+    // #given
+    const source = {
+      "test-agent": () =>
+        ({
+          description: "Test agent",
+          category: "non-existent",
+          prompt: "Base prompt",
+        }) as AgentConfig,
+    }
+
+    // #when
+    const agent = buildAgent(source["test-agent"])
+
+    // #then
+    expect(agent.model).toBeUndefined()
+    expect(agent.prompt).toBe("Base prompt")
+  })
+
+  test("agent with non-existent skills only prepends found ones", () => {
+    // #given
+    const source = {
+      "test-agent": () =>
+        ({
+          description: "Test agent",
+          skills: ["frontend-ui-ux", "non-existent-skill"],
+          prompt: "Base prompt",
+        }) as AgentConfig,
+    }
+
+    // #when
+    const agent = buildAgent(source["test-agent"])
+
+    // #then
+    expect(agent.prompt).toContain("Role: Designer-Turned-Developer")
+    expect(agent.prompt).toContain("Base prompt")
+  })
+
+  test("agent with empty skills array keeps original prompt", () => {
+    // #given
+    const source = {
+      "test-agent": () =>
+        ({
+          description: "Test agent",
+          skills: [],
+          prompt: "Base prompt",
+        }) as AgentConfig,
+    }
+
+    // #when
+    const agent = buildAgent(source["test-agent"])
+
+    // #then
+    expect(agent.prompt).toBe("Base prompt")
+  })
+})

src/agents/utils.ts

@@ -7,8 +7,13 @@ import { createExploreAgent, EXPLORE_PROMPT_METADATA } from "./explore"
 import { createFrontendUiUxEngineerAgent, FRONTEND_PROMPT_METADATA } from "./frontend-ui-ux-engineer"
 import { createDocumentWriterAgent, DOCUMENT_WRITER_PROMPT_METADATA } from "./document-writer"
 import { createMultimodalLookerAgent, MULTIMODAL_LOOKER_PROMPT_METADATA } from "./multimodal-looker"
+import { createMetisAgent } from "./metis"
+import { createOrchestratorSisyphusAgent, orchestratorSisyphusAgent } from "./orchestrator-sisyphus"
+import { createMomusAgent } from "./momus"
 import type { AvailableAgent } from "./sisyphus-prompt-builder"
 import { deepMerge } from "../shared"
+import { DEFAULT_CATEGORIES } from "../tools/sisyphus-task/constants"
+import { resolveMultipleSkills } from "../features/opencode-skill-loader/skill-content"
 
 type AgentSource = AgentFactory | AgentConfig
 
@@ -20,6 +25,9 @@ const agentSources: Record<BuiltinAgentName, AgentSource> = {
   "frontend-ui-ux-engineer": createFrontendUiUxEngineerAgent,
   "document-writer": createDocumentWriterAgent,
   "multimodal-looker": createMultimodalLookerAgent,
+  "Metis (Plan Consultant)": createMetisAgent,
+  "Momus (Plan Reviewer)": createMomusAgent,
+  "orchestrator-sisyphus": orchestratorSisyphusAgent,
 }
 
 /**
@@ -39,8 +47,31 @@ function isFactory(source: AgentSource): source is AgentFactory {
   return typeof source === "function"
 }
 
-function buildAgent(source: AgentSource, model?: string): AgentConfig {
-  return isFactory(source) ? source(model) : source
+export function buildAgent(source: AgentSource, model?: string): AgentConfig {
+  const base = isFactory(source) ? source(model) : source
+
+  const agentWithCategory = base as AgentConfig & { category?: string; skills?: string[] }
+  if (agentWithCategory.category) {
+    const categoryConfig = DEFAULT_CATEGORIES[agentWithCategory.category]
+    if (categoryConfig) {
+      if (!base.model) {
+        base.model = categoryConfig.model
+      }
+      if (base.temperature === undefined && categoryConfig.temperature !== undefined) {
+        base.temperature = categoryConfig.temperature
+      }
+    }
+  }
+
+  if (agentWithCategory.skills?.length) {
+    const { resolved } = resolveMultipleSkills(agentWithCategory.skills)
+    if (resolved.size > 0) {
+      const skillContent = Array.from(resolved.values()).join("\n\n")
+      base.prompt = skillContent + (base.prompt ? "\n\n" + base.prompt : "")
+    }
+  }
+
+  return base
 }
 
 /**
@@ -96,6 +127,7 @@ export function createBuiltinAgents(
     const agentName = name as BuiltinAgentName
 
     if (agentName === "Sisyphus") continue
+    if (agentName === "orchestrator-sisyphus") continue
     if (disabledAgents.includes(agentName)) continue
 
     const override = agentOverrides[agentName]
@@ -142,5 +174,20 @@ export function createBuiltinAgents(
     result["Sisyphus"] = sisyphusConfig
   }
 
+  if (!disabledAgents.includes("orchestrator-sisyphus")) {
+    const orchestratorOverride = agentOverrides["orchestrator-sisyphus"]
+    const orchestratorModel = orchestratorOverride?.model
+    let orchestratorConfig = createOrchestratorSisyphusAgent({
+      model: orchestratorModel,
+      availableAgents,
+    })
+
+    if (orchestratorOverride) {
+      orchestratorConfig = mergeAgentConfig(orchestratorConfig, orchestratorOverride)
+    }
+
+    result["orchestrator-sisyphus"] = orchestratorConfig
+  }
+
   return result
 }

src/auth/AGENTS.md

@@ -9,16 +9,20 @@ Google Antigravity OAuth for Gemini models. Token management, fetch interception
 ```
 auth/
 └── antigravity/
-    ├── plugin.ts         # Main export, hooks registration
+    ├── plugin.ts         # Main export, hooks registration (554 lines)
     ├── oauth.ts          # OAuth flow, token acquisition
     ├── token.ts          # Token storage, refresh logic
-    ├── fetch.ts          # Fetch interceptor (621 lines)
-    ├── response.ts       # Response transformation (598 lines)
-    ├── thinking.ts       # Thinking block extraction (571 lines)
+    ├── fetch.ts          # Fetch interceptor (798 lines)
+    ├── response.ts       # Response transformation (599 lines)
+    ├── thinking.ts       # Thinking block extraction (755 lines)
     ├── thought-signature-store.ts  # Signature caching
     ├── message-converter.ts        # Format conversion
+    ├── accounts.ts       # Multi-account management
+    ├── browser.ts        # Browser automation for OAuth
+    ├── cli.ts            # CLI interaction
     ├── request.ts        # Request building
     ├── project.ts        # Project ID management
+    ├── storage.ts        # Token persistence
     ├── tools.ts          # OAuth tool registration
     ├── constants.ts      # API endpoints, model mappings
     └── types.ts

src/auth/antigravity/constants.test.ts

@@ -0,0 +1,69 @@
+import { describe, it, expect } from "bun:test"
+import {
+  ANTIGRAVITY_TOKEN_REFRESH_BUFFER_MS,
+  ANTIGRAVITY_ENDPOINT_FALLBACKS,
+  ANTIGRAVITY_CALLBACK_PORT,
+} from "./constants"
+
+describe("Antigravity Constants", () => {
+  describe("ANTIGRAVITY_TOKEN_REFRESH_BUFFER_MS", () => {
+    it("should be 60 seconds (60,000ms) to refresh before expiry", () => {
+      // #given
+      const SIXTY_SECONDS_MS = 60 * 1000 // 60,000
+
+      // #when
+      const actual = ANTIGRAVITY_TOKEN_REFRESH_BUFFER_MS
+
+      // #then
+      expect(actual).toBe(SIXTY_SECONDS_MS)
+    })
+  })
+
+  describe("ANTIGRAVITY_ENDPOINT_FALLBACKS", () => {
+    it("should have exactly 3 endpoints (sandbox → daily → prod)", () => {
+      // #given
+      const expectedCount = 3
+
+      // #when
+      const actual = ANTIGRAVITY_ENDPOINT_FALLBACKS
+
+      // #then
+      expect(actual).toHaveLength(expectedCount)
+    })
+
+    it("should have sandbox endpoint first", () => {
+      // #then
+      expect(ANTIGRAVITY_ENDPOINT_FALLBACKS[0]).toBe(
+        "https://daily-cloudcode-pa.sandbox.googleapis.com"
+      )
+    })
+
+    it("should have daily endpoint second", () => {
+      // #then
+      expect(ANTIGRAVITY_ENDPOINT_FALLBACKS[1]).toBe(
+        "https://daily-cloudcode-pa.googleapis.com"
+      )
+    })
+
+    it("should have prod endpoint third", () => {
+      // #then
+      expect(ANTIGRAVITY_ENDPOINT_FALLBACKS[2]).toBe(
+        "https://cloudcode-pa.googleapis.com"
+      )
+    })
+
+    it("should NOT include autopush endpoint", () => {
+      // #then
+      const endpointsJoined = ANTIGRAVITY_ENDPOINT_FALLBACKS.join(",")
+      const hasAutopush = endpointsJoined.includes("autopush-cloudcode-pa")
+      expect(hasAutopush).toBe(false)
+    })
+  })
+
+  describe("ANTIGRAVITY_CALLBACK_PORT", () => {
+    it("should be 51121 to match CLIProxyAPI", () => {
+      // #then
+      expect(ANTIGRAVITY_CALLBACK_PORT).toBe(51121)
+    })
+  })
+})

src/auth/antigravity/oauth.test.ts

@@ -0,0 +1,262 @@
+import { describe, it, expect, beforeEach, afterEach, mock } from "bun:test"
+import { buildAuthURL, exchangeCode, startCallbackServer } from "./oauth"
+import { ANTIGRAVITY_CLIENT_ID, GOOGLE_TOKEN_URL, ANTIGRAVITY_CALLBACK_PORT } from "./constants"
+
+describe("OAuth PKCE Removal", () => {
+  describe("buildAuthURL", () => {
+    it("should NOT include code_challenge parameter", async () => {
+      // #given
+      const projectId = "test-project"
+
+      // #when
+      const result = await buildAuthURL(projectId)
+      const url = new URL(result.url)
+
+      // #then
+      expect(url.searchParams.has("code_challenge")).toBe(false)
+    })
+
+    it("should NOT include code_challenge_method parameter", async () => {
+      // #given
+      const projectId = "test-project"
+
+      // #when
+      const result = await buildAuthURL(projectId)
+      const url = new URL(result.url)
+
+      // #then
+      expect(url.searchParams.has("code_challenge_method")).toBe(false)
+    })
+
+    it("should include state parameter for CSRF protection", async () => {
+      // #given
+      const projectId = "test-project"
+
+      // #when
+      const result = await buildAuthURL(projectId)
+      const url = new URL(result.url)
+      const state = url.searchParams.get("state")
+
+      // #then
+      expect(state).toBeTruthy()
+    })
+
+    it("should have state as simple random string (not JSON/base64)", async () => {
+      // #given
+      const projectId = "test-project"
+
+      // #when
+      const result = await buildAuthURL(projectId)
+      const url = new URL(result.url)
+      const state = url.searchParams.get("state")!
+
+      // #then - positive assertions for simple random string
+      expect(state.length).toBeGreaterThanOrEqual(16)
+      expect(state.length).toBeLessThanOrEqual(64)
+      // Should be URL-safe (alphanumeric, no special chars like { } " :)
+      expect(state).toMatch(/^[a-zA-Z0-9_-]+$/)
+      // Should NOT contain JSON indicators
+      expect(state).not.toContain("{")
+      expect(state).not.toContain("}")
+      expect(state).not.toContain('"')
+    })
+
+    it("should include access_type=offline", async () => {
+      // #given
+      const projectId = "test-project"
+
+      // #when
+      const result = await buildAuthURL(projectId)
+      const url = new URL(result.url)
+
+      // #then
+      expect(url.searchParams.get("access_type")).toBe("offline")
+    })
+
+    it("should include prompt=consent", async () => {
+      // #given
+      const projectId = "test-project"
+
+      // #when
+      const result = await buildAuthURL(projectId)
+      const url = new URL(result.url)
+
+      // #then
+      expect(url.searchParams.get("prompt")).toBe("consent")
+    })
+
+    it("should NOT return verifier property (PKCE removed)", async () => {
+      // #given
+      const projectId = "test-project"
+
+      // #when
+      const result = await buildAuthURL(projectId)
+
+      // #then
+      expect(result).not.toHaveProperty("verifier")
+      expect(result).toHaveProperty("url")
+      expect(result).toHaveProperty("state")
+    })
+
+    it("should return state that matches URL state param", async () => {
+      // #given
+      const projectId = "test-project"
+
+      // #when
+      const result = await buildAuthURL(projectId)
+      const url = new URL(result.url)
+
+      // #then
+      expect(result.state).toBe(url.searchParams.get("state")!)
+    })
+  })
+
+  describe("exchangeCode", () => {
+    let originalFetch: typeof fetch
+
+    beforeEach(() => {
+      originalFetch = globalThis.fetch
+    })
+
+    afterEach(() => {
+      globalThis.fetch = originalFetch
+    })
+
+    it("should NOT send code_verifier in token exchange", async () => {
+      // #given
+      let capturedBody: string | null = null
+      globalThis.fetch = mock(async (url: string, init?: RequestInit) => {
+        if (url === GOOGLE_TOKEN_URL) {
+          capturedBody = init?.body as string
+          return new Response(JSON.stringify({
+            access_token: "test-access",
+            refresh_token: "test-refresh",
+            expires_in: 3600,
+            token_type: "Bearer"
+          }))
+        }
+        return new Response("", { status: 404 })
+      }) as unknown as typeof fetch
+
+      // #when
+      await exchangeCode("test-code", "http://localhost:51121/oauth-callback")
+
+      // #then
+      expect(capturedBody).toBeTruthy()
+      const params = new URLSearchParams(capturedBody!)
+      expect(params.has("code_verifier")).toBe(false)
+    })
+
+    it("should send required OAuth parameters", async () => {
+      // #given
+      let capturedBody: string | null = null
+      globalThis.fetch = mock(async (url: string, init?: RequestInit) => {
+        if (url === GOOGLE_TOKEN_URL) {
+          capturedBody = init?.body as string
+          return new Response(JSON.stringify({
+            access_token: "test-access",
+            refresh_token: "test-refresh",
+            expires_in: 3600,
+            token_type: "Bearer"
+          }))
+        }
+        return new Response("", { status: 404 })
+      }) as unknown as typeof fetch
+
+      // #when
+      await exchangeCode("test-code", "http://localhost:51121/oauth-callback")
+
+      // #then
+      const params = new URLSearchParams(capturedBody!)
+      expect(params.get("grant_type")).toBe("authorization_code")
+      expect(params.get("code")).toBe("test-code")
+      expect(params.get("client_id")).toBe(ANTIGRAVITY_CLIENT_ID)
+      expect(params.get("redirect_uri")).toBe("http://localhost:51121/oauth-callback")
+    })
+  })
+
+  describe("State/CSRF Validation", () => {
+    it("should generate unique state for each call", async () => {
+      // #given
+      const projectId = "test-project"
+
+      // #when
+      const result1 = await buildAuthURL(projectId)
+      const result2 = await buildAuthURL(projectId)
+
+      // #then
+      expect(result1.state).not.toBe(result2.state)
+    })
+  })
+
+  describe("startCallbackServer Port Handling", () => {
+    it("should prefer port 51121", () => {
+      // #given
+      // Port 51121 should be free
+
+      // #when
+      const handle = startCallbackServer()
+
+      // #then
+      // If 51121 is available, should use it
+      // If not available, should use valid fallback
+      expect(handle.port).toBeGreaterThan(0)
+      expect(handle.port).toBeLessThan(65536)
+      handle.close()
+    })
+
+    it("should return actual bound port", () => {
+      // #when
+      const handle = startCallbackServer()
+
+      // #then
+      expect(typeof handle.port).toBe("number")
+      expect(handle.port).toBeGreaterThan(0)
+      handle.close()
+    })
+
+    it("should fallback to OS-assigned port if 51121 is occupied (EADDRINUSE)", async () => {
+      // #given - Occupy port 51121 first
+      const blocker = Bun.serve({
+        port: ANTIGRAVITY_CALLBACK_PORT,
+        fetch: () => new Response("blocked")
+      })
+
+      try {
+        // #when
+        const handle = startCallbackServer()
+
+        // #then
+        expect(handle.port).not.toBe(ANTIGRAVITY_CALLBACK_PORT)
+        expect(handle.port).toBeGreaterThan(0)
+        handle.close()
+      } finally {
+        // Cleanup blocker
+        blocker.stop()
+      }
+    })
+
+    it("should cleanup server on close", () => {
+      // #given
+      const handle = startCallbackServer()
+      const port = handle.port
+
+      // #when
+      handle.close()
+
+      // #then - port should be released (can bind again)
+      const testServer = Bun.serve({ port, fetch: () => new Response("test") })
+      expect(testServer.port).toBe(port)
+      testServer.stop()
+    })
+
+    it("should provide redirect URI with actual port", () => {
+      // #given
+      const handle = startCallbackServer()
+
+      // #then
+      expect(handle.redirectUri).toBe(`http://localhost:${handle.port}/oauth-callback`)
+      handle.close()
+    })
+  })
+})

src/auth/antigravity/oauth.ts

@@ -1,9 +1,7 @@
 /**
- * Antigravity OAuth 2.0 flow implementation with PKCE.
+ * Antigravity OAuth 2.0 flow implementation.
  * Handles Google OAuth for Antigravity authentication.
  */
-import { generatePKCE } from "@openauthjs/openauth/pkce"
-
 import {
   ANTIGRAVITY_CLIENT_ID,
   ANTIGRAVITY_CLIENT_SECRET,
@@ -19,37 +17,14 @@ import type {
   AntigravityUserInfo,
 } from "./types"
 
-/**
- * PKCE pair containing verifier and challenge.
- */
-export interface PKCEPair {
-  /** PKCE verifier - used during token exchange */
-  verifier: string
-  /** PKCE challenge - sent in auth URL */
-  challenge: string
-  /** Challenge method - always "S256" */
-  method: string
-}
-
-/**
- * OAuth state encoded in the auth URL.
- * Contains the PKCE verifier for later retrieval.
- */
-export interface OAuthState {
-  /** PKCE verifier */
-  verifier: string
-  /** Optional project ID */
-  projectId?: string
-}
-
 /**
  * Result from building an OAuth authorization URL.
  */
 export interface AuthorizationResult {
   /** Full OAuth URL to open in browser */
   url: string
-  /** PKCE verifier to use during code exchange */
-  verifier: string
+  /** State for CSRF protection */
+  state: string
 }
 
 /**
@@ -64,70 +39,12 @@ export interface CallbackResult {
   error?: string
 }
 
-/**
- * Generate PKCE verifier and challenge pair.
- * Uses @openauthjs/openauth for cryptographically secure generation.
- *
- * @returns PKCE pair with verifier, challenge, and method
- */
-export async function generatePKCEPair(): Promise<PKCEPair> {
-  const pkce = await generatePKCE()
-  return {
-    verifier: pkce.verifier,
-    challenge: pkce.challenge,
-    method: pkce.method,
-  }
-}
-
-/**
- * Encode OAuth state into a URL-safe base64 string.
- *
- * @param state - OAuth state object
- * @returns Base64URL encoded state
- */
-function encodeState(state: OAuthState): string {
-  const json = JSON.stringify(state)
-  return Buffer.from(json, "utf8").toString("base64url")
-}
-
-/**
- * Decode OAuth state from a base64 string.
- *
- * @param encoded - Base64URL or Base64 encoded state
- * @returns Decoded OAuth state
- */
-export function decodeState(encoded: string): OAuthState {
-  // Handle both base64url and standard base64
-  const normalized = encoded.replace(/-/g, "+").replace(/_/g, "/")
-  const padded = normalized.padEnd(
-    normalized.length + ((4 - (normalized.length % 4)) % 4),
-    "="
-  )
-  const json = Buffer.from(padded, "base64").toString("utf8")
-  const parsed = JSON.parse(json)
-
-  if (typeof parsed.verifier !== "string") {
-    throw new Error("Missing PKCE verifier in state")
-  }
-
-  return {
-    verifier: parsed.verifier,
-    projectId:
-      typeof parsed.projectId === "string" ? parsed.projectId : undefined,
-  }
-}
-
 export async function buildAuthURL(
   projectId?: string,
   clientId: string = ANTIGRAVITY_CLIENT_ID,
   port: number = ANTIGRAVITY_CALLBACK_PORT
 ): Promise<AuthorizationResult> {
-  const pkce = await generatePKCEPair()
-
-  const state: OAuthState = {
-    verifier: pkce.verifier,
-    projectId,
-  }
+  const state = crypto.randomUUID().replace(/-/g, "")
 
   const redirectUri = `http://localhost:${port}/oauth-callback`
 
@@ -136,15 +53,13 @@ export async function buildAuthURL(
   url.searchParams.set("redirect_uri", redirectUri)
   url.searchParams.set("response_type", "code")
   url.searchParams.set("scope", ANTIGRAVITY_SCOPES.join(" "))
-  url.searchParams.set("state", encodeState(state))
-  url.searchParams.set("code_challenge", pkce.challenge)
-  url.searchParams.set("code_challenge_method", "S256")
+  url.searchParams.set("state", state)
   url.searchParams.set("access_type", "offline")
   url.searchParams.set("prompt", "consent")
 
   return {
     url: url.toString(),
-    verifier: pkce.verifier,
+    state,
   }
 }
 
@@ -152,26 +67,23 @@ export async function buildAuthURL(
  * Exchange authorization code for tokens.
  *
  * @param code - Authorization code from OAuth callback
- * @param verifier - PKCE verifier from initial auth request
+ * @param redirectUri - OAuth redirect URI
  * @param clientId - Optional custom client ID (defaults to ANTIGRAVITY_CLIENT_ID)
  * @param clientSecret - Optional custom client secret (defaults to ANTIGRAVITY_CLIENT_SECRET)
  * @returns Token exchange result with access and refresh tokens
  */
 export async function exchangeCode(
   code: string,
-  verifier: string,
+  redirectUri: string,
   clientId: string = ANTIGRAVITY_CLIENT_ID,
-  clientSecret: string = ANTIGRAVITY_CLIENT_SECRET,
-  port: number = ANTIGRAVITY_CALLBACK_PORT
+  clientSecret: string = ANTIGRAVITY_CLIENT_SECRET
 ): Promise<AntigravityTokenExchangeResult> {
-  const redirectUri = `http://localhost:${port}/oauth-callback`
   const params = new URLSearchParams({
     client_id: clientId,
     client_secret: clientSecret,
     code,
     grant_type: "authorization_code",
     redirect_uri: redirectUri,
-    code_verifier: verifier,
   })
 
   const response = await fetch(GOOGLE_TOKEN_URL, {
@@ -236,6 +148,7 @@ export async function fetchUserInfo(
 
 export interface CallbackServerHandle {
   port: number
+  redirectUri: string
   waitForCallback: () => Promise<CallbackResult>
   close: () => void
 }
@@ -259,43 +172,53 @@ export function startCallbackServer(
     }
   }
 
-  server = Bun.serve({
-    port: 0,
-    fetch(request: Request): Response {
-      const url = new URL(request.url)
+  const fetchHandler = (request: Request): Response => {
+    const url = new URL(request.url)
 
-      if (url.pathname === "/oauth-callback") {
-        const code = url.searchParams.get("code") || ""
-        const state = url.searchParams.get("state") || ""
-        const error = url.searchParams.get("error") || undefined
+    if (url.pathname === "/oauth-callback") {
+      const code = url.searchParams.get("code") || ""
+      const state = url.searchParams.get("state") || ""
+      const error = url.searchParams.get("error") || undefined
 
-        let responseBody: string
-        if (code && !error) {
-          responseBody =
-            "<html><body><h1>Login successful</h1><p>You can close this window.</p></body></html>"
-        } else {
-          responseBody =
-            "<html><body><h1>Login failed</h1><p>Please check the CLI output.</p></body></html>"
-        }
-
-        setTimeout(() => {
-          cleanup()
-          if (resolveCallback) {
-            resolveCallback({ code, state, error })
-          }
-        }, 100)
-
-        return new Response(responseBody, {
-          status: 200,
-          headers: { "Content-Type": "text/html" },
-        })
+      let responseBody: string
+      if (code && !error) {
+        responseBody =
+          "<html><body><h1>Login successful</h1><p>You can close this window.</p></body></html>"
+      } else {
+        responseBody =
+          "<html><body><h1>Login failed</h1><p>Please check the CLI output.</p></body></html>"
       }
 
-      return new Response("Not Found", { status: 404 })
-    },
-  })
+      setTimeout(() => {
+        cleanup()
+        if (resolveCallback) {
+          resolveCallback({ code, state, error })
+        }
+      }, 100)
+
+      return new Response(responseBody, {
+        status: 200,
+        headers: { "Content-Type": "text/html" },
+      })
+    }
+
+    return new Response("Not Found", { status: 404 })
+  }
+
+  try {
+    server = Bun.serve({
+      port: ANTIGRAVITY_CALLBACK_PORT,
+      fetch: fetchHandler,
+    })
+  } catch (error) {
+    server = Bun.serve({
+      port: 0,
+      fetch: fetchHandler,
+    })
+  }
 
   const actualPort = server.port as number
+  const redirectUri = `http://localhost:${actualPort}/oauth-callback`
 
   const waitForCallback = (): Promise<CallbackResult> => {
     return new Promise((resolve, reject) => {
@@ -311,6 +234,7 @@ export function startCallbackServer(
 
   return {
     port: actualPort,
+    redirectUri,
     waitForCallback,
     close: cleanup,
   }
@@ -324,7 +248,7 @@ export async function performOAuthFlow(
 ): Promise<{
   tokens: AntigravityTokenExchangeResult
   userInfo: AntigravityUserInfo
-  verifier: string
+  state: string
 }> {
   const serverHandle = startCallbackServer()
 
@@ -345,15 +269,15 @@ export async function performOAuthFlow(
       throw new Error("No authorization code received")
     }
 
-    const state = decodeState(callback.state)
-    if (state.verifier !== auth.verifier) {
-      throw new Error("PKCE verifier mismatch - possible CSRF attack")
+    if (callback.state !== auth.state) {
+      throw new Error("State mismatch - possible CSRF attack")
     }
 
-    const tokens = await exchangeCode(callback.code, auth.verifier, clientId, clientSecret, serverHandle.port)
+    const redirectUri = `http://localhost:${serverHandle.port}/oauth-callback`
+    const tokens = await exchangeCode(callback.code, redirectUri, clientId, clientSecret)
     const userInfo = await fetchUserInfo(tokens.access_token)
 
-    return { tokens, userInfo, verifier: auth.verifier }
+    return { tokens, userInfo, state: auth.state }
   } catch (err) {
     serverHandle.close()
     throw err

src/auth/antigravity/plugin.ts

@@ -33,7 +33,6 @@ import {
   exchangeCode,
   startCallbackServer,
   fetchUserInfo,
-  decodeState,
 } from "./oauth"
 import { createAntigravityFetch } from "./fetch"
 import { fetchProjectContext } from "./project"
@@ -248,7 +247,7 @@ export async function createGoogleAntigravityAuthPlugin({
          */
         authorize: async (): Promise<AuthOuathResult> => {
           const serverHandle = startCallbackServer()
-          const { url, verifier } = await buildAuthURL(undefined, cachedClientId, serverHandle.port)
+          const { url, state: expectedState } = await buildAuthURL(undefined, cachedClientId, serverHandle.port)
 
           const browserOpened = await openBrowserURL(url)
 
@@ -277,15 +276,15 @@ export async function createGoogleAntigravityAuthPlugin({
                   return { type: "failed" as const }
                 }
 
-                const state = decodeState(result.state)
-                if (state.verifier !== verifier) {
+                if (result.state !== expectedState) {
                   if (process.env.ANTIGRAVITY_DEBUG === "1") {
-                    console.error("[antigravity-plugin] PKCE verifier mismatch")
+                    console.error("[antigravity-plugin] State mismatch - possible CSRF attack")
                   }
                   return { type: "failed" as const }
                 }
 
-                const tokens = await exchangeCode(result.code, verifier, cachedClientId, cachedClientSecret, serverHandle.port)
+                const redirectUri = `http://localhost:${serverHandle.port}/oauth-callback`
+                const tokens = await exchangeCode(result.code, redirectUri, cachedClientId, cachedClientSecret)
 
                 if (!tokens.refresh_token) {
                   serverHandle.close()
@@ -343,7 +342,7 @@ export async function createGoogleAntigravityAuthPlugin({
                   if (!addAnother) break
 
                   const additionalServerHandle = startCallbackServer()
-                  const { url: additionalUrl, verifier: additionalVerifier } = await buildAuthURL(
+                  const { url: additionalUrl, state: expectedAdditionalState } = await buildAuthURL(
                     undefined,
                     cachedClientId,
                     additionalServerHandle.port
@@ -373,24 +372,23 @@ export async function createGoogleAntigravityAuthPlugin({
                       continue
                     }
 
-                    const additionalState = decodeState(additionalResult.state)
-                    if (additionalState.verifier !== additionalVerifier) {
+                    if (additionalResult.state !== expectedAdditionalState) {
                       additionalServerHandle.close()
                       await client.tui.showToast({
                         body: {
-                          message: "Verification failed, skipping...",
+                          message: "State mismatch, skipping...",
                           variant: "warning",
                         },
                       })
                       continue
                     }
 
+                    const additionalRedirectUri = `http://localhost:${additionalServerHandle.port}/oauth-callback`
                     const additionalTokens = await exchangeCode(
                       additionalResult.code,
-                      additionalVerifier,
+                      additionalRedirectUri,
                       cachedClientId,
-                      cachedClientSecret,
-                      additionalServerHandle.port
+                      cachedClientSecret
                     )
 
                     if (!additionalTokens.refresh_token) {

src/auth/antigravity/token.test.ts

@@ -0,0 +1,78 @@
+import { describe, it, expect } from "bun:test"
+import { isTokenExpired } from "./token"
+import type { AntigravityTokens } from "./types"
+
+describe("Token Expiry with 60-second Buffer", () => {
+  const createToken = (expiresInSeconds: number): AntigravityTokens => ({
+    type: "antigravity",
+    access_token: "test-access",
+    refresh_token: "test-refresh",
+    expires_in: expiresInSeconds,
+    timestamp: Date.now(),
+  })
+
+  it("should NOT be expired if token expires in 2 minutes", () => {
+    // #given
+    const twoMinutes = 2 * 60
+    const token = createToken(twoMinutes)
+
+    // #when
+    const expired = isTokenExpired(token)
+
+    // #then
+    expect(expired).toBe(false)
+  })
+
+  it("should be expired if token expires in 30 seconds", () => {
+    // #given
+    const thirtySeconds = 30
+    const token = createToken(thirtySeconds)
+
+    // #when
+    const expired = isTokenExpired(token)
+
+    // #then
+    expect(expired).toBe(true)
+  })
+
+  it("should be expired at exactly 60 seconds (boundary)", () => {
+    // #given
+    const sixtySeconds = 60
+    const token = createToken(sixtySeconds)
+
+    // #when
+    const expired = isTokenExpired(token)
+
+    // #then - at boundary, should trigger refresh
+    expect(expired).toBe(true)
+  })
+
+  it("should be expired if token already expired", () => {
+    // #given
+    const alreadyExpired: AntigravityTokens = {
+      type: "antigravity",
+      access_token: "test-access",
+      refresh_token: "test-refresh",
+      expires_in: 3600,
+      timestamp: Date.now() - 4000 * 1000,
+    }
+
+    // #when
+    const expired = isTokenExpired(alreadyExpired)
+
+    // #then
+    expect(expired).toBe(true)
+  })
+
+  it("should NOT be expired if token has plenty of time", () => {
+    // #given
+    const twoHours = 2 * 60 * 60
+    const token = createToken(twoHours)
+
+    // #when
+    const expired = isTokenExpired(token)
+
+    // #then
+    expect(expired).toBe(false)
+  })
+})

src/cli/AGENTS.md

@@ -9,16 +9,20 @@ CLI for oh-my-opencode: interactive installer, health diagnostics (doctor), runt
 ```
 cli/
 ├── index.ts              # Commander.js entry, subcommand routing
-├── install.ts            # Interactive TUI installer (477 lines)
-├── config-manager.ts     # JSONC parsing, env detection (669 lines)
+├── install.ts            # Interactive TUI installer (436 lines)
+├── config-manager.ts     # JSONC parsing, env detection (725 lines)
 ├── types.ts              # CLI-specific types
+├── commands/             # CLI subcommands
 ├── doctor/               # Health check system
 │   ├── index.ts          # Doctor command entry
+│   ├── runner.ts         # Health check orchestration
 │   ├── constants.ts      # Check categories
 │   ├── types.ts          # Check result interfaces
-│   └── checks/           # 17+ individual checks
+│   └── checks/           # 17+ individual checks (auth, config, dependencies, gh, lsp, mcp, opencode, plugin, version)
 ├── get-local-version/    # Version detection
 └── run/                  # OpenCode session launcher
+    ├── completion.ts     # Completion logic
+    └── events.ts         # Event handling
 ```
 
 ## CLI COMMANDS

src/cli/config-manager.ts

@@ -310,6 +310,15 @@ export function generateOmoConfig(installConfig: InstallConfig): Record<string,
     config.agents = agents
   }
 
+  // Categories: override model for Antigravity auth (gemini-3-pro-preview → gemini-3-pro-high)
+  if (installConfig.hasGemini) {
+    config.categories = {
+      "visual-engineering": { model: "google/gemini-3-pro-high" },
+      artistry: { model: "google/gemini-3-pro-high" },
+      writing: { model: "google/gemini-3-flash-high" },
+    }
+  }
+
   return config
 }
 

src/config/schema.test.ts

@@ -1,5 +1,5 @@
 import { describe, expect, test } from "bun:test"
-import { OhMyOpenCodeConfigSchema } from "./schema"
+import { AgentOverrideConfigSchema, BuiltinCategoryNameSchema, OhMyOpenCodeConfigSchema } from "./schema"
 
 describe("disabled_mcps schema", () => {
   test("should accept built-in MCP names", () => {
@@ -134,3 +134,184 @@ describe("disabled_mcps schema", () => {
     }
   })
 })
+
+describe("AgentOverrideConfigSchema", () => {
+  describe("category field", () => {
+    test("accepts category as optional string", () => {
+      // #given
+      const config = { category: "visual-engineering" }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(true)
+      if (result.success) {
+        expect(result.data.category).toBe("visual-engineering")
+      }
+    })
+
+    test("accepts config without category", () => {
+      // #given
+      const config = { temperature: 0.5 }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(true)
+    })
+
+    test("rejects non-string category", () => {
+      // #given
+      const config = { category: 123 }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(false)
+    })
+  })
+
+  describe("skills field", () => {
+    test("accepts skills as optional string array", () => {
+      // #given
+      const config = { skills: ["frontend-ui-ux", "code-reviewer"] }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(true)
+      if (result.success) {
+        expect(result.data.skills).toEqual(["frontend-ui-ux", "code-reviewer"])
+      }
+    })
+
+    test("accepts empty skills array", () => {
+      // #given
+      const config = { skills: [] }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(true)
+      if (result.success) {
+        expect(result.data.skills).toEqual([])
+      }
+    })
+
+    test("accepts config without skills", () => {
+      // #given
+      const config = { temperature: 0.5 }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(true)
+    })
+
+    test("rejects non-array skills", () => {
+      // #given
+      const config = { skills: "frontend-ui-ux" }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(false)
+    })
+  })
+
+  describe("backward compatibility", () => {
+    test("still accepts model field (deprecated)", () => {
+      // #given
+      const config = { model: "openai/gpt-5.2" }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(true)
+      if (result.success) {
+        expect(result.data.model).toBe("openai/gpt-5.2")
+      }
+    })
+
+    test("accepts both model and category (deprecated usage)", () => {
+      // #given - category should take precedence at runtime, but both should validate
+      const config = { 
+        model: "openai/gpt-5.2",
+        category: "ultrabrain"
+      }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(true)
+      if (result.success) {
+        expect(result.data.model).toBe("openai/gpt-5.2")
+        expect(result.data.category).toBe("ultrabrain")
+      }
+    })
+  })
+
+  describe("combined fields", () => {
+    test("accepts category with skills", () => {
+      // #given
+      const config = { 
+        category: "visual-engineering",
+        skills: ["frontend-ui-ux"]
+      }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(true)
+      if (result.success) {
+        expect(result.data.category).toBe("visual-engineering")
+        expect(result.data.skills).toEqual(["frontend-ui-ux"])
+      }
+    })
+
+    test("accepts category with skills and other fields", () => {
+      // #given
+      const config = { 
+        category: "ultrabrain",
+        skills: ["code-reviewer"],
+        temperature: 0.3,
+        prompt_append: "Extra instructions"
+      }
+
+      // #when
+      const result = AgentOverrideConfigSchema.safeParse(config)
+
+      // #then
+      expect(result.success).toBe(true)
+      if (result.success) {
+        expect(result.data.category).toBe("ultrabrain")
+        expect(result.data.skills).toEqual(["code-reviewer"])
+        expect(result.data.temperature).toBe(0.3)
+        expect(result.data.prompt_append).toBe("Extra instructions")
+      }
+    })
+  })
+})
+
+describe("BuiltinCategoryNameSchema", () => {
+  test("accepts all builtin category names", () => {
+    // #given
+    const categories = ["visual-engineering", "ultrabrain", "artistry", "quick", "most-capable", "writing", "general"]
+
+    // #when / #then
+    for (const cat of categories) {
+      const result = BuiltinCategoryNameSchema.safeParse(cat)
+      expect(result.success).toBe(true)
+    }
+  })
+})

src/config/schema.ts

@@ -24,10 +24,15 @@ export const BuiltinAgentNameSchema = z.enum([
   "frontend-ui-ux-engineer",
   "document-writer",
   "multimodal-looker",
+  "Metis (Plan Consultant)",
+  "Momus (Plan Reviewer)",
+  "orchestrator-sisyphus",
 ])
 
 export const BuiltinSkillNameSchema = z.enum([
   "playwright",
+  "frontend-ui-ux",
+  "git-master",
 ])
 
 export const OverridableAgentNameSchema = z.enum([
@@ -35,13 +40,16 @@ export const OverridableAgentNameSchema = z.enum([
   "plan",
   "Sisyphus",
   "OpenCode-Builder",
-  "Planner-Sisyphus",
+  "Prometheus (Planner)",
+  "Metis (Plan Consultant)",
+  "Momus (Plan Reviewer)",
   "oracle",
   "librarian",
   "explore",
   "frontend-ui-ux-engineer",
   "document-writer",
   "multimodal-looker",
+  "orchestrator-sisyphus",
 ])
 
 export const AgentNameSchema = BuiltinAgentNameSchema
@@ -75,14 +83,23 @@ export const HookNameSchema = z.enum([
   "claude-code-hooks",
   "auto-slash-command",
   "edit-error-recovery",
+  "prometheus-md-only",
+  "start-work",
+  "sisyphus-orchestrator",
 ])
 
 export const BuiltinCommandNameSchema = z.enum([
   "init-deep",
+  "start-work",
 ])
 
 export const AgentOverrideConfigSchema = z.object({
+  /** @deprecated Use `category` instead. Model is inherited from category defaults. */
   model: z.string().optional(),
+  /** Category name to inherit model and other settings from CategoryConfig */
+  category: z.string().optional(),
+  /** Skill names to inject into agent prompt */
+  skills: z.array(z.string()).optional(),
   temperature: z.number().min(0).max(2).optional(),
   top_p: z.number().min(0).max(1).optional(),
   prompt: z.string().optional(),
@@ -103,13 +120,16 @@ export const AgentOverridesSchema = z.object({
   plan: AgentOverrideConfigSchema.optional(),
   Sisyphus: AgentOverrideConfigSchema.optional(),
   "OpenCode-Builder": AgentOverrideConfigSchema.optional(),
-  "Planner-Sisyphus": AgentOverrideConfigSchema.optional(),
+  "Prometheus (Planner)": AgentOverrideConfigSchema.optional(),
+  "Metis (Plan Consultant)": AgentOverrideConfigSchema.optional(),
+  "Momus (Plan Reviewer)": AgentOverrideConfigSchema.optional(),
   oracle: AgentOverrideConfigSchema.optional(),
   librarian: AgentOverrideConfigSchema.optional(),
   explore: AgentOverrideConfigSchema.optional(),
   "frontend-ui-ux-engineer": AgentOverrideConfigSchema.optional(),
   "document-writer": AgentOverrideConfigSchema.optional(),
   "multimodal-looker": AgentOverrideConfigSchema.optional(),
+  "orchestrator-sisyphus": AgentOverrideConfigSchema.optional(),
 })
 
 export const ClaudeCodeConfigSchema = z.object({
@@ -129,6 +149,33 @@ export const SisyphusAgentConfigSchema = z.object({
   replace_plan: z.boolean().optional(),
 })
 
+export const CategoryConfigSchema = z.object({
+  model: z.string(),
+  temperature: z.number().min(0).max(2).optional(),
+  top_p: z.number().min(0).max(1).optional(),
+  maxTokens: z.number().optional(),
+  thinking: z.object({
+    type: z.enum(["enabled", "disabled"]),
+    budgetTokens: z.number().optional(),
+  }).optional(),
+  reasoningEffort: z.enum(["low", "medium", "high"]).optional(),
+  textVerbosity: z.enum(["low", "medium", "high"]).optional(),
+  tools: z.record(z.string(), z.boolean()).optional(),
+  prompt_append: z.string().optional(),
+})
+
+export const BuiltinCategoryNameSchema = z.enum([
+  "visual-engineering",
+  "ultrabrain",
+  "artistry",
+  "quick",
+  "most-capable",
+  "writing",
+  "general",
+])
+
+export const CategoriesConfigSchema = z.record(z.string(), CategoryConfigSchema)
+
 export const CommentCheckerConfigSchema = z.object({
   /** Custom prompt to replace the default warning message. Use {{comments}} placeholder for detected comments XML. */
   custom_prompt: z.string().optional(),
@@ -243,6 +290,13 @@ export const NotificationConfigSchema = z.object({
   force_enable: z.boolean().optional(),
 })
 
+export const GitMasterConfigSchema = z.object({
+  /** Add "Ultraworked with Sisyphus" footer to commit messages (default: true) */
+  commit_footer: z.boolean().default(true),
+  /** Add "Co-authored-by: Sisyphus" trailer to commit messages (default: true) */
+  include_co_authored_by: z.boolean().default(true),
+})
+
 export const OhMyOpenCodeConfigSchema = z.object({
   $schema: z.string().optional(),
   disabled_mcps: z.array(AnyMcpNameSchema).optional(),
@@ -251,6 +305,7 @@ export const OhMyOpenCodeConfigSchema = z.object({
   disabled_hooks: z.array(HookNameSchema).optional(),
   disabled_commands: z.array(BuiltinCommandNameSchema).optional(),
   agents: AgentOverridesSchema.optional(),
+  categories: CategoriesConfigSchema.optional(),
   claude_code: ClaudeCodeConfigSchema.optional(),
   google_auth: z.boolean().optional(),
   sisyphus_agent: SisyphusAgentConfigSchema.optional(),
@@ -261,6 +316,7 @@ export const OhMyOpenCodeConfigSchema = z.object({
   ralph_loop: RalphLoopConfigSchema.optional(),
   background_task: BackgroundTaskConfigSchema.optional(),
   notification: NotificationConfigSchema.optional(),
+  git_master: GitMasterConfigSchema.optional(),
 })
 
 export type OhMyOpenCodeConfig = z.infer<typeof OhMyOpenCodeConfigSchema>
@@ -279,5 +335,9 @@ export type SkillsConfig = z.infer<typeof SkillsConfigSchema>
 export type SkillDefinition = z.infer<typeof SkillDefinitionSchema>
 export type RalphLoopConfig = z.infer<typeof RalphLoopConfigSchema>
 export type NotificationConfig = z.infer<typeof NotificationConfigSchema>
+export type CategoryConfig = z.infer<typeof CategoryConfigSchema>
+export type CategoriesConfig = z.infer<typeof CategoriesConfigSchema>
+export type BuiltinCategoryName = z.infer<typeof BuiltinCategoryNameSchema>
+export type GitMasterConfig = z.infer<typeof GitMasterConfigSchema>
 
 export { AnyMcpNameSchema, type AnyMcpName, McpNameSchema, type McpName } from "../mcp/types"

src/features/AGENTS.md

@@ -8,17 +8,23 @@ Claude Code compatibility layer + core feature modules. Commands, skills, agents
 
 ```
 features/
-├── background-agent/           # Task lifecycle, notifications (460 lines)
+├── background-agent/           # Task lifecycle, notifications (608 lines)
+├── boulder-state/              # Boulder state persistence
 ├── builtin-commands/           # Built-in slash commands
-├── builtin-skills/             # Built-in skills (playwright)
+│   └── templates/              # start-work, refactor, init-deep, ralph-loop
+├── builtin-skills/             # Built-in skills
+│   ├── git-master/             # Atomic commits, rebase, history search
+│   └── frontend-ui-ux/         # Designer-turned-developer skill
 ├── claude-code-agent-loader/   # ~/.claude/agents/*.md
 ├── claude-code-command-loader/ # ~/.claude/commands/*.md
 ├── claude-code-mcp-loader/     # .mcp.json files
 │   └── env-expander.ts         # ${VAR} expansion
-├── claude-code-plugin-loader/  # installed_plugins.json (484 lines)
+├── claude-code-plugin-loader/  # installed_plugins.json (486 lines)
 ├── claude-code-session-state/  # Session state persistence
+├── context-injector/           # Context collection and injection
 ├── opencode-skill-loader/      # Skills from OpenCode + Claude paths
 ├── skill-mcp-manager/          # MCP servers in skill YAML
+├── task-toast-manager/         # Task toast notifications
 └── hook-message-injector/      # Inject messages into conversation
 ```
 

src/features/background-agent/manager.test.ts

@@ -1,11 +1,12 @@
 import { describe, test, expect, beforeEach } from "bun:test"
-import type { BackgroundTask } from "./types"
+import type { BackgroundTask, ResumeInput } from "./types"
 
 const TASK_TTL_MS = 30 * 60 * 1000
 
 class MockBackgroundManager {
   private tasks: Map<string, BackgroundTask> = new Map()
   private notifications: Map<string, BackgroundTask[]> = new Map()
+  public resumeCalls: Array<{ sessionId: string; prompt: string }> = []
 
   addTask(task: BackgroundTask): void {
     this.tasks.set(task.id, task)
@@ -15,6 +16,15 @@ class MockBackgroundManager {
     return this.tasks.get(id)
   }
 
+  findBySession(sessionID: string): BackgroundTask | undefined {
+    for (const task of this.tasks.values()) {
+      if (task.sessionID === sessionID) {
+        return task
+      }
+    }
+    return undefined
+  }
+
   getTasksByParentSession(sessionID: string): BackgroundTask[] {
     const result: BackgroundTask[] = []
     for (const task of this.tasks.values()) {
@@ -105,6 +115,29 @@ class MockBackgroundManager {
     }
     return count
   }
+
+  resume(input: ResumeInput): BackgroundTask {
+    const existingTask = this.findBySession(input.sessionId)
+    if (!existingTask) {
+      throw new Error(`Task not found for session: ${input.sessionId}`)
+    }
+
+    this.resumeCalls.push({ sessionId: input.sessionId, prompt: input.prompt })
+
+    existingTask.status = "running"
+    existingTask.completedAt = undefined
+    existingTask.error = undefined
+    existingTask.parentSessionID = input.parentSessionID
+    existingTask.parentMessageID = input.parentMessageID
+    existingTask.parentModel = input.parentModel
+
+    existingTask.progress = {
+      toolCalls: existingTask.progress?.toolCalls ?? 0,
+      lastUpdate: new Date(),
+    }
+
+    return existingTask
+  }
 }
 
 function createMockTask(overrides: Partial<BackgroundTask> & { id: string; sessionID: string; parentSessionID: string }): BackgroundTask {
@@ -482,3 +515,254 @@ describe("BackgroundManager.pruneStaleTasksAndNotifications", () => {
     expect(manager.getTask("task-fresh")).toBeDefined()
   })
 })
+
+describe("BackgroundManager.resume", () => {
+  let manager: MockBackgroundManager
+
+  beforeEach(() => {
+    // #given
+    manager = new MockBackgroundManager()
+  })
+
+  test("should throw error when task not found", () => {
+    // #given - empty manager
+
+    // #when / #then
+    expect(() => manager.resume({
+      sessionId: "non-existent",
+      prompt: "continue",
+      parentSessionID: "session-new",
+      parentMessageID: "msg-new",
+    })).toThrow("Task not found for session: non-existent")
+  })
+
+  test("should resume existing task and reset state to running", () => {
+    // #given
+    const completedTask = createMockTask({
+      id: "task-a",
+      sessionID: "session-a",
+      parentSessionID: "session-parent",
+      status: "completed",
+    })
+    completedTask.completedAt = new Date()
+    completedTask.error = "previous error"
+    manager.addTask(completedTask)
+
+    // #when
+    const result = manager.resume({
+      sessionId: "session-a",
+      prompt: "continue the work",
+      parentSessionID: "session-new-parent",
+      parentMessageID: "msg-new",
+    })
+
+    // #then
+    expect(result.status).toBe("running")
+    expect(result.completedAt).toBeUndefined()
+    expect(result.error).toBeUndefined()
+    expect(result.parentSessionID).toBe("session-new-parent")
+    expect(result.parentMessageID).toBe("msg-new")
+  })
+
+  test("should preserve task identity while updating parent context", () => {
+    // #given
+    const existingTask = createMockTask({
+      id: "task-a",
+      sessionID: "session-a",
+      parentSessionID: "old-parent",
+      description: "original description",
+      agent: "explore",
+    })
+    manager.addTask(existingTask)
+
+    // #when
+    const result = manager.resume({
+      sessionId: "session-a",
+      prompt: "new prompt",
+      parentSessionID: "new-parent",
+      parentMessageID: "new-msg",
+      parentModel: { providerID: "anthropic", modelID: "claude-opus" },
+    })
+
+    // #then
+    expect(result.id).toBe("task-a")
+    expect(result.sessionID).toBe("session-a")
+    expect(result.description).toBe("original description")
+    expect(result.agent).toBe("explore")
+    expect(result.parentModel).toEqual({ providerID: "anthropic", modelID: "claude-opus" })
+  })
+
+  test("should track resume calls with prompt", () => {
+    // #given
+    const task = createMockTask({
+      id: "task-a",
+      sessionID: "session-a",
+      parentSessionID: "session-parent",
+    })
+    manager.addTask(task)
+
+    // #when
+    manager.resume({
+      sessionId: "session-a",
+      prompt: "continue with additional context",
+      parentSessionID: "session-new",
+      parentMessageID: "msg-new",
+    })
+
+    // #then
+    expect(manager.resumeCalls).toHaveLength(1)
+    expect(manager.resumeCalls[0]).toEqual({
+      sessionId: "session-a",
+      prompt: "continue with additional context",
+    })
+  })
+
+  test("should preserve existing tool call count in progress", () => {
+    // #given
+    const taskWithProgress = createMockTask({
+      id: "task-a",
+      sessionID: "session-a",
+      parentSessionID: "session-parent",
+    })
+    taskWithProgress.progress = {
+      toolCalls: 42,
+      lastTool: "read",
+      lastUpdate: new Date(),
+    }
+    manager.addTask(taskWithProgress)
+
+    // #when
+    const result = manager.resume({
+      sessionId: "session-a",
+      prompt: "continue",
+      parentSessionID: "session-new",
+      parentMessageID: "msg-new",
+    })
+
+    // #then
+    expect(result.progress?.toolCalls).toBe(42)
+  })
+})
+
+describe("LaunchInput.skillContent", () => {
+  test("skillContent should be optional in LaunchInput type", () => {
+    // #given
+    const input: import("./types").LaunchInput = {
+      description: "test",
+      prompt: "test prompt",
+      agent: "explore",
+      parentSessionID: "parent-session",
+      parentMessageID: "parent-msg",
+    }
+
+    // #when / #then - should compile without skillContent
+    expect(input.skillContent).toBeUndefined()
+  })
+
+  test("skillContent can be provided in LaunchInput", () => {
+    // #given
+    const input: import("./types").LaunchInput = {
+      description: "test",
+      prompt: "test prompt",
+      agent: "explore",
+      parentSessionID: "parent-session",
+      parentMessageID: "parent-msg",
+      skillContent: "You are a playwright expert",
+    }
+
+    // #when / #then
+    expect(input.skillContent).toBe("You are a playwright expert")
+  })
+})
+
+describe("BackgroundManager.notifyParentSession - agent context preservation", () => {
+  test("should not pass agent field when parentAgent is undefined", async () => {
+    // #given
+    const task: BackgroundTask = {
+      id: "task-no-agent",
+      sessionID: "session-child",
+      parentSessionID: "session-parent",
+      parentMessageID: "msg-parent",
+      description: "task without agent context",
+      prompt: "test",
+      agent: "explore",
+      status: "completed",
+      startedAt: new Date(),
+      completedAt: new Date(),
+      parentAgent: undefined,
+      parentModel: { providerID: "anthropic", modelID: "claude-opus" },
+    }
+
+    // #when
+    const promptBody = buildNotificationPromptBody(task)
+
+    // #then
+    expect("agent" in promptBody).toBe(false)
+    expect(promptBody.model).toEqual({ providerID: "anthropic", modelID: "claude-opus" })
+  })
+
+  test("should include agent field when parentAgent is defined", async () => {
+    // #given
+    const task: BackgroundTask = {
+      id: "task-with-agent",
+      sessionID: "session-child",
+      parentSessionID: "session-parent",
+      parentMessageID: "msg-parent",
+      description: "task with agent context",
+      prompt: "test",
+      agent: "explore",
+      status: "completed",
+      startedAt: new Date(),
+      completedAt: new Date(),
+      parentAgent: "Sisyphus",
+      parentModel: { providerID: "anthropic", modelID: "claude-opus" },
+    }
+
+    // #when
+    const promptBody = buildNotificationPromptBody(task)
+
+    // #then
+    expect(promptBody.agent).toBe("Sisyphus")
+  })
+
+  test("should not pass model field when parentModel is undefined", async () => {
+    // #given
+    const task: BackgroundTask = {
+      id: "task-no-model",
+      sessionID: "session-child",
+      parentSessionID: "session-parent",
+      parentMessageID: "msg-parent",
+      description: "task without model context",
+      prompt: "test",
+      agent: "explore",
+      status: "completed",
+      startedAt: new Date(),
+      completedAt: new Date(),
+      parentAgent: "Sisyphus",
+      parentModel: undefined,
+    }
+
+    // #when
+    const promptBody = buildNotificationPromptBody(task)
+
+    // #then
+    expect("model" in promptBody).toBe(false)
+    expect(promptBody.agent).toBe("Sisyphus")
+  })
+})
+
+function buildNotificationPromptBody(task: BackgroundTask): Record<string, unknown> {
+  const body: Record<string, unknown> = {
+    parts: [{ type: "text", text: `[BACKGROUND TASK COMPLETED] Task "${task.description}" finished.` }],
+  }
+
+  if (task.parentAgent !== undefined) {
+    body.agent = task.parentAgent
+  }
+
+  if (task.parentModel?.providerID && task.parentModel?.modelID) {
+    body.model = { providerID: task.parentModel.providerID, modelID: task.parentModel.modelID }
+  }
+
+  return body
+}

src/features/background-agent/manager.ts

@@ -1,20 +1,19 @@
-import { existsSync, readdirSync } from "node:fs"
-import { join } from "node:path"
+
 import type { PluginInput } from "@opencode-ai/plugin"
 import type {
   BackgroundTask,
   LaunchInput,
+  ResumeInput,
 } from "./types"
 import { log } from "../../shared/logger"
 import { ConcurrencyManager } from "./concurrency"
 import type { BackgroundTaskConfig } from "../../config/schema"
-import {
-  findNearestMessageWithFields,
-  MESSAGE_STORAGE,
-} from "../hook-message-injector"
+
 import { subagentSessions } from "../claude-code-session-state"
+import { getTaskToastManager } from "../task-toast-manager"
 
 const TASK_TTL_MS = 30 * 60 * 1000
+const MIN_STABILITY_TIME_MS = 10 * 1000  // Must run at least 10s before stability detection kicks in
 
 type OpencodeClient = PluginInput["client"]
 
@@ -42,23 +41,10 @@ interface Todo {
   id: string
 }
 
-function getMessageDir(sessionID: string): string | null {
-  if (!existsSync(MESSAGE_STORAGE)) return null
-
-  const directPath = join(MESSAGE_STORAGE, sessionID)
-  if (existsSync(directPath)) return directPath
-
-  for (const dir of readdirSync(MESSAGE_STORAGE)) {
-    const sessionPath = join(MESSAGE_STORAGE, dir, sessionID)
-    if (existsSync(sessionPath)) return sessionPath
-  }
-
-  return null
-}
-
 export class BackgroundManager {
   private tasks: Map<string, BackgroundTask>
   private notifications: Map<string, BackgroundTask[]>
+  private pendingByParent: Map<string, Set<string>>  // Track pending tasks per parent for batching
   private client: OpencodeClient
   private directory: string
   private pollingInterval?: ReturnType<typeof setInterval>
@@ -67,19 +53,27 @@ export class BackgroundManager {
   constructor(ctx: PluginInput, config?: BackgroundTaskConfig) {
     this.tasks = new Map()
     this.notifications = new Map()
+    this.pendingByParent = new Map()
     this.client = ctx.client
     this.directory = ctx.directory
     this.concurrencyManager = new ConcurrencyManager(config)
   }
 
   async launch(input: LaunchInput): Promise<BackgroundTask> {
+    log("[background-agent] launch() called with:", {
+      agent: input.agent,
+      model: input.model,
+      description: input.description,
+      parentSessionID: input.parentSessionID,
+    })
+
     if (!input.agent || input.agent.trim() === "") {
       throw new Error("Agent parameter is required")
     }
 
-    const model = input.agent
+    const concurrencyKey = input.agent
 
-    await this.concurrencyManager.acquire(model)
+    await this.concurrencyManager.acquire(concurrencyKey)
 
     const createResult = await this.client.session.create({
       body: {
@@ -87,12 +81,12 @@ export class BackgroundManager {
         title: `Background: ${input.description}`,
       },
     }).catch((error) => {
-      this.concurrencyManager.release(model)
+      this.concurrencyManager.release(concurrencyKey)
       throw error
     })
 
     if (createResult.error) {
-      this.concurrencyManager.release(model)
+      this.concurrencyManager.release(concurrencyKey)
       throw new Error(`Failed to create background session: ${createResult.error}`)
     }
 
@@ -114,21 +108,50 @@ export class BackgroundManager {
         lastUpdate: new Date(),
       },
       parentModel: input.parentModel,
-      model,
+      parentAgent: input.parentAgent,
+      model: input.model,
+      concurrencyKey,
     }
 
     this.tasks.set(task.id, task)
     this.startPolling()
 
+    // Track for batched notifications
+    const pending = this.pendingByParent.get(input.parentSessionID) ?? new Set()
+    pending.add(task.id)
+    this.pendingByParent.set(input.parentSessionID, pending)
+
     log("[background-agent] Launching task:", { taskId: task.id, sessionID, agent: input.agent })
 
-    this.client.session.promptAsync({
+    const toastManager = getTaskToastManager()
+    if (toastManager) {
+      toastManager.addTask({
+        id: task.id,
+        description: input.description,
+        agent: input.agent,
+        isBackground: true,
+        skills: input.skills,
+      })
+    }
+
+    log("[background-agent] Calling prompt (fire-and-forget) for launch with:", {
+      sessionID,
+      agent: input.agent,
+      model: input.model,
+      hasSkillContent: !!input.skillContent,
+      promptLength: input.prompt.length,
+    })
+
+    // Use prompt() instead of promptAsync() to properly initialize agent loop (fire-and-forget)
+    // Include model if caller provided one (e.g., from Sisyphus category configs)
+    this.client.session.prompt({
       path: { id: sessionID },
       body: {
         agent: input.agent,
+        ...(input.model ? { model: input.model } : {}),
+        system: input.skillContent,
         tools: {
           task: false,
-          background_task: false,
           call_omo_agent: false,
         },
         parts: [{ type: "text", text: input.prompt }],
@@ -145,11 +168,13 @@ export class BackgroundManager {
           existingTask.error = errorMessage
         }
         existingTask.completedAt = new Date()
-        if (existingTask.model) {
-          this.concurrencyManager.release(existingTask.model)
+        if (existingTask.concurrencyKey) {
+          this.concurrencyManager.release(existingTask.concurrencyKey)
         }
         this.markForNotification(existingTask)
-        this.notifyParentSession(existingTask)
+        this.notifyParentSession(existingTask).catch(err => {
+          log("[background-agent] Failed to notify on error:", err)
+        })
       }
     })
 
@@ -192,6 +217,121 @@ export class BackgroundManager {
     return undefined
   }
 
+  /**
+   * Register an external task (e.g., from sisyphus_task) for notification tracking.
+   * This allows tasks created by external tools to receive the same toast/prompt notifications.
+   */
+  registerExternalTask(input: {
+    taskId: string
+    sessionID: string
+    parentSessionID: string
+    description: string
+    agent?: string
+    parentAgent?: string
+  }): BackgroundTask {
+    const task: BackgroundTask = {
+      id: input.taskId,
+      sessionID: input.sessionID,
+      parentSessionID: input.parentSessionID,
+      parentMessageID: "",
+      description: input.description,
+      prompt: "",
+      agent: input.agent || "sisyphus_task",
+      status: "running",
+      startedAt: new Date(),
+      progress: {
+        toolCalls: 0,
+        lastUpdate: new Date(),
+      },
+      parentAgent: input.parentAgent,
+    }
+
+    this.tasks.set(task.id, task)
+    subagentSessions.add(input.sessionID)
+    this.startPolling()
+
+    // Track for batched notifications (external tasks need tracking too)
+    const pending = this.pendingByParent.get(input.parentSessionID) ?? new Set()
+    pending.add(task.id)
+    this.pendingByParent.set(input.parentSessionID, pending)
+
+    log("[background-agent] Registered external task:", { taskId: task.id, sessionID: input.sessionID })
+
+    return task
+  }
+
+  async resume(input: ResumeInput): Promise<BackgroundTask> {
+    const existingTask = this.findBySession(input.sessionId)
+    if (!existingTask) {
+      throw new Error(`Task not found for session: ${input.sessionId}`)
+    }
+
+    existingTask.status = "running"
+    existingTask.completedAt = undefined
+    existingTask.error = undefined
+    existingTask.parentSessionID = input.parentSessionID
+    existingTask.parentMessageID = input.parentMessageID
+    existingTask.parentModel = input.parentModel
+    existingTask.parentAgent = input.parentAgent
+
+    existingTask.progress = {
+      toolCalls: existingTask.progress?.toolCalls ?? 0,
+      lastUpdate: new Date(),
+    }
+
+    this.startPolling()
+    subagentSessions.add(existingTask.sessionID)
+
+    // Track for batched notifications (P2 fix: resumed tasks need tracking too)
+    const pending = this.pendingByParent.get(input.parentSessionID) ?? new Set()
+    pending.add(existingTask.id)
+    this.pendingByParent.set(input.parentSessionID, pending)
+
+    const toastManager = getTaskToastManager()
+    if (toastManager) {
+      toastManager.addTask({
+        id: existingTask.id,
+        description: existingTask.description,
+        agent: existingTask.agent,
+        isBackground: true,
+      })
+    }
+
+    log("[background-agent] Resuming task:", { taskId: existingTask.id, sessionID: existingTask.sessionID })
+
+    log("[background-agent] Resuming task - calling prompt (fire-and-forget) with:", {
+      sessionID: existingTask.sessionID,
+      agent: existingTask.agent,
+      promptLength: input.prompt.length,
+    })
+
+    // Note: Don't pass model in body - use agent's configured model instead
+    // Use prompt() instead of promptAsync() to properly initialize agent loop
+    this.client.session.prompt({
+      path: { id: existingTask.sessionID },
+      body: {
+        agent: existingTask.agent,
+        tools: {
+          task: false,
+          call_omo_agent: false,
+        },
+        parts: [{ type: "text", text: input.prompt }],
+      },
+    }).catch((error) => {
+      log("[background-agent] resume prompt error:", error)
+      existingTask.status = "error"
+      const errorMessage = error instanceof Error ? error.message : String(error)
+      existingTask.error = errorMessage
+      existingTask.completedAt = new Date()
+      this.markForNotification(existingTask)
+      this.notifyParentSession(existingTask).catch(err => {
+        log("[background-agent] Failed to notify on resume error:", err)
+      })
+    })
+
+    return existingTask
+  }
+
   private async checkSessionTodos(sessionID: string): Promise<boolean> {
     try {
       const response = await this.client.session.todo({
@@ -241,7 +381,22 @@ export class BackgroundManager {
       const task = this.findBySession(sessionID)
       if (!task || task.status !== "running") return
 
-      this.checkSessionTodos(sessionID).then((hasIncompleteTodos) => {
+      // Edge guard: Require minimum elapsed time (5 seconds) before accepting idle
+      const elapsedMs = Date.now() - task.startedAt.getTime()
+      const MIN_IDLE_TIME_MS = 5000
+      if (elapsedMs < MIN_IDLE_TIME_MS) {
+        log("[background-agent] Ignoring early session.idle, elapsed:", { elapsedMs, taskId: task.id })
+        return
+      }
+
+      // Edge guard: Verify session has actual assistant output before completing
+      this.validateSessionHasOutput(sessionID).then(async (hasValidOutput) => {
+        if (!hasValidOutput) {
+          log("[background-agent] Session.idle but no valid output yet, waiting:", task.id)
+          return
+        }
+
+        const hasIncompleteTodos = await this.checkSessionTodos(sessionID)
         if (hasIncompleteTodos) {
           log("[background-agent] Task has incomplete todos, waiting for todo-continuation:", task.id)
           return
@@ -250,8 +405,10 @@ export class BackgroundManager {
         task.status = "completed"
         task.completedAt = new Date()
         this.markForNotification(task)
-        this.notifyParentSession(task)
+        await this.notifyParentSession(task)
         log("[background-agent] Task completed via session.idle event:", task.id)
+      }).catch(err => {
+        log("[background-agent] Error in session.idle handler:", err)
       })
     }
 
@@ -269,8 +426,8 @@ export class BackgroundManager {
         task.error = "Session deleted"
       }
 
-      if (task.model) {
-        this.concurrencyManager.release(task.model)
+      if (task.concurrencyKey) {
+        this.concurrencyManager.release(task.concurrencyKey)
       }
       this.tasks.delete(task.id)
       this.clearNotificationsForTask(task.id)
@@ -292,6 +449,66 @@ export class BackgroundManager {
     this.notifications.delete(sessionID)
   }
 
+  /**
+   * Validates that a session has actual assistant/tool output before marking complete.
+   * Prevents premature completion when session.idle fires before agent responds.
+   */
+  private async validateSessionHasOutput(sessionID: string): Promise<boolean> {
+    try {
+      const response = await this.client.session.messages({
+        path: { id: sessionID },
+      })
+
+      const messages = response.data ?? []
+      
+      // Check for at least one assistant or tool message
+      const hasAssistantOrToolMessage = messages.some(
+        (m: { info?: { role?: string } }) => 
+          m.info?.role === "assistant" || m.info?.role === "tool"
+      )
+
+      if (!hasAssistantOrToolMessage) {
+        log("[background-agent] No assistant/tool messages found in session:", sessionID)
+        return false
+      }
+
+      // Additionally check that at least one message has content (not just empty)
+      // OpenCode API uses different part types than Anthropic's API:
+      // - "reasoning" with .text property (thinking/reasoning content)
+      // - "tool" with .state.output property (tool call results)
+      // - "text" with .text property (final text output)
+      // - "step-start"/"step-finish" (metadata, no content)
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      const hasContent = messages.some((m: any) => {
+        if (m.info?.role !== "assistant" && m.info?.role !== "tool") return false
+        const parts = m.parts ?? []
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      return parts.some((p: any) => 
+        // Text content (final output)
+        (p.type === "text" && p.text && p.text.trim().length > 0) ||
+        // Reasoning content (thinking blocks)
+        (p.type === "reasoning" && p.text && p.text.trim().length > 0) ||
+        // Tool calls (indicates work was done)
+        p.type === "tool" ||
+        // Tool results (output from executed tools) - important for tool-only tasks
+        (p.type === "tool_result" && p.content && 
+          (typeof p.content === "string" ? p.content.trim().length > 0 : p.content.length > 0))
+      )
+      })
+
+      if (!hasContent) {
+        log("[background-agent] Messages exist but no content found in session:", sessionID)
+        return false
+      }
+
+      return true
+    } catch (error) {
+      log("[background-agent] Error validating session output:", error)
+      // On error, allow completion to proceed (don't block indefinitely)
+      return true
+    }
+  }
+
   private clearNotificationsForTask(taskId: string): void {
     for (const [sessionID, tasks] of this.notifications.entries()) {
       const filtered = tasks.filter((t) => t.id !== taskId)
@@ -319,67 +536,119 @@ export class BackgroundManager {
     }
   }
 
-  cleanup(): void {
+cleanup(): void {
     this.stopPolling()
     this.tasks.clear()
     this.notifications.clear()
+    this.pendingByParent.clear()
   }
 
-  private notifyParentSession(task: BackgroundTask): void {
+  /**
+   * Get all running tasks (for compaction hook)
+   */
+  getRunningTasks(): BackgroundTask[] {
+    return Array.from(this.tasks.values()).filter(t => t.status === "running")
+  }
+
+  /**
+   * Get all completed tasks still in memory (for compaction hook)
+   */
+  getCompletedTasks(): BackgroundTask[] {
+    return Array.from(this.tasks.values()).filter(t => t.status !== "running")
+  }
+
+private async notifyParentSession(task: BackgroundTask): Promise<void> {
     const duration = this.formatDuration(task.startedAt, task.completedAt)
 
     log("[background-agent] notifyParentSession called for task:", task.id)
 
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    const tuiClient = this.client as any
-    if (tuiClient.tui?.showToast) {
-      tuiClient.tui.showToast({
-        body: {
-          title: "Background Task Completed",
-          message: `Task "${task.description}" finished in ${duration}.`,
-          variant: "success",
-          duration: 5000,
-        },
-      }).catch(() => {})
+    // Show toast notification
+    const toastManager = getTaskToastManager()
+    if (toastManager) {
+      toastManager.showCompletionToast({
+        id: task.id,
+        description: task.description,
+        duration,
+      })
     }
 
-    const message = `[BACKGROUND TASK COMPLETED] Task "${task.description}" finished in ${duration}. Use background_output with task_id="${task.id}" to get results.`
+    // Update pending tracking and check if all tasks complete
+    const pendingSet = this.pendingByParent.get(task.parentSessionID)
+    if (pendingSet) {
+      pendingSet.delete(task.id)
+      if (pendingSet.size === 0) {
+        this.pendingByParent.delete(task.parentSessionID)
+      }
+    }
 
-    log("[background-agent] Sending notification to parent session:", { parentSessionID: task.parentSessionID })
+    const allComplete = !pendingSet || pendingSet.size === 0
+    const remainingCount = pendingSet?.size ?? 0
 
+    // Build notification message
+    const statusText = task.status === "error" ? "FAILED" : "COMPLETED"
+    const errorInfo = task.error ? `\n**Error:** ${task.error}` : ""
+    
+    let notification: string
+    if (allComplete) {
+      // All tasks complete - build summary
+      const completedTasks = Array.from(this.tasks.values())
+        .filter(t => t.parentSessionID === task.parentSessionID && t.status !== "running")
+        .map(t => `- \`${t.id}\`: ${t.description}`)
+        .join("\n")
+
+      notification = `<system-reminder>
+[ALL BACKGROUND TASKS COMPLETE]
+
+**Completed:**
+${completedTasks || `- \`${task.id}\`: ${task.description}`}
+
+Use \`background_output(task_id="<id>")\` to retrieve each result.
+</system-reminder>`
+    } else {
+      // Individual completion - silent notification
+      notification = `<system-reminder>
+[BACKGROUND TASK ${statusText}]
+**ID:** \`${task.id}\`
+**Description:** ${task.description}
+**Duration:** ${duration}${errorInfo}
+
+**${remainingCount} task${remainingCount === 1 ? "" : "s"} still in progress.** You WILL be notified when ALL complete.
+Do NOT poll - continue productive work.
+
+Use \`background_output(task_id="${task.id}")\` to retrieve this result when ready.
+</system-reminder>`
+    }
+
+    // Inject notification via session.prompt with noReply
+    try {
+      await this.client.session.prompt({
+        path: { id: task.parentSessionID },
+        body: {
+          noReply: !allComplete,  // Silent unless all complete
+          agent: task.parentAgent,
+          parts: [{ type: "text", text: notification }],
+        },
+      })
+      log("[background-agent] Sent notification to parent session:", {
+        taskId: task.id,
+        allComplete,
+        noReply: !allComplete,
+      })
+    } catch (error) {
+      log("[background-agent] Failed to send notification:", error)
+    }
+
+    // Cleanup after retention period
     const taskId = task.id
-    setTimeout(async () => {
-      if (task.model) {
-        this.concurrencyManager.release(task.model)
+    setTimeout(() => {
+      if (task.concurrencyKey) {
+        this.concurrencyManager.release(task.concurrencyKey)
+        task.concurrencyKey = undefined
       }
-
-      try {
-        const messageDir = getMessageDir(task.parentSessionID)
-        const prevMessage = messageDir ? findNearestMessageWithFields(messageDir) : null
-
-        const modelContext = task.parentModel ?? prevMessage?.model
-        const modelField = modelContext?.providerID && modelContext?.modelID
-          ? { providerID: modelContext.providerID, modelID: modelContext.modelID }
-          : undefined
-
-        await this.client.session.prompt({
-          path: { id: task.parentSessionID },
-          body: {
-            agent: prevMessage?.agent,
-            model: modelField,
-            parts: [{ type: "text", text: message }],
-          },
-          query: { directory: this.directory },
-        })
-        log("[background-agent] Successfully sent prompt to parent session:", { parentSessionID: task.parentSessionID })
-      } catch (error) {
-        log("[background-agent] prompt failed:", String(error))
-      } finally {
-        this.clearNotificationsForTask(taskId)
-        this.tasks.delete(taskId)
-        log("[background-agent] Removed completed task from memory:", taskId)
-      }
-    }, 200)
+      this.clearNotificationsForTask(taskId)
+      this.tasks.delete(taskId)
+      log("[background-agent] Removed completed task from memory:", taskId)
+    }, 5 * 60 * 1000)
   }
 
   private formatDuration(start: Date, end?: Date): string {
@@ -413,8 +682,8 @@ export class BackgroundManager {
         task.status = "error"
         task.error = "Task timed out after 30 minutes"
         task.completedAt = new Date()
-        if (task.model) {
-          this.concurrencyManager.release(task.model)
+        if (task.concurrencyKey) {
+          this.concurrencyManager.release(task.concurrencyKey)
         }
         this.clearNotificationsForTask(taskId)
         this.tasks.delete(taskId)
@@ -448,15 +717,18 @@ export class BackgroundManager {
     for (const task of this.tasks.values()) {
       if (task.status !== "running") continue
 
-      try {
+try {
         const sessionStatus = allStatuses[task.sessionID]
         
-        if (!sessionStatus) {
-          log("[background-agent] Session not found in status:", task.sessionID)
-          continue
-        }
+        // Don't skip if session not in status - fall through to message-based detection
+        if (sessionStatus?.type === "idle") {
+          // Edge guard: Validate session has actual output before completing
+          const hasValidOutput = await this.validateSessionHasOutput(task.sessionID)
+          if (!hasValidOutput) {
+            log("[background-agent] Polling idle but no valid output yet, waiting:", task.id)
+            continue
+          }
 
-        if (sessionStatus.type === "idle") {
           const hasIncompleteTodos = await this.checkSessionTodos(task.sessionID)
           if (hasIncompleteTodos) {
             log("[background-agent] Task has incomplete todos via polling, waiting:", task.id)
@@ -466,7 +738,7 @@ export class BackgroundManager {
           task.status = "completed"
           task.completedAt = new Date()
           this.markForNotification(task)
-          this.notifyParentSession(task)
+          await this.notifyParentSession(task)
           log("[background-agent] Task completed via polling:", task.id)
           continue
         }
@@ -507,10 +779,41 @@ export class BackgroundManager {
           task.progress.toolCalls = toolCalls
           task.progress.lastTool = lastTool
           task.progress.lastUpdate = new Date()
-          if (lastMessage) {
+if (lastMessage) {
             task.progress.lastMessage = lastMessage
             task.progress.lastMessageAt = new Date()
           }
+
+          // Stability detection: complete when message count unchanged for 3 polls
+          const currentMsgCount = messages.length
+          const elapsedMs = Date.now() - task.startedAt.getTime()
+
+          if (elapsedMs >= MIN_STABILITY_TIME_MS) {
+            if (task.lastMsgCount === currentMsgCount) {
+              task.stablePolls = (task.stablePolls ?? 0) + 1
+              if (task.stablePolls >= 3) {
+                // Edge guard: Validate session has actual output before completing
+                const hasValidOutput = await this.validateSessionHasOutput(task.sessionID)
+                if (!hasValidOutput) {
+                  log("[background-agent] Stability reached but no valid output, waiting:", task.id)
+                  continue
+                }
+
+                const hasIncompleteTodos = await this.checkSessionTodos(task.sessionID)
+                if (!hasIncompleteTodos) {
+                  task.status = "completed"
+                  task.completedAt = new Date()
+                  this.markForNotification(task)
+                  await this.notifyParentSession(task)
+                  log("[background-agent] Task completed via stability detection:", task.id)
+                  continue
+                }
+              }
+            } else {
+              task.stablePolls = 0
+            }
+          }
+          task.lastMsgCount = currentMsgCount
         }
       } catch (error) {
         log("[background-agent] Poll error for task:", { taskId: task.id, error })

src/features/background-agent/types.ts

@@ -27,7 +27,15 @@ export interface BackgroundTask {
   error?: string
   progress?: TaskProgress
   parentModel?: { providerID: string; modelID: string }
-  model?: string
+  model?: { providerID: string; modelID: string }
+  /** Agent name used for concurrency tracking */
+  concurrencyKey?: string
+  /** Parent session's agent name for notification */
+  parentAgent?: string
+  /** Last message count for stability detection */
+  lastMsgCount?: number
+  /** Number of consecutive polls with stable message count */
+  stablePolls?: number
 }
 
 export interface LaunchInput {
@@ -37,4 +45,17 @@ export interface LaunchInput {
   parentSessionID: string
   parentMessageID: string
   parentModel?: { providerID: string; modelID: string }
+  parentAgent?: string
+  model?: { providerID: string; modelID: string }
+  skills?: string[]
+  skillContent?: string
+}
+
+export interface ResumeInput {
+  sessionId: string
+  prompt: string
+  parentSessionID: string
+  parentMessageID: string
+  parentModel?: { providerID: string; modelID: string }
+  parentAgent?: string
 }

src/features/boulder-state/constants.ts

@@ -0,0 +1,13 @@
+/**
+ * Boulder State Constants
+ */
+
+export const BOULDER_DIR = ".sisyphus"
+export const BOULDER_FILE = "boulder.json"
+export const BOULDER_STATE_PATH = `${BOULDER_DIR}/${BOULDER_FILE}`
+
+export const NOTEPAD_DIR = "notepads"
+export const NOTEPAD_BASE_PATH = `${BOULDER_DIR}/${NOTEPAD_DIR}`
+
+/** Prometheus plan directory pattern */
+export const PROMETHEUS_PLANS_DIR = ".sisyphus/plans"

src/features/boulder-state/index.ts

@@ -0,0 +1,3 @@
+export * from "./types"
+export * from "./constants"
+export * from "./storage"

src/features/boulder-state/storage.test.ts

@@ -0,0 +1,250 @@
+import { describe, expect, test, beforeEach, afterEach } from "bun:test"
+import { existsSync, mkdirSync, rmSync, writeFileSync } from "node:fs"
+import { join } from "node:path"
+import { tmpdir } from "node:os"
+import {
+  readBoulderState,
+  writeBoulderState,
+  appendSessionId,
+  clearBoulderState,
+  getPlanProgress,
+  getPlanName,
+  createBoulderState,
+  findPrometheusPlans,
+} from "./storage"
+import type { BoulderState } from "./types"
+
+describe("boulder-state", () => {
+  const TEST_DIR = join(tmpdir(), "boulder-state-test-" + Date.now())
+  const SISYPHUS_DIR = join(TEST_DIR, ".sisyphus")
+
+  beforeEach(() => {
+    if (!existsSync(TEST_DIR)) {
+      mkdirSync(TEST_DIR, { recursive: true })
+    }
+    if (!existsSync(SISYPHUS_DIR)) {
+      mkdirSync(SISYPHUS_DIR, { recursive: true })
+    }
+    clearBoulderState(TEST_DIR)
+  })
+
+  afterEach(() => {
+    if (existsSync(TEST_DIR)) {
+      rmSync(TEST_DIR, { recursive: true, force: true })
+    }
+  })
+
+  describe("readBoulderState", () => {
+    test("should return null when no boulder.json exists", () => {
+      // #given - no boulder.json file
+      // #when
+      const result = readBoulderState(TEST_DIR)
+      // #then
+      expect(result).toBeNull()
+    })
+
+    test("should read valid boulder state", () => {
+      // #given - valid boulder.json
+      const state: BoulderState = {
+        active_plan: "/path/to/plan.md",
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1", "session-2"],
+        plan_name: "my-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      // #when
+      const result = readBoulderState(TEST_DIR)
+
+      // #then
+      expect(result).not.toBeNull()
+      expect(result?.active_plan).toBe("/path/to/plan.md")
+      expect(result?.session_ids).toEqual(["session-1", "session-2"])
+      expect(result?.plan_name).toBe("my-plan")
+    })
+  })
+
+  describe("writeBoulderState", () => {
+    test("should write state and create .sisyphus directory if needed", () => {
+      // #given - state to write
+      const state: BoulderState = {
+        active_plan: "/test/plan.md",
+        started_at: "2026-01-02T12:00:00Z",
+        session_ids: ["ses-123"],
+        plan_name: "test-plan",
+      }
+
+      // #when
+      const success = writeBoulderState(TEST_DIR, state)
+      const readBack = readBoulderState(TEST_DIR)
+
+      // #then
+      expect(success).toBe(true)
+      expect(readBack).not.toBeNull()
+      expect(readBack?.active_plan).toBe("/test/plan.md")
+    })
+  })
+
+  describe("appendSessionId", () => {
+    test("should append new session id to existing state", () => {
+      // #given - existing state with one session
+      const state: BoulderState = {
+        active_plan: "/plan.md",
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1"],
+        plan_name: "plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      // #when
+      const result = appendSessionId(TEST_DIR, "session-2")
+
+      // #then
+      expect(result).not.toBeNull()
+      expect(result?.session_ids).toEqual(["session-1", "session-2"])
+    })
+
+    test("should not duplicate existing session id", () => {
+      // #given - state with session-1 already
+      const state: BoulderState = {
+        active_plan: "/plan.md",
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1"],
+        plan_name: "plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      // #when
+      appendSessionId(TEST_DIR, "session-1")
+      const result = readBoulderState(TEST_DIR)
+
+      // #then
+      expect(result?.session_ids).toEqual(["session-1"])
+    })
+
+    test("should return null when no state exists", () => {
+      // #given - no boulder.json
+      // #when
+      const result = appendSessionId(TEST_DIR, "new-session")
+      // #then
+      expect(result).toBeNull()
+    })
+  })
+
+  describe("clearBoulderState", () => {
+    test("should remove boulder.json", () => {
+      // #given - existing state
+      const state: BoulderState = {
+        active_plan: "/plan.md",
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1"],
+        plan_name: "plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      // #when
+      const success = clearBoulderState(TEST_DIR)
+      const result = readBoulderState(TEST_DIR)
+
+      // #then
+      expect(success).toBe(true)
+      expect(result).toBeNull()
+    })
+
+    test("should succeed even when no file exists", () => {
+      // #given - no boulder.json
+      // #when
+      const success = clearBoulderState(TEST_DIR)
+      // #then
+      expect(success).toBe(true)
+    })
+  })
+
+  describe("getPlanProgress", () => {
+    test("should count completed and uncompleted checkboxes", () => {
+      // #given - plan file with checkboxes
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, `# Plan
+- [ ] Task 1
+- [x] Task 2  
+- [ ] Task 3
+- [X] Task 4
+`)
+
+      // #when
+      const progress = getPlanProgress(planPath)
+
+      // #then
+      expect(progress.total).toBe(4)
+      expect(progress.completed).toBe(2)
+      expect(progress.isComplete).toBe(false)
+    })
+
+    test("should return isComplete true when all checked", () => {
+      // #given - all tasks completed
+      const planPath = join(TEST_DIR, "complete-plan.md")
+      writeFileSync(planPath, `# Plan
+- [x] Task 1
+- [X] Task 2
+`)
+
+      // #when
+      const progress = getPlanProgress(planPath)
+
+      // #then
+      expect(progress.total).toBe(2)
+      expect(progress.completed).toBe(2)
+      expect(progress.isComplete).toBe(true)
+    })
+
+    test("should return isComplete true for empty plan", () => {
+      // #given - plan with no checkboxes
+      const planPath = join(TEST_DIR, "empty-plan.md")
+      writeFileSync(planPath, "# Plan\nNo tasks here")
+
+      // #when
+      const progress = getPlanProgress(planPath)
+
+      // #then
+      expect(progress.total).toBe(0)
+      expect(progress.isComplete).toBe(true)
+    })
+
+    test("should handle non-existent file", () => {
+      // #given - non-existent file
+      // #when
+      const progress = getPlanProgress("/non/existent/file.md")
+      // #then
+      expect(progress.total).toBe(0)
+      expect(progress.isComplete).toBe(true)
+    })
+  })
+
+  describe("getPlanName", () => {
+    test("should extract plan name from path", () => {
+      // #given
+      const path = "/home/user/.sisyphus/plans/project/my-feature.md"
+      // #when
+      const name = getPlanName(path)
+      // #then
+      expect(name).toBe("my-feature")
+    })
+  })
+
+  describe("createBoulderState", () => {
+    test("should create state with correct fields", () => {
+      // #given
+      const planPath = "/path/to/auth-refactor.md"
+      const sessionId = "ses-abc123"
+
+      // #when
+      const state = createBoulderState(planPath, sessionId)
+
+      // #then
+      expect(state.active_plan).toBe(planPath)
+      expect(state.session_ids).toEqual([sessionId])
+      expect(state.plan_name).toBe("auth-refactor")
+      expect(state.started_at).toBeDefined()
+    })
+  })
+})

src/features/boulder-state/storage.ts

@@ -0,0 +1,150 @@
+/**
+ * Boulder State Storage
+ *
+ * Handles reading/writing boulder.json for active plan tracking.
+ */
+
+import { existsSync, readFileSync, writeFileSync, mkdirSync, readdirSync } from "node:fs"
+import { dirname, join, basename } from "node:path"
+import type { BoulderState, PlanProgress } from "./types"
+import { BOULDER_DIR, BOULDER_FILE, PROMETHEUS_PLANS_DIR } from "./constants"
+
+export function getBoulderFilePath(directory: string): string {
+  return join(directory, BOULDER_DIR, BOULDER_FILE)
+}
+
+export function readBoulderState(directory: string): BoulderState | null {
+  const filePath = getBoulderFilePath(directory)
+
+  if (!existsSync(filePath)) {
+    return null
+  }
+
+  try {
+    const content = readFileSync(filePath, "utf-8")
+    return JSON.parse(content) as BoulderState
+  } catch {
+    return null
+  }
+}
+
+export function writeBoulderState(directory: string, state: BoulderState): boolean {
+  const filePath = getBoulderFilePath(directory)
+
+  try {
+    const dir = dirname(filePath)
+    if (!existsSync(dir)) {
+      mkdirSync(dir, { recursive: true })
+    }
+
+    writeFileSync(filePath, JSON.stringify(state, null, 2), "utf-8")
+    return true
+  } catch {
+    return false
+  }
+}
+
+export function appendSessionId(directory: string, sessionId: string): BoulderState | null {
+  const state = readBoulderState(directory)
+  if (!state) return null
+
+  if (!state.session_ids.includes(sessionId)) {
+    state.session_ids.push(sessionId)
+    if (writeBoulderState(directory, state)) {
+      return state
+    }
+  }
+
+  return state
+}
+
+export function clearBoulderState(directory: string): boolean {
+  const filePath = getBoulderFilePath(directory)
+
+  try {
+    if (existsSync(filePath)) {
+      const { unlinkSync } = require("node:fs")
+      unlinkSync(filePath)
+    }
+    return true
+  } catch {
+    return false
+  }
+}
+
+/**
+ * Find Prometheus plan files for this project.
+ * Prometheus stores plans at: {project}/.sisyphus/plans/{name}.md
+ */
+export function findPrometheusPlans(directory: string): string[] {
+  const plansDir = join(directory, PROMETHEUS_PLANS_DIR)
+
+  if (!existsSync(plansDir)) {
+    return []
+  }
+
+  try {
+    const files = readdirSync(plansDir)
+    return files
+      .filter((f) => f.endsWith(".md"))
+      .map((f) => join(plansDir, f))
+      .sort((a, b) => {
+        // Sort by modification time, newest first
+        const aStat = require("node:fs").statSync(a)
+        const bStat = require("node:fs").statSync(b)
+        return bStat.mtimeMs - aStat.mtimeMs
+      })
+  } catch {
+    return []
+  }
+}
+
+/**
+ * Parse a plan file and count checkbox progress.
+ */
+export function getPlanProgress(planPath: string): PlanProgress {
+  if (!existsSync(planPath)) {
+    return { total: 0, completed: 0, isComplete: true }
+  }
+
+  try {
+    const content = readFileSync(planPath, "utf-8")
+    
+    // Match markdown checkboxes: - [ ] or - [x] or - [X]
+    const uncheckedMatches = content.match(/^[-*]\s*\[\s*\]/gm) || []
+    const checkedMatches = content.match(/^[-*]\s*\[[xX]\]/gm) || []
+
+    const total = uncheckedMatches.length + checkedMatches.length
+    const completed = checkedMatches.length
+
+    return {
+      total,
+      completed,
+      isComplete: total === 0 || completed === total,
+    }
+  } catch {
+    return { total: 0, completed: 0, isComplete: true }
+  }
+}
+
+/**
+ * Extract plan name from file path.
+ */
+export function getPlanName(planPath: string): string {
+  return basename(planPath, ".md")
+}
+
+/**
+ * Create a new boulder state for a plan.
+ */
+export function createBoulderState(
+  planPath: string,
+  sessionId: string
+): BoulderState {
+  return {
+    active_plan: planPath,
+    started_at: new Date().toISOString(),
+    session_ids: [sessionId],
+    plan_name: getPlanName(planPath),
+  }
+}

src/features/boulder-state/types.ts

@@ -0,0 +1,26 @@
+/**
+ * Boulder State Types
+ *
+ * Manages the active work plan state for Sisyphus orchestrator.
+ * Named after Sisyphus's boulder - the eternal task that must be rolled.
+ */
+
+export interface BoulderState {
+  /** Absolute path to the active plan file */
+  active_plan: string
+  /** ISO timestamp when work started */
+  started_at: string
+  /** Session IDs that have worked on this plan */
+  session_ids: string[]
+  /** Plan name derived from filename */
+  plan_name: string
+}
+
+export interface PlanProgress {
+  /** Total number of checkboxes */
+  total: number
+  /** Number of completed checkboxes */
+  completed: number
+  /** Whether all tasks are done */
+  isComplete: boolean
+}

src/features/builtin-commands/commands.ts

@@ -3,6 +3,7 @@ import type { BuiltinCommandName, BuiltinCommands } from "./types"
 import { INIT_DEEP_TEMPLATE } from "./templates/init-deep"
 import { RALPH_LOOP_TEMPLATE, CANCEL_RALPH_TEMPLATE } from "./templates/ralph-loop"
 import { REFACTOR_TEMPLATE } from "./templates/refactor"
+import { START_WORK_TEMPLATE } from "./templates/start-work"
 
 const BUILTIN_COMMAND_DEFINITIONS: Record<BuiltinCommandName, Omit<CommandDefinition, "name">> = {
   "init-deep": {
@@ -41,6 +42,23 @@ ${REFACTOR_TEMPLATE}
 </command-instruction>`,
     argumentHint: "<refactoring-target> [--scope=<file|module|project>] [--strategy=<safe|aggressive>]",
   },
+  "start-work": {
+    description: "(builtin) Start Sisyphus work session from Prometheus plan",
+    agent: "orchestrator-sisyphus",
+    template: `<command-instruction>
+${START_WORK_TEMPLATE}
+</command-instruction>
+
+<session-context>
+Session ID: $SESSION_ID
+Timestamp: $TIMESTAMP
+</session-context>
+
+<user-request>
+$ARGUMENTS
+</user-request>`,
+    argumentHint: "[plan-name]",
+  },
 }
 
 export function loadBuiltinCommands(

src/features/builtin-commands/templates/init-deep.ts

@@ -45,12 +45,12 @@ Don't wait—these run async while main session works.
 
 \`\`\`
 // Fire all at once, collect results later
-background_task(agent="explore", prompt="Project structure: PREDICT standard patterns for detected language → REPORT deviations only")
-background_task(agent="explore", prompt="Entry points: FIND main files → REPORT non-standard organization")
-background_task(agent="explore", prompt="Conventions: FIND config files (.eslintrc, pyproject.toml, .editorconfig) → REPORT project-specific rules")
-background_task(agent="explore", prompt="Anti-patterns: FIND 'DO NOT', 'NEVER', 'ALWAYS', 'DEPRECATED' comments → LIST forbidden patterns")
-background_task(agent="explore", prompt="Build/CI: FIND .github/workflows, Makefile → REPORT non-standard patterns")
-background_task(agent="explore", prompt="Test patterns: FIND test configs, test structure → REPORT unique conventions")
+sisyphus_task(agent="explore", prompt="Project structure: PREDICT standard patterns for detected language → REPORT deviations only")
+sisyphus_task(agent="explore", prompt="Entry points: FIND main files → REPORT non-standard organization")
+sisyphus_task(agent="explore", prompt="Conventions: FIND config files (.eslintrc, pyproject.toml, .editorconfig) → REPORT project-specific rules")
+sisyphus_task(agent="explore", prompt="Anti-patterns: FIND 'DO NOT', 'NEVER', 'ALWAYS', 'DEPRECATED' comments → LIST forbidden patterns")
+sisyphus_task(agent="explore", prompt="Build/CI: FIND .github/workflows, Makefile → REPORT non-standard patterns")
+sisyphus_task(agent="explore", prompt="Test patterns: FIND test configs, test structure → REPORT unique conventions")
 \`\`\`
 
 <dynamic-agents>
@@ -76,9 +76,9 @@ max_depth=$(find . -type d -not -path '*/node_modules/*' -not -path '*/.git/*' |
 Example spawning:
 \`\`\`
 // 500 files, 50k lines, depth 6, 15 large files → spawn 5+5+2+1 = 13 additional agents
-background_task(agent="explore", prompt="Large file analysis: FIND files >500 lines, REPORT complexity hotspots")
-background_task(agent="explore", prompt="Deep modules at depth 4+: FIND hidden patterns, internal conventions")
-background_task(agent="explore", prompt="Cross-cutting concerns: FIND shared utilities across directories")
+sisyphus_task(agent="explore", prompt="Large file analysis: FIND files >500 lines, REPORT complexity hotspots")
+sisyphus_task(agent="explore", prompt="Deep modules at depth 4+: FIND hidden patterns, internal conventions")
+sisyphus_task(agent="explore", prompt="Cross-cutting concerns: FIND shared utilities across directories")
 // ... more based on calculation
 \`\`\`
 </dynamic-agents>
@@ -240,7 +240,7 @@ Launch document-writer agents for each location:
 
 \`\`\`
 for loc in AGENTS_LOCATIONS (except root):
-  background_task(agent="document-writer", prompt=\\\`
+  sisyphus_task(agent="document-writer", prompt=\\\`
     Generate AGENTS.md for: \${loc.path}
     - Reason: \${loc.reason}
     - 30-80 lines max

src/features/builtin-commands/templates/refactor.ts

@@ -605,7 +605,7 @@ Use \`ast_grep_search\` and \`ast_grep_replace\` for structural transformations.
 ## Agents
 - \`explore\`: Parallel codebase pattern discovery
 - \`plan\`: Detailed refactoring plan generation
-- \`oracle\`: Consult for complex architectural decisions
+- \`oracle\`: Read-only consultation for complex architectural decisions and debugging
 - \`librarian\`: **Use proactively** when encountering deprecated methods or library migration tasks. Query official docs and OSS examples for modern replacements.
 
 ## Deprecated Code & Library Migration

src/features/builtin-commands/templates/start-work.ts

@@ -0,0 +1,72 @@
+export const START_WORK_TEMPLATE = `You are starting a Sisyphus work session.
+
+## WHAT TO DO
+
+1. **Find available plans**: Search for Prometheus-generated plan files at \`.sisyphus/plans/\`
+
+2. **Check for active boulder state**: Read \`.sisyphus/boulder.json\` if it exists
+
+3. **Decision logic**:
+   - If \`.sisyphus/boulder.json\` exists AND plan is NOT complete (has unchecked boxes):
+     - **APPEND** current session to session_ids
+     - Continue work on existing plan
+   - If no active plan OR plan is complete:
+     - List available plan files
+     - If ONE plan: auto-select it
+     - If MULTIPLE plans: show list with timestamps, ask user to select
+
+4. **Create/Update boulder.json**:
+   \`\`\`json
+   {
+     "active_plan": "/absolute/path/to/plan.md",
+     "started_at": "ISO_TIMESTAMP",
+     "session_ids": ["session_id_1", "session_id_2"],
+     "plan_name": "plan-name"
+   }
+   \`\`\`
+
+5. **Read the plan file** and start executing tasks according to Orchestrator Sisyphus workflow
+
+## OUTPUT FORMAT
+
+When listing plans for selection:
+\`\`\`
+📋 Available Work Plans
+
+Current Time: {ISO timestamp}
+Session ID: {current session id}
+
+1. [plan-name-1.md] - Modified: {date} - Progress: 3/10 tasks
+2. [plan-name-2.md] - Modified: {date} - Progress: 0/5 tasks
+
+Which plan would you like to work on? (Enter number or plan name)
+\`\`\`
+
+When resuming existing work:
+\`\`\`
+🔄 Resuming Work Session
+
+Active Plan: {plan-name}
+Progress: {completed}/{total} tasks
+Sessions: {count} (appending current session)
+
+Reading plan and continuing from last incomplete task...
+\`\`\`
+
+When auto-selecting single plan:
+\`\`\`
+🚀 Starting Work Session
+
+Plan: {plan-name}
+Session ID: {session_id}
+Started: {timestamp}
+
+Reading plan and beginning execution...
+\`\`\`
+
+## CRITICAL
+
+- The session_id is injected by the hook - use it directly
+- Always update boulder.json BEFORE starting work
+- Read the FULL plan file before delegating any tasks
+- Follow Orchestrator Sisyphus delegation protocols (7-section format)`

src/features/builtin-commands/types.ts

@@ -1,6 +1,6 @@
 import type { CommandDefinition } from "../claude-code-command-loader"
 
-export type BuiltinCommandName = "init-deep" | "ralph-loop" | "cancel-ralph" | "refactor"
+export type BuiltinCommandName = "init-deep" | "ralph-loop" | "cancel-ralph" | "refactor" | "start-work"
 
 export interface BuiltinCommandConfig {
   disabled_commands?: BuiltinCommandName[]

src/features/builtin-skills/frontend-ui-ux/SKILL.md

@@ -0,0 +1,78 @@
+---
+name: frontend-ui-ux
+description: Designer-turned-developer who crafts stunning UI/UX even without design mockups
+---
+
+# Role: Designer-Turned-Developer
+
+You are a designer who learned to code. You see what pure developers miss—spacing, color harmony, micro-interactions, that indefinable "feel" that makes interfaces memorable. Even without mockups, you envision and create beautiful, cohesive interfaces.
+
+**Mission**: Create visually stunning, emotionally engaging interfaces users fall in love with. Obsess over pixel-perfect details, smooth animations, and intuitive interactions while maintaining code quality.
+
+---
+
+# Work Principles
+
+1. **Complete what's asked** — Execute the exact task. No scope creep. Work until it works. Never mark work complete without proper verification.
+2. **Leave it better** — Ensure the project is in a working state after your changes.
+3. **Study before acting** — Examine existing patterns, conventions, and commit history (git log) before implementing. Understand why code is structured the way it is.
+4. **Blend seamlessly** — Match existing code patterns. Your code should look like the team wrote it.
+5. **Be transparent** — Announce each step. Explain reasoning. Report both successes and failures.
+
+---
+
+# Design Process
+
+Before coding, commit to a **BOLD aesthetic direction**:
+
+1. **Purpose**: What problem does this solve? Who uses it?
+2. **Tone**: Pick an extreme—brutally minimal, maximalist chaos, retro-futuristic, organic/natural, luxury/refined, playful/toy-like, editorial/magazine, brutalist/raw, art deco/geometric, soft/pastel, industrial/utilitarian
+3. **Constraints**: Technical requirements (framework, performance, accessibility)
+4. **Differentiation**: What's the ONE thing someone will remember?
+
+**Key**: Choose a clear direction and execute with precision. Intentionality > intensity.
+
+Then implement working code (HTML/CSS/JS, React, Vue, Angular, etc.) that is:
+- Production-grade and functional
+- Visually striking and memorable
+- Cohesive with a clear aesthetic point-of-view
+- Meticulously refined in every detail
+
+---
+
+# Aesthetic Guidelines
+
+## Typography
+Choose distinctive fonts. **Avoid**: Arial, Inter, Roboto, system fonts, Space Grotesk. Pair a characterful display font with a refined body font.
+
+## Color
+Commit to a cohesive palette. Use CSS variables. Dominant colors with sharp accents outperform timid, evenly-distributed palettes. **Avoid**: purple gradients on white (AI slop).
+
+## Motion
+Focus on high-impact moments. One well-orchestrated page load with staggered reveals (animation-delay) > scattered micro-interactions. Use scroll-triggering and hover states that surprise. Prioritize CSS-only. Use Motion library for React when available.
+
+## Spatial Composition
+Unexpected layouts. Asymmetry. Overlap. Diagonal flow. Grid-breaking elements. Generous negative space OR controlled density.
+
+## Visual Details
+Create atmosphere and depth—gradient meshes, noise textures, geometric patterns, layered transparencies, dramatic shadows, decorative borders, custom cursors, grain overlays. Never default to solid colors.
+
+---
+
+# Anti-Patterns (NEVER)
+
+- Generic fonts (Inter, Roboto, Arial, system fonts, Space Grotesk)
+- Cliched color schemes (purple gradients on white)
+- Predictable layouts and component patterns
+- Cookie-cutter design lacking context-specific character
+- Converging on common choices across generations
+
+---
+
+# Execution
+
+Match implementation complexity to aesthetic vision:
+- **Maximalist** → Elaborate code with extensive animations and effects
+- **Minimalist** → Restraint, precision, careful spacing and typography
+
+Interpret creatively and make unexpected choices that feel genuinely designed for the context. No design should be the same. Vary between light and dark themes, different fonts, different aesthetics. You are capable of extraordinary creative work—don't hold back.

src/features/context-injector/injector.test.ts

@@ -207,7 +207,7 @@ describe("createContextInjectorMessagesTransformHook", () => {
     ],
   })
 
-  it("inserts synthetic message before last user message", async () => {
+  it("prepends context to last user message", async () => {
     // #given
     const hook = createContextInjectorMessagesTransformHook(collector)
     const sessionID = "ses_transform1"
@@ -228,10 +228,8 @@ describe("createContextInjectorMessagesTransformHook", () => {
     await hook["experimental.chat.messages.transform"]!({}, output)
 
     // #then
-    expect(output.messages.length).toBe(4)
-    expect(output.messages[2].parts[0].text).toBe("Ultrawork context")
-    expect(output.messages[2].parts[0].synthetic).toBe(true)
-    expect(output.messages[3].parts[0].text).toBe("Second message")
+    expect(output.messages.length).toBe(3)
+    expect(output.messages[2].parts[0].text).toBe("Ultrawork context\n\n---\n\nSecond message")
   })
 
   it("does nothing when no pending context", async () => {

src/features/context-injector/injector.ts

@@ -78,6 +78,9 @@ export function createContextInjectorMessagesTransformHook(
   return {
     "experimental.chat.messages.transform": async (_input, output) => {
       const { messages } = output
+      log("[DEBUG] experimental.chat.messages.transform called", {
+        messageCount: messages.length,
+      })
       if (messages.length === 0) {
         return
       }
@@ -91,16 +94,28 @@ export function createContextInjectorMessagesTransformHook(
       }
 
       if (lastUserMessageIndex === -1) {
+        log("[DEBUG] No user message found in messages")
         return
       }
 
       const lastUserMessage = messages[lastUserMessageIndex]
       const sessionID = (lastUserMessage.info as unknown as { sessionID?: string }).sessionID
+      log("[DEBUG] Extracted sessionID from lastUserMessage.info", {
+        sessionID,
+        infoKeys: Object.keys(lastUserMessage.info),
+        lastUserMessageInfo: JSON.stringify(lastUserMessage.info).slice(0, 200),
+      })
       if (!sessionID) {
+        log("[DEBUG] sessionID is undefined or empty")
         return
       }
 
-      if (!collector.hasPending(sessionID)) {
+      const hasPending = collector.hasPending(sessionID)
+      log("[DEBUG] Checking hasPending", {
+        sessionID,
+        hasPending,
+      })
+      if (!hasPending) {
         return
       }
 
@@ -109,47 +124,26 @@ export function createContextInjectorMessagesTransformHook(
         return
       }
 
-      const refInfo = lastUserMessage.info as unknown as {
-        sessionID?: string
-        agent?: string
-        model?: { providerID?: string; modelID?: string }
-        path?: { cwd?: string; root?: string }
+      const textPartIndex = lastUserMessage.parts.findIndex(
+        (p) => p.type === "text" && (p as { text?: string }).text
+      )
+
+      if (textPartIndex === -1) {
+        log("[context-injector] No text part found in last user message, skipping injection", {
+          sessionID,
+          partsCount: lastUserMessage.parts.length,
+        })
+        return
       }
 
-      const syntheticMessageId = `synthetic_ctx_${Date.now()}`
-      const syntheticPartId = `synthetic_ctx_part_${Date.now()}`
-      const now = Date.now()
+      const textPart = lastUserMessage.parts[textPartIndex] as { text?: string }
+      const originalText = textPart.text ?? ""
+      textPart.text = `${pending.merged}\n\n---\n\n${originalText}`
 
-      const syntheticMessage: MessageWithParts = {
-        info: {
-          id: syntheticMessageId,
-          sessionID: sessionID,
-          role: "user",
-          time: { created: now },
-          agent: refInfo.agent ?? "Sisyphus",
-          model: refInfo.model ?? { providerID: "unknown", modelID: "unknown" },
-          path: refInfo.path ?? { cwd: "/", root: "/" },
-        } as unknown as Message,
-        parts: [
-          {
-            id: syntheticPartId,
-            sessionID: sessionID,
-            messageID: syntheticMessageId,
-            type: "text",
-            text: pending.merged,
-            synthetic: true,
-            time: { start: now, end: now },
-          } as Part,
-        ],
-      }
-
-      messages.splice(lastUserMessageIndex, 0, syntheticMessage)
-
-      log("[context-injector] Injected synthetic message from collector", {
+      log("[context-injector] Prepended context to last user message", {
         sessionID,
-        insertIndex: lastUserMessageIndex,
         contextLength: pending.merged.length,
-        newMessageCount: messages.length,
+        originalTextLength: originalText.length,
       })
     },
   }

src/features/hook-message-injector/injector.ts

@@ -1,12 +1,12 @@
 import { existsSync, mkdirSync, readFileSync, readdirSync, writeFileSync } from "node:fs"
 import { join } from "node:path"
 import { MESSAGE_STORAGE, PART_STORAGE } from "./constants"
-import type { MessageMeta, OriginalMessageContext, TextPart } from "./types"
+import type { MessageMeta, OriginalMessageContext, TextPart, ToolPermission } from "./types"
 
 export interface StoredMessage {
   agent?: string
   model?: { providerID?: string; modelID?: string }
-  tools?: Record<string, boolean>
+  tools?: Record<string, ToolPermission>
 }
 
 export function findNearestMessageWithFields(messageDir: string): StoredMessage | null {
@@ -16,6 +16,7 @@ export function findNearestMessageWithFields(messageDir: string): StoredMessage
       .sort()
       .reverse()
 
+    // First pass: find message with ALL fields (ideal)
     for (const file of files) {
       try {
         const content = readFileSync(join(messageDir, file), "utf-8")
@@ -27,6 +28,20 @@ export function findNearestMessageWithFields(messageDir: string): StoredMessage
         continue
       }
     }
+
+    // Second pass: find message with ANY useful field (fallback)
+    // This ensures agent info isn't lost when model info is missing
+    for (const file of files) {
+      try {
+        const content = readFileSync(join(messageDir, file), "utf-8")
+        const msg = JSON.parse(content) as StoredMessage
+        if (msg.agent || (msg.model?.providerID && msg.model?.modelID)) {
+          return msg
+        }
+      } catch {
+        continue
+      }
+    }
   } catch {
     return null
   }

src/features/hook-message-injector/types.ts

@@ -1,3 +1,5 @@
+export type ToolPermission = boolean | "allow" | "deny" | "ask"
+
 export interface MessageMeta {
   id: string
   sessionID: string
@@ -15,7 +17,7 @@ export interface MessageMeta {
     cwd: string
     root: string
   }
-  tools?: Record<string, boolean>
+  tools?: Record<string, ToolPermission>
 }
 
 export interface OriginalMessageContext {
@@ -28,7 +30,7 @@ export interface OriginalMessageContext {
     cwd?: string
     root?: string
   }
-  tools?: Record<string, boolean>
+  tools?: Record<string, ToolPermission>
 }
 
 export interface TextPart {

src/features/opencode-skill-loader/index.ts

@@ -1,3 +1,4 @@
 export * from "./types"
 export * from "./loader"
 export * from "./merger"
+export * from "./skill-content"

src/features/opencode-skill-loader/skill-content.test.ts

@@ -0,0 +1,111 @@
+import { describe, it, expect } from "bun:test"
+import { resolveSkillContent, resolveMultipleSkills } from "./skill-content"
+
+describe("resolveSkillContent", () => {
+	it("should return template for existing skill", () => {
+		// #given: builtin skills with 'frontend-ui-ux' skill
+		// #when: resolving content for 'frontend-ui-ux'
+		const result = resolveSkillContent("frontend-ui-ux")
+
+		// #then: returns template string
+		expect(result).not.toBeNull()
+		expect(typeof result).toBe("string")
+		expect(result).toContain("Role: Designer-Turned-Developer")
+	})
+
+	it("should return template for 'playwright' skill", () => {
+		// #given: builtin skills with 'playwright' skill
+		// #when: resolving content for 'playwright'
+		const result = resolveSkillContent("playwright")
+
+		// #then: returns template string
+		expect(result).not.toBeNull()
+		expect(typeof result).toBe("string")
+		expect(result).toContain("Playwright Browser Automation")
+	})
+
+	it("should return null for non-existent skill", () => {
+		// #given: builtin skills without 'nonexistent' skill
+		// #when: resolving content for 'nonexistent'
+		const result = resolveSkillContent("nonexistent")
+
+		// #then: returns null
+		expect(result).toBeNull()
+	})
+
+	it("should return null for empty string", () => {
+		// #given: builtin skills
+		// #when: resolving content for empty string
+		const result = resolveSkillContent("")
+
+		// #then: returns null
+		expect(result).toBeNull()
+	})
+})
+
+describe("resolveMultipleSkills", () => {
+	it("should resolve all existing skills", () => {
+		// #given: list of existing skill names
+		const skillNames = ["frontend-ui-ux", "playwright"]
+
+		// #when: resolving multiple skills
+		const result = resolveMultipleSkills(skillNames)
+
+		// #then: all skills resolved, none not found
+		expect(result.resolved.size).toBe(2)
+		expect(result.notFound).toEqual([])
+		expect(result.resolved.get("frontend-ui-ux")).toContain("Designer-Turned-Developer")
+		expect(result.resolved.get("playwright")).toContain("Playwright Browser Automation")
+	})
+
+	it("should handle partial success - some skills not found", () => {
+		// #given: list with existing and non-existing skills
+		const skillNames = ["frontend-ui-ux", "nonexistent", "playwright", "another-missing"]
+
+		// #when: resolving multiple skills
+		const result = resolveMultipleSkills(skillNames)
+
+		// #then: resolves existing skills, lists not found skills
+		expect(result.resolved.size).toBe(2)
+		expect(result.notFound).toEqual(["nonexistent", "another-missing"])
+		expect(result.resolved.get("frontend-ui-ux")).toContain("Designer-Turned-Developer")
+		expect(result.resolved.get("playwright")).toContain("Playwright Browser Automation")
+	})
+
+	it("should handle empty array", () => {
+		// #given: empty skill names list
+		const skillNames: string[] = []
+
+		// #when: resolving multiple skills
+		const result = resolveMultipleSkills(skillNames)
+
+		// #then: returns empty resolved and notFound
+		expect(result.resolved.size).toBe(0)
+		expect(result.notFound).toEqual([])
+	})
+
+	it("should handle all skills not found", () => {
+		// #given: list of non-existing skills
+		const skillNames = ["skill-one", "skill-two", "skill-three"]
+
+		// #when: resolving multiple skills
+		const result = resolveMultipleSkills(skillNames)
+
+		// #then: no skills resolved, all in notFound
+		expect(result.resolved.size).toBe(0)
+		expect(result.notFound).toEqual(["skill-one", "skill-two", "skill-three"])
+	})
+
+	it("should preserve skill order in resolved map", () => {
+		// #given: list of skill names in specific order
+		const skillNames = ["playwright", "frontend-ui-ux"]
+
+		// #when: resolving multiple skills
+		const result = resolveMultipleSkills(skillNames)
+
+		// #then: map contains skills with expected keys
+		expect(result.resolved.has("playwright")).toBe(true)
+		expect(result.resolved.has("frontend-ui-ux")).toBe(true)
+		expect(result.resolved.size).toBe(2)
+	})
+})

src/features/opencode-skill-loader/skill-content.ts

@@ -0,0 +1,29 @@
+import { createBuiltinSkills } from "../builtin-skills/skills"
+
+export function resolveSkillContent(skillName: string): string | null {
+	const skills = createBuiltinSkills()
+	const skill = skills.find((s) => s.name === skillName)
+	return skill?.template ?? null
+}
+
+export function resolveMultipleSkills(skillNames: string[]): {
+	resolved: Map<string, string>
+	notFound: string[]
+} {
+	const skills = createBuiltinSkills()
+	const skillMap = new Map(skills.map((s) => [s.name, s.template]))
+
+	const resolved = new Map<string, string>()
+	const notFound: string[] = []
+
+	for (const name of skillNames) {
+		const template = skillMap.get(name)
+		if (template) {
+			resolved.set(name, template)
+		} else {
+			notFound.push(name)
+		}
+	}
+
+	return { resolved, notFound }
+}

src/features/task-toast-manager/index.ts

@@ -0,0 +1,2 @@
+export { TaskToastManager, getTaskToastManager, initTaskToastManager } from "./manager"
+export type { TrackedTask, TaskStatus, TaskToastOptions } from "./types"

src/features/task-toast-manager/manager.test.ts

@@ -0,0 +1,145 @@
+import { describe, test, expect, beforeEach, mock } from "bun:test"
+import { TaskToastManager } from "./manager"
+import type { ConcurrencyManager } from "../background-agent/concurrency"
+
+describe("TaskToastManager", () => {
+  let mockClient: {
+    tui: {
+      showToast: ReturnType<typeof mock>
+    }
+  }
+  let toastManager: TaskToastManager
+  let mockConcurrencyManager: ConcurrencyManager
+
+  beforeEach(() => {
+    mockClient = {
+      tui: {
+        showToast: mock(() => Promise.resolve()),
+      },
+    }
+    mockConcurrencyManager = {
+      getConcurrencyLimit: mock(() => 5),
+    } as unknown as ConcurrencyManager
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    toastManager = new TaskToastManager(mockClient as any, mockConcurrencyManager)
+  })
+
+  describe("skills in toast message", () => {
+    test("should display skills when provided", () => {
+      // #given - a task with skills
+      const task = {
+        id: "task_1",
+        description: "Test task",
+        agent: "Sisyphus-Junior",
+        isBackground: true,
+        skills: ["playwright", "git-master"],
+      }
+
+      // #when - addTask is called
+      toastManager.addTask(task)
+
+      // #then - toast message should include skills
+      expect(mockClient.tui.showToast).toHaveBeenCalled()
+      const call = mockClient.tui.showToast.mock.calls[0][0]
+      expect(call.body.message).toContain("playwright")
+      expect(call.body.message).toContain("git-master")
+    })
+
+    test("should not display skills section when no skills provided", () => {
+      // #given - a task without skills
+      const task = {
+        id: "task_2",
+        description: "Test task without skills",
+        agent: "explore",
+        isBackground: true,
+      }
+
+      // #when - addTask is called
+      toastManager.addTask(task)
+
+      // #then - toast message should not include skills prefix
+      expect(mockClient.tui.showToast).toHaveBeenCalled()
+      const call = mockClient.tui.showToast.mock.calls[0][0]
+      expect(call.body.message).not.toContain("Skills:")
+    })
+  })
+
+  describe("concurrency info in toast message", () => {
+    test("should display concurrency status in toast", () => {
+      // #given - multiple running tasks
+      toastManager.addTask({
+        id: "task_1",
+        description: "First task",
+        agent: "explore",
+        isBackground: true,
+      })
+      toastManager.addTask({
+        id: "task_2",
+        description: "Second task",
+        agent: "librarian",
+        isBackground: true,
+      })
+
+      // #when - third task is added
+      toastManager.addTask({
+        id: "task_3",
+        description: "Third task",
+        agent: "explore",
+        isBackground: true,
+      })
+
+      // #then - toast should show concurrency info
+      expect(mockClient.tui.showToast).toHaveBeenCalledTimes(3)
+      const lastCall = mockClient.tui.showToast.mock.calls[2][0]
+      // Should show "Running (3):" header
+      expect(lastCall.body.message).toContain("Running (3):")
+    })
+
+    test("should display concurrency limit info when available", () => {
+      // #given - a concurrency manager with known limit
+      const mockConcurrencyWithCounts = {
+        getConcurrencyLimit: mock(() => 5),
+        getRunningCount: mock(() => 2),
+        getQueuedCount: mock(() => 1),
+      } as unknown as ConcurrencyManager
+
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      const managerWithConcurrency = new TaskToastManager(mockClient as any, mockConcurrencyWithCounts)
+
+      // #when - a task is added
+      managerWithConcurrency.addTask({
+        id: "task_1",
+        description: "Test task",
+        agent: "explore",
+        isBackground: true,
+      })
+
+      // #then - toast should show concurrency status like "2/5 slots"
+      expect(mockClient.tui.showToast).toHaveBeenCalled()
+      const call = mockClient.tui.showToast.mock.calls[0][0]
+      expect(call.body.message).toMatch(/\d+\/\d+/)
+    })
+  })
+
+  describe("combined skills and concurrency display", () => {
+    test("should display both skills and concurrency info together", () => {
+      // #given - a task with skills and concurrency manager
+      const task = {
+        id: "task_1",
+        description: "Full info task",
+        agent: "Sisyphus-Junior",
+        isBackground: true,
+        skills: ["frontend-ui-ux"],
+      }
+
+      // #when - addTask is called
+      toastManager.addTask(task)
+
+      // #then - toast should include both skills and task count
+      expect(mockClient.tui.showToast).toHaveBeenCalled()
+      const call = mockClient.tui.showToast.mock.calls[0][0]
+      expect(call.body.message).toContain("frontend-ui-ux")
+      expect(call.body.message).toContain("Running (1):")
+    })
+  })
+})

src/features/task-toast-manager/manager.ts

@@ -0,0 +1,199 @@
+import type { PluginInput } from "@opencode-ai/plugin"
+import type { TrackedTask, TaskStatus } from "./types"
+import type { ConcurrencyManager } from "../background-agent/concurrency"
+
+type OpencodeClient = PluginInput["client"]
+
+export class TaskToastManager {
+  private tasks: Map<string, TrackedTask> = new Map()
+  private client: OpencodeClient
+  private concurrencyManager?: ConcurrencyManager
+
+  constructor(client: OpencodeClient, concurrencyManager?: ConcurrencyManager) {
+    this.client = client
+    this.concurrencyManager = concurrencyManager
+  }
+
+  setConcurrencyManager(manager: ConcurrencyManager): void {
+    this.concurrencyManager = manager
+  }
+
+  addTask(task: {
+    id: string
+    description: string
+    agent: string
+    isBackground: boolean
+    status?: TaskStatus
+    skills?: string[]
+  }): void {
+    const trackedTask: TrackedTask = {
+      id: task.id,
+      description: task.description,
+      agent: task.agent,
+      status: task.status ?? "running",
+      startedAt: new Date(),
+      isBackground: task.isBackground,
+      skills: task.skills,
+    }
+
+    this.tasks.set(task.id, trackedTask)
+    this.showTaskListToast(trackedTask)
+  }
+
+  /**
+   * Update task status
+   */
+  updateTask(id: string, status: TaskStatus): void {
+    const task = this.tasks.get(id)
+    if (task) {
+      task.status = status
+    }
+  }
+
+  /**
+   * Remove completed/error task
+   */
+  removeTask(id: string): void {
+    this.tasks.delete(id)
+  }
+
+  /**
+   * Get all running tasks (newest first)
+   */
+  getRunningTasks(): TrackedTask[] {
+    const running = Array.from(this.tasks.values())
+      .filter((t) => t.status === "running")
+      .sort((a, b) => b.startedAt.getTime() - a.startedAt.getTime())
+    return running
+  }
+
+  /**
+   * Get all queued tasks
+   */
+  getQueuedTasks(): TrackedTask[] {
+    return Array.from(this.tasks.values())
+      .filter((t) => t.status === "queued")
+      .sort((a, b) => a.startedAt.getTime() - b.startedAt.getTime())
+  }
+
+  /**
+   * Format duration since task started
+   */
+  private formatDuration(startedAt: Date): string {
+    const seconds = Math.floor((Date.now() - startedAt.getTime()) / 1000)
+    if (seconds < 60) return `${seconds}s`
+    const minutes = Math.floor(seconds / 60)
+    if (minutes < 60) return `${minutes}m ${seconds % 60}s`
+    const hours = Math.floor(minutes / 60)
+    return `${hours}h ${minutes % 60}m`
+  }
+
+  private getConcurrencyInfo(): string {
+    if (!this.concurrencyManager) return ""
+    const running = this.getRunningTasks()
+    const queued = this.getQueuedTasks()
+    const total = running.length + queued.length
+    const limit = this.concurrencyManager.getConcurrencyLimit("default")
+    if (limit === Infinity) return ""
+    return ` [${total}/${limit}]`
+  }
+
+  private buildTaskListMessage(newTask: TrackedTask): string {
+    const running = this.getRunningTasks()
+    const queued = this.getQueuedTasks()
+    const concurrencyInfo = this.getConcurrencyInfo()
+
+    const lines: string[] = []
+
+    if (running.length > 0) {
+      lines.push(`Running (${running.length}):${concurrencyInfo}`)
+      for (const task of running) {
+        const duration = this.formatDuration(task.startedAt)
+        const bgIcon = task.isBackground ? "⚡" : "🔄"
+        const isNew = task.id === newTask.id ? " ← NEW" : ""
+        const skillsInfo = task.skills?.length ? ` [${task.skills.join(", ")}]` : ""
+        lines.push(`${bgIcon} ${task.description} (${task.agent})${skillsInfo} - ${duration}${isNew}`)
+      }
+    }
+
+    if (queued.length > 0) {
+      if (lines.length > 0) lines.push("")
+      lines.push(`Queued (${queued.length}):`)
+      for (const task of queued) {
+        const bgIcon = task.isBackground ? "⏳" : "⏸️"
+        const skillsInfo = task.skills?.length ? ` [${task.skills.join(", ")}]` : ""
+        lines.push(`${bgIcon} ${task.description} (${task.agent})${skillsInfo}`)
+      }
+    }
+
+    return lines.join("\n")
+  }
+
+  /**
+   * Show consolidated toast with all running/queued tasks
+   */
+  private showTaskListToast(newTask: TrackedTask): void {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    const tuiClient = this.client as any
+    if (!tuiClient.tui?.showToast) return
+
+    const message = this.buildTaskListMessage(newTask)
+    const running = this.getRunningTasks()
+    const queued = this.getQueuedTasks()
+
+    const title = newTask.isBackground
+      ? `⚡ New Background Task`
+      : `🔄 New Task Executed`
+
+    tuiClient.tui.showToast({
+      body: {
+        title,
+        message: message || `${newTask.description} (${newTask.agent})`,
+        variant: "info",
+        duration: running.length + queued.length > 2 ? 5000 : 3000,
+      },
+    }).catch(() => {})
+  }
+
+  /**
+   * Show task completion toast
+   */
+  showCompletionToast(task: { id: string; description: string; duration: string }): void {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    const tuiClient = this.client as any
+    if (!tuiClient.tui?.showToast) return
+
+    this.removeTask(task.id)
+
+    const remaining = this.getRunningTasks()
+    const queued = this.getQueuedTasks()
+
+    let message = `✅ "${task.description}" finished in ${task.duration}`
+    if (remaining.length > 0 || queued.length > 0) {
+      message += `\n\nStill running: ${remaining.length} | Queued: ${queued.length}`
+    }
+
+    tuiClient.tui.showToast({
+      body: {
+        title: "Task Completed",
+        message,
+        variant: "success",
+        duration: 5000,
+      },
+    }).catch(() => {})
+  }
+}
+
+let instance: TaskToastManager | null = null
+
+export function getTaskToastManager(): TaskToastManager | null {
+  return instance
+}
+
+export function initTaskToastManager(
+  client: OpencodeClient,
+  concurrencyManager?: ConcurrencyManager
+): TaskToastManager {
+  instance = new TaskToastManager(client, concurrencyManager)
+  return instance
+}

src/features/task-toast-manager/types.ts

@@ -0,0 +1,18 @@
+export type TaskStatus = "running" | "queued" | "completed" | "error"
+
+export interface TrackedTask {
+  id: string
+  description: string
+  agent: string
+  status: TaskStatus
+  startedAt: Date
+  isBackground: boolean
+  skills?: string[]
+}
+
+export interface TaskToastOptions {
+  title: string
+  message: string
+  variant: "info" | "success" | "warning" | "error"
+  duration?: number
+}

src/hooks/AGENTS.md

@@ -2,35 +2,42 @@
 
 ## OVERVIEW
 
-22 lifecycle hooks intercepting/modifying agent behavior. Context injection, error recovery, output control, notifications.
+22+ lifecycle hooks intercepting/modifying agent behavior. Context injection, error recovery, output control, notifications.
 
 ## STRUCTURE
 
 ```
 hooks/
-├── anthropic-context-window-limit-recovery/  # Auto-compact at token limit (554 lines)
+├── anthropic-context-window-limit-recovery/  # Auto-compact at token limit (556 lines)
 ├── auto-slash-command/         # Detect and execute /command patterns
 ├── auto-update-checker/        # Version notifications, startup toast
 ├── background-notification/    # OS notify on task complete
-├── claude-code-hooks/          # settings.json PreToolUse/PostToolUse/etc
+├── claude-code-hooks/          # settings.json PreToolUse/PostToolUse/etc (408 lines)
 ├── comment-checker/            # Prevent excessive AI comments
-│   └── filters/                # docstring, directive, bdd, etc
+│   ├── filters/                # docstring, directive, bdd, shebang
+│   └── output/                 # XML builder, formatter
 ├── compaction-context-injector/ # Preserve context during compaction
 ├── directory-agents-injector/  # Auto-inject AGENTS.md
 ├── directory-readme-injector/  # Auto-inject README.md
+├── edit-error-recovery/        # Recover from edit failures
 ├── empty-message-sanitizer/    # Sanitize empty messages
 ├── interactive-bash-session/   # Tmux session management
 ├── keyword-detector/           # ultrawork/search keyword activation
 ├── non-interactive-env/        # CI/headless handling
 ├── preemptive-compaction/      # Pre-emptive at 85% usage
+├── prometheus-md-only/         # Restrict prometheus to read-only
 ├── ralph-loop/                 # Self-referential dev loop
 ├── rules-injector/             # Conditional rules from .claude/rules/
-├── session-recovery/           # Recover from errors (430 lines)
+├── session-recovery/           # Recover from errors (432 lines)
+├── sisyphus-orchestrator/      # Main orchestration hook (660 lines)
+├── start-work/                 # Initialize Sisyphus work session
+├── task-resume-info/           # Track task resume state
 ├── think-mode/                 # Auto-detect thinking triggers
+├── thinking-block-validator/   # Validate thinking block format
 ├── agent-usage-reminder/       # Remind to use specialists
 ├── context-window-monitor.ts   # Monitor usage (standalone)
 ├── session-notification.ts     # OS notify on idle
-├── todo-continuation-enforcer.ts # Force TODO completion
+├── todo-continuation-enforcer.ts # Force TODO completion (413 lines)
 └── tool-output-truncator.ts    # Truncate verbose outputs
 ```
 

src/hooks/agent-usage-reminder/constants.ts

@@ -15,6 +15,8 @@ export const TARGET_TOOLS = new Set([
   "safe_glob",
   "webfetch",
   "context7_resolve-library-id",
+  "context7_query-docs",
+  "websearch_web_search_exa",
   "context7_get-library-docs",
   "grep_app_searchgithub",
 ]);
@@ -22,7 +24,7 @@ export const TARGET_TOOLS = new Set([
 export const AGENT_TOOLS = new Set([
   "task",
   "call_omo_agent",
-  "background_task",
+  "sisyphus_task",
 ]);
 
 export const REMINDER_MESSAGE = `
@@ -30,13 +32,13 @@ export const REMINDER_MESSAGE = `
 
 You called a search/fetch tool directly without leveraging specialized agents.
 
-RECOMMENDED: Use background_task with explore/librarian agents for better results:
+RECOMMENDED: Use sisyphus_task with explore/librarian agents for better results:
 
 \`\`\`
 // Parallel exploration - fire multiple agents simultaneously
-background_task(agent="explore", prompt="Find all files matching pattern X")
-background_task(agent="explore", prompt="Search for implementation of Y") 
-background_task(agent="librarian", prompt="Lookup documentation for Z")
+sisyphus_task(agent="explore", prompt="Find all files matching pattern X")
+sisyphus_task(agent="explore", prompt="Search for implementation of Y") 
+sisyphus_task(agent="librarian", prompt="Lookup documentation for Z")
 
 // Then continue your work while they run in background
 // System will notify you when each completes
@@ -48,5 +50,5 @@ WHY:
 - Specialized agents have domain expertise
 - Reduces context window usage in main session
 
-ALWAYS prefer: Multiple parallel background_task calls > Direct tool calls
+ALWAYS prefer: Multiple parallel sisyphus_task calls > Direct tool calls
 `;

src/hooks/anthropic-context-window-limit-recovery/executor.test.ts

@@ -257,7 +257,7 @@ describe("executeCompact lock management", () => {
     expect(mockClient.session.summarize).toHaveBeenCalledWith(
       expect.objectContaining({
         path: { id: sessionID },
-        body: { providerID: "anthropic", modelID: "claude-opus-4-5" },
+        body: { providerID: "anthropic", modelID: "claude-opus-4-5", auto: true },
       }),
     )
 

src/hooks/anthropic-context-window-limit-recovery/executor.ts

@@ -409,7 +409,7 @@ export async function executeCompact(
             try {
               await (client as Client).session.prompt_async({
                 path: { id: sessionID },
-                body: { parts: [{ type: "text", text: "Continue" }] },
+                body: { auto: true } as never,
                 query: { directory },
               });
             } catch {}
@@ -497,21 +497,12 @@ export async function executeCompact(
             })
             .catch(() => {});
 
+          const summarizeBody = { providerID, modelID, auto: true }
           await (client as Client).session.summarize({
             path: { id: sessionID },
-            body: { providerID, modelID },
+            body: summarizeBody as never,
             query: { directory },
           });
-
-          setTimeout(async () => {
-            try {
-              await (client as Client).session.prompt_async({
-                path: { id: sessionID },
-                body: { parts: [{ type: "text", text: "Continue" }] },
-                query: { directory },
-              });
-            } catch {}
-          }, 500);
           return;
         } catch {
           const delay =

src/hooks/auto-update-checker/index.test.ts

@@ -0,0 +1,153 @@
+import { describe, test, expect } from "bun:test"
+import { isPrereleaseVersion, isDistTag, isPrereleaseOrDistTag } from "./index"
+
+describe("auto-update-checker", () => {
+  describe("isPrereleaseVersion", () => {
+    test("returns true for beta versions", () => {
+      // #given a beta version
+      const version = "3.0.0-beta.1"
+
+      // #when checking if prerelease
+      const result = isPrereleaseVersion(version)
+
+      // #then returns true
+      expect(result).toBe(true)
+    })
+
+    test("returns true for alpha versions", () => {
+      // #given an alpha version
+      const version = "1.0.0-alpha"
+
+      // #when checking if prerelease
+      const result = isPrereleaseVersion(version)
+
+      // #then returns true
+      expect(result).toBe(true)
+    })
+
+    test("returns true for rc versions", () => {
+      // #given an rc version
+      const version = "2.0.0-rc.1"
+
+      // #when checking if prerelease
+      const result = isPrereleaseVersion(version)
+
+      // #then returns true
+      expect(result).toBe(true)
+    })
+
+    test("returns false for stable versions", () => {
+      // #given a stable version
+      const version = "2.14.0"
+
+      // #when checking if prerelease
+      const result = isPrereleaseVersion(version)
+
+      // #then returns false
+      expect(result).toBe(false)
+    })
+  })
+
+  describe("isDistTag", () => {
+    test("returns true for beta dist-tag", () => {
+      // #given beta dist-tag
+      const version = "beta"
+
+      // #when checking if dist-tag
+      const result = isDistTag(version)
+
+      // #then returns true
+      expect(result).toBe(true)
+    })
+
+    test("returns true for next dist-tag", () => {
+      // #given next dist-tag
+      const version = "next"
+
+      // #when checking if dist-tag
+      const result = isDistTag(version)
+
+      // #then returns true
+      expect(result).toBe(true)
+    })
+
+    test("returns true for canary dist-tag", () => {
+      // #given canary dist-tag
+      const version = "canary"
+
+      // #when checking if dist-tag
+      const result = isDistTag(version)
+
+      // #then returns true
+      expect(result).toBe(true)
+    })
+
+    test("returns false for semver versions", () => {
+      // #given a semver version
+      const version = "2.14.0"
+
+      // #when checking if dist-tag
+      const result = isDistTag(version)
+
+      // #then returns false
+      expect(result).toBe(false)
+    })
+
+    test("returns false for latest (handled separately)", () => {
+      // #given latest tag
+      const version = "latest"
+
+      // #when checking if dist-tag
+      const result = isDistTag(version)
+
+      // #then returns true (but latest is filtered before this check)
+      expect(result).toBe(true)
+    })
+  })
+
+  describe("isPrereleaseOrDistTag", () => {
+    test("returns false for null", () => {
+      // #given null version
+      const version = null
+
+      // #when checking
+      const result = isPrereleaseOrDistTag(version)
+
+      // #then returns false
+      expect(result).toBe(false)
+    })
+
+    test("returns true for prerelease version", () => {
+      // #given prerelease version
+      const version = "3.0.0-beta.1"
+
+      // #when checking
+      const result = isPrereleaseOrDistTag(version)
+
+      // #then returns true
+      expect(result).toBe(true)
+    })
+
+    test("returns true for dist-tag", () => {
+      // #given dist-tag
+      const version = "beta"
+
+      // #when checking
+      const result = isPrereleaseOrDistTag(version)
+
+      // #then returns true
+      expect(result).toBe(true)
+    })
+
+    test("returns false for stable version", () => {
+      // #given stable version
+      const version = "2.14.0"
+
+      // #when checking
+      const result = isPrereleaseOrDistTag(version)
+
+      // #then returns false
+      expect(result).toBe(false)
+    })
+  })
+})

src/hooks/auto-update-checker/index.ts

@@ -9,6 +9,20 @@ import type { AutoUpdateCheckerOptions } from "./types"
 
 const SISYPHUS_SPINNER = ["·", "•", "●", "○", "◌", "◦", " "]
 
+export function isPrereleaseVersion(version: string): boolean {
+  return version.includes("-")
+}
+
+export function isDistTag(version: string): boolean {
+  const startsWithDigit = /^\d/.test(version)
+  return !startsWithDigit
+}
+
+export function isPrereleaseOrDistTag(pinnedVersion: string | null): boolean {
+  if (!pinnedVersion) return false
+  return isPrereleaseVersion(pinnedVersion) || isDistTag(pinnedVersion)
+}
+
 export function createAutoUpdateCheckerHook(ctx: PluginInput, options: AutoUpdateCheckerOptions = {}) {
   const { showStartupToast = true, isSisyphusEnabled = false, autoUpdate = true } = options
 
@@ -63,7 +77,7 @@ export function createAutoUpdateCheckerHook(ctx: PluginInput, options: AutoUpdat
 }
 
 async function runBackgroundUpdateCheck(
-  ctx: PluginInput, 
+  ctx: PluginInput,
   autoUpdate: boolean,
   getToastMessage: (isUpdate: boolean, latestVersion?: string) => string
 ): Promise<void> {
@@ -99,7 +113,18 @@ async function runBackgroundUpdateCheck(
     return
   }
 
+  // Check if current version is a prerelease - don't auto-downgrade prerelease to stable
+  if (isPrereleaseVersion(currentVersion)) {
+    log(`[auto-update-checker] Skipping auto-update for prerelease version: ${currentVersion}`)
+    return
+  }
+
   if (pluginInfo.isPinned) {
+    if (isPrereleaseOrDistTag(pluginInfo.pinnedVersion)) {
+      log(`[auto-update-checker] Skipping auto-update for prerelease/dist-tag: ${pluginInfo.pinnedVersion}`)
+      return
+    }
+
     const updated = updatePinnedVersion(pluginInfo.configPath, pluginInfo.entry, latestVersion)
     if (!updated) {
       await showUpdateAvailableToast(ctx, latestVersion, getToastMessage)
@@ -112,7 +137,7 @@ async function runBackgroundUpdateCheck(
   invalidatePackage(PACKAGE_NAME)
 
   const installSuccess = await runBunInstallSafe()
-  
+
   if (installSuccess) {
     await showAutoUpdatedToast(ctx, currentVersion, latestVersion)
     log(`[auto-update-checker] Update installed: ${currentVersion} → ${latestVersion}`)
@@ -180,7 +205,7 @@ async function showSpinnerToast(ctx: PluginInput, version: string, message: stri
 }
 
 async function showUpdateAvailableToast(
-  ctx: PluginInput, 
+  ctx: PluginInput,
   latestVersion: string,
   getToastMessage: (isUpdate: boolean, latestVersion?: string) => string
 ): Promise<void> {

src/hooks/background-compaction/index.ts

@@ -0,0 +1,85 @@
+import type { BackgroundManager } from "../../features/background-agent"
+
+interface CompactingInput {
+  sessionID: string
+}
+
+interface CompactingOutput {
+  context: string[]
+  prompt?: string
+}
+
+/**
+ * Background agent compaction hook - preserves task state during context compaction.
+ * 
+ * When OpenCode compacts session context to save tokens, this hook injects
+ * information about running and recently completed background tasks so the
+ * agent doesn't lose awareness of delegated work.
+ */
+export function createBackgroundCompactionHook(manager: BackgroundManager) {
+  return {
+    "experimental.session.compacting": async (
+      input: CompactingInput,
+      output: CompactingOutput
+    ): Promise<void> => {
+      const { sessionID } = input
+
+      // Get running tasks for this session
+      const running = manager.getRunningTasks()
+        .filter(t => t.parentSessionID === sessionID)
+        .map(t => ({
+          id: t.id,
+          agent: t.agent,
+          description: t.description,
+          startedAt: t.startedAt,
+        }))
+
+      // Get recently completed tasks (still in memory within 5-min retention)
+      const completed = manager.getCompletedTasks()
+        .filter(t => t.parentSessionID === sessionID)
+        .slice(-10) // Last 10 completed
+        .map(t => ({
+          id: t.id,
+          agent: t.agent,
+          description: t.description,
+          status: t.status,
+        }))
+
+      // Early exit if nothing to preserve
+      if (running.length === 0 && completed.length === 0) return
+
+      const sections: string[] = ["<background-tasks>"]
+
+      // Running tasks section
+      if (running.length > 0) {
+        sections.push("## Running Background Tasks")
+        sections.push("")
+        for (const t of running) {
+          const elapsed = Math.floor((Date.now() - t.startedAt.getTime()) / 1000)
+          sections.push(`- **\`${t.id}\`** (${t.agent}): ${t.description} [${elapsed}s elapsed]`)
+        }
+        sections.push("")
+        sections.push("> **Note:** You WILL be notified when tasks complete.")
+        sections.push("> Do NOT poll - continue productive work.")
+        sections.push("")
+      }
+
+      // Completed tasks section
+      if (completed.length > 0) {
+        sections.push("## Recently Completed Tasks")
+        sections.push("")
+        for (const t of completed) {
+          const statusEmoji = t.status === "completed" ? "✅" : t.status === "error" ? "❌" : "⏱️"
+          sections.push(`- ${statusEmoji} **\`${t.id}\`**: ${t.description}`)
+        }
+        sections.push("")
+      }
+
+      sections.push("## Retrieval")
+      sections.push('Use `background_output(task_id="<id>")` to retrieve task results.')
+      sections.push("</background-tasks>")
+
+      output.context.push(sections.join("\n"))
+    }
+  }
+}

src/hooks/background-notification/index.ts

@@ -9,6 +9,12 @@ interface EventInput {
   event: Event
 }
 
+/**
+ * Background notification hook - handles event routing to BackgroundManager.
+ * 
+ * Notifications are now delivered directly via session.prompt({ noReply }) 
+ * from the manager, so this hook only needs to handle event routing.
+ */
 export function createBackgroundNotificationHook(manager: BackgroundManager) {
   const eventHandler = async ({ event }: EventInput) => {
     manager.handleEvent(event)

src/hooks/claude-code-hooks/index.ts

@@ -27,14 +27,18 @@ import { cacheToolInput, getToolInput } from "./tool-input-cache"
 import { recordToolUse, recordToolResult, getTranscriptPath, recordUserMessage } from "./transcript"
 import type { PluginConfig } from "./types"
 import { log, isHookDisabled } from "../../shared"
-import { injectHookMessage } from "../../features/hook-message-injector"
 import { detectKeywordsWithType, removeCodeBlocks } from "../keyword-detector"
+import type { ContextCollector } from "../../features/context-injector"
 
 const sessionFirstMessageProcessed = new Set<string>()
 const sessionErrorState = new Map<string, { hasError: boolean; errorMessage?: string }>()
 const sessionInterruptState = new Map<string, { interrupted: boolean }>()
 
-export function createClaudeCodeHooksHook(ctx: PluginInput, config: PluginConfig = {}) {
+export function createClaudeCodeHooksHook(
+  ctx: PluginInput,
+  config: PluginConfig = {},
+  contextCollector?: ContextCollector
+) {
   return {
     "experimental.session.compacting": async (
       input: { sessionID: string },
@@ -164,24 +168,31 @@ export function createClaudeCodeHooksHook(ctx: PluginInput, config: PluginConfig
               output.parts[idx].text = `${hookContent}\n\n${output.parts[idx].text ?? ""}`
               log("UserPromptSubmit hooks prepended to first message parts directly", { sessionID: input.sessionID })
             }
-          } else {
-            const message = output.message as {
-              agent?: string
-              model?: { modelID?: string; providerID?: string }
-              path?: { cwd?: string; root?: string }
-              tools?: Record<string, boolean>
-            }
-
-            const success = injectHookMessage(input.sessionID, hookContent, {
-              agent: message.agent,
-              model: message.model,
-              path: message.path ?? { cwd: ctx.directory, root: "/" },
-              tools: message.tools,
-            })
-
-            log(success ? "Hook message injected via file system" : "File injection failed", {
+          } else if (contextCollector) {
+            log("[DEBUG] Registering hook content to contextCollector", {
               sessionID: input.sessionID,
+              contentLength: hookContent.length,
+              contentPreview: hookContent.slice(0, 100),
             })
+            contextCollector.register(input.sessionID, {
+              id: "hook-context",
+              source: "custom",
+              content: hookContent,
+              priority: "high",
+            })
+
+            log("Hook content registered for synthetic message injection", {
+              sessionID: input.sessionID,
+              contentLength: hookContent.length,
+            })
+          } else {
+            const idx = output.parts.findIndex((p) => p.type === "text" && p.text)
+            if (idx >= 0) {
+              output.parts[idx].text = `${hookContent}\n\n${output.parts[idx].text ?? ""}`
+              log("Hook content prepended to message (fallback)", {
+                sessionID: input.sessionID,
+              })
+            }
           }
         }
       }
@@ -239,7 +250,7 @@ export function createClaudeCodeHooksHook(ctx: PluginInput, config: PluginConfig
       const cachedInput = getToolInput(input.sessionID, input.tool, input.callID) || {}
 
       // Use metadata if available and non-empty, otherwise wrap output.output in a structured object
-      // This ensures plugin tools (call_omo_agent, background_task, task) that return strings
+      // This ensures plugin tools (call_omo_agent, sisyphus_task, task) that return strings
       // get their results properly recorded in transcripts instead of empty {}
       const metadata = output.metadata as Record<string, unknown> | undefined
       const hasMetadata = metadata && typeof metadata === "object" && Object.keys(metadata).length > 0

src/hooks/index.ts

@@ -14,6 +14,7 @@ export { createThinkModeHook } from "./think-mode";
 export { createClaudeCodeHooksHook } from "./claude-code-hooks";
 export { createRulesInjectorHook } from "./rules-injector";
 export { createBackgroundNotificationHook } from "./background-notification"
+export { createBackgroundCompactionHook } from "./background-compaction"
 export { createAutoUpdateCheckerHook } from "./auto-update-checker";
 
 export { createAgentUsageReminderHook } from "./agent-usage-reminder";
@@ -25,3 +26,7 @@ export { createThinkingBlockValidatorHook } from "./thinking-block-validator";
 export { createRalphLoopHook, type RalphLoopHook } from "./ralph-loop";
 export { createAutoSlashCommandHook } from "./auto-slash-command";
 export { createEditErrorRecoveryHook } from "./edit-error-recovery";
+export { createPrometheusMdOnlyHook } from "./prometheus-md-only";
+export { createTaskResumeInfoHook } from "./task-resume-info";
+export { createStartWorkHook } from "./start-work";
+export { createSisyphusOrchestratorHook } from "./sisyphus-orchestrator";

src/hooks/keyword-detector/constants.ts

@@ -101,14 +101,14 @@ TELL THE USER WHAT AGENTS YOU WILL LEVERAGE NOW TO SATISFY USER'S REQUEST.
 
 ## EXECUTION RULES
 - **TODO**: Track EVERY step. Mark complete IMMEDIATELY after each.
-- **PARALLEL**: Fire independent agent calls simultaneously via background_task - NEVER wait sequentially.
-- **BACKGROUND FIRST**: Use background_task for exploration/research agents (10+ concurrent if needed).
+- **PARALLEL**: Fire independent agent calls simultaneously via sisyphus_task(background=true) - NEVER wait sequentially.
+- **BACKGROUND FIRST**: Use sisyphus_task for exploration/research agents (10+ concurrent if needed).
 - **VERIFY**: Re-read request after completion. Check ALL requirements met before reporting done.
 - **DELEGATE**: Don't do everything yourself - orchestrate specialized agents for their strengths.
 
 ## WORKFLOW
 1. Analyze the request and identify required capabilities
-2. Spawn exploration/librarian agents via background_task in PARALLEL (10+ if needed)
+2. Spawn exploration/librarian agents via sisyphus_task(background=true) in PARALLEL (10+ if needed)
 3. Always Use Plan agent with gathered context to create detailed work breakdown
 4. Execute with continuous verification against original requirements
 

src/hooks/preemptive-compaction/index.ts

@@ -169,9 +169,10 @@ export function createPreemptiveCompactionHook(
         })
       }
 
+      const summarizeBody = { providerID, modelID, auto: true }
       await ctx.client.session.summarize({
         path: { id: sessionID },
-        body: { providerID, modelID },
+        body: summarizeBody as never,
         query: { directory: ctx.directory },
       })
 
@@ -187,22 +188,6 @@ export function createPreemptiveCompactionHook(
         .catch(() => {})
 
       state.compactionInProgress.delete(sessionID)
-
-      setTimeout(async () => {
-        try {
-          const messageDir = getMessageDir(sessionID)
-          const storedMessage = messageDir ? findNearestMessageWithFields(messageDir) : null
-
-          await ctx.client.session.promptAsync({
-            path: { id: sessionID },
-            body: {
-              agent: storedMessage?.agent,
-              parts: [{ type: "text", text: "Continue" }],
-            },
-            query: { directory: ctx.directory },
-          })
-        } catch {}
-      }, 500)
       return
     } catch (err) {
       log("[preemptive-compaction] compaction failed", { sessionID, error: err })

src/hooks/prometheus-md-only/constants.ts

@@ -0,0 +1,30 @@
+export const HOOK_NAME = "prometheus-md-only"
+
+export const PROMETHEUS_AGENTS = ["Prometheus (Planner)"]
+
+export const ALLOWED_EXTENSIONS = [".md"]
+
+export const ALLOWED_PATH_PREFIX = ".sisyphus/"
+
+export const BLOCKED_TOOLS = ["Write", "Edit", "write", "edit"]
+
+export const PLANNING_CONSULT_WARNING = `
+
+---
+
+[SYSTEM DIRECTIVE - READ-ONLY PLANNING CONSULTATION]
+
+You are being invoked by Prometheus (Planner), a READ-ONLY planning agent.
+
+**CRITICAL CONSTRAINTS:**
+- DO NOT modify any files (no Write, Edit, or any file mutations)
+- DO NOT execute commands that change system state
+- DO NOT create, delete, or rename files
+- ONLY provide analysis, recommendations, and information
+
+**YOUR ROLE**: Provide consultation, research, and analysis to assist with planning.
+Return your findings and recommendations. The actual implementation will be handled separately after planning is complete.
+
+---
+
+`

src/hooks/prometheus-md-only/index.test.ts

@@ -0,0 +1,298 @@
+import { describe, expect, test, beforeEach, afterEach, mock } from "bun:test"
+import { mkdirSync, rmSync, writeFileSync } from "node:fs"
+import { join } from "node:path"
+import { createPrometheusMdOnlyHook } from "./index"
+import { MESSAGE_STORAGE } from "../../features/hook-message-injector"
+
+describe("prometheus-md-only", () => {
+  const TEST_SESSION_ID = "test-session-prometheus"
+  let testMessageDir: string
+
+  function createMockPluginInput() {
+    return {
+      client: {},
+      directory: "/tmp/test",
+    } as never
+  }
+
+  function setupMessageStorage(sessionID: string, agent: string): void {
+    testMessageDir = join(MESSAGE_STORAGE, sessionID)
+    mkdirSync(testMessageDir, { recursive: true })
+    const messageContent = {
+      agent,
+      model: { providerID: "test", modelID: "test-model" },
+    }
+    writeFileSync(
+      join(testMessageDir, "msg_001.json"),
+      JSON.stringify(messageContent)
+    )
+  }
+
+  afterEach(() => {
+    if (testMessageDir) {
+      try {
+        rmSync(testMessageDir, { recursive: true, force: true })
+      } catch {
+        // ignore
+      }
+    }
+  })
+
+  describe("with Prometheus agent in message storage", () => {
+    beforeEach(() => {
+      setupMessageStorage(TEST_SESSION_ID, "Prometheus (Planner)")
+    })
+
+    test("should block Prometheus from writing non-.md files", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "Write",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const output = {
+        args: { filePath: "/path/to/file.ts" },
+      }
+
+      // #when / #then
+      await expect(
+        hook["tool.execute.before"](input, output)
+      ).rejects.toThrow("can only write/edit .md files")
+    })
+
+    test("should allow Prometheus to write .md files inside .sisyphus/", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "Write",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const output = {
+        args: { filePath: "/project/.sisyphus/plans/work-plan.md" },
+      }
+
+      // #when / #then
+      await expect(
+        hook["tool.execute.before"](input, output)
+      ).resolves.toBeUndefined()
+    })
+
+    test("should block Prometheus from writing .md files outside .sisyphus/", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "Write",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const output = {
+        args: { filePath: "/path/to/README.md" },
+      }
+
+      // #when / #then
+      await expect(
+        hook["tool.execute.before"](input, output)
+      ).rejects.toThrow("can only write/edit .md files inside .sisyphus/")
+    })
+
+    test("should block Edit tool for non-.md files", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "Edit",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const output = {
+        args: { filePath: "/path/to/code.py" },
+      }
+
+      // #when / #then
+      await expect(
+        hook["tool.execute.before"](input, output)
+      ).rejects.toThrow("can only write/edit .md files")
+    })
+
+    test("should not affect non-Write/Edit tools", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "Read",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const output = {
+        args: { filePath: "/path/to/file.ts" },
+      }
+
+      // #when / #then
+      await expect(
+        hook["tool.execute.before"](input, output)
+      ).resolves.toBeUndefined()
+    })
+
+    test("should handle missing filePath gracefully", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "Write",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const output = {
+        args: {},
+      }
+
+      // #when / #then
+      await expect(
+        hook["tool.execute.before"](input, output)
+      ).resolves.toBeUndefined()
+    })
+
+    test("should inject read-only warning when Prometheus calls sisyphus_task", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "sisyphus_task",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const output = {
+        args: { prompt: "Analyze this codebase" },
+      }
+
+      // #when
+      await hook["tool.execute.before"](input, output)
+
+      // #then
+      expect(output.args.prompt).toContain("[SYSTEM DIRECTIVE - READ-ONLY PLANNING CONSULTATION]")
+      expect(output.args.prompt).toContain("DO NOT modify any files")
+    })
+
+    test("should inject read-only warning when Prometheus calls task", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "task",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const output = {
+        args: { prompt: "Research this library" },
+      }
+
+      // #when
+      await hook["tool.execute.before"](input, output)
+
+      // #then
+      expect(output.args.prompt).toContain("[SYSTEM DIRECTIVE - READ-ONLY PLANNING CONSULTATION]")
+    })
+
+    test("should inject read-only warning when Prometheus calls call_omo_agent", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "call_omo_agent",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const output = {
+        args: { prompt: "Find implementation examples" },
+      }
+
+      // #when
+      await hook["tool.execute.before"](input, output)
+
+      // #then
+      expect(output.args.prompt).toContain("[SYSTEM DIRECTIVE - READ-ONLY PLANNING CONSULTATION]")
+    })
+
+    test("should not double-inject warning if already present", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "sisyphus_task",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const promptWithWarning = "Some prompt [SYSTEM DIRECTIVE - READ-ONLY PLANNING CONSULTATION] already here"
+      const output = {
+        args: { prompt: promptWithWarning },
+      }
+
+      // #when
+      await hook["tool.execute.before"](input, output)
+
+      // #then
+      const occurrences = (output.args.prompt as string).split("[SYSTEM DIRECTIVE - READ-ONLY PLANNING CONSULTATION]").length - 1
+      expect(occurrences).toBe(1)
+    })
+  })
+
+  describe("with non-Prometheus agent in message storage", () => {
+    beforeEach(() => {
+      setupMessageStorage(TEST_SESSION_ID, "Sisyphus")
+    })
+
+    test("should not affect non-Prometheus agents", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "Write",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const output = {
+        args: { filePath: "/path/to/file.ts" },
+      }
+
+      // #when / #then
+      await expect(
+        hook["tool.execute.before"](input, output)
+      ).resolves.toBeUndefined()
+    })
+
+    test("should not inject warning for non-Prometheus agents calling sisyphus_task", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "sisyphus_task",
+        sessionID: TEST_SESSION_ID,
+        callID: "call-1",
+      }
+      const originalPrompt = "Implement this feature"
+      const output = {
+        args: { prompt: originalPrompt },
+      }
+
+      // #when
+      await hook["tool.execute.before"](input, output)
+
+      // #then
+      expect(output.args.prompt).toBe(originalPrompt)
+      expect(output.args.prompt).not.toContain("[SYSTEM DIRECTIVE - READ-ONLY PLANNING CONSULTATION]")
+    })
+  })
+
+  describe("without message storage", () => {
+    test("should handle missing session gracefully (no agent found)", async () => {
+      // #given
+      const hook = createPrometheusMdOnlyHook(createMockPluginInput())
+      const input = {
+        tool: "Write",
+        sessionID: "non-existent-session",
+        callID: "call-1",
+      }
+      const output = {
+        args: { filePath: "/path/to/file.ts" },
+      }
+
+      // #when / #then
+      await expect(
+        hook["tool.execute.before"](input, output)
+      ).resolves.toBeUndefined()
+    })
+  })
+})

src/hooks/prometheus-md-only/index.ts

@@ -0,0 +1,97 @@
+import type { PluginInput } from "@opencode-ai/plugin"
+import { existsSync, readdirSync } from "node:fs"
+import { join } from "node:path"
+import { HOOK_NAME, PROMETHEUS_AGENTS, ALLOWED_EXTENSIONS, ALLOWED_PATH_PREFIX, BLOCKED_TOOLS, PLANNING_CONSULT_WARNING } from "./constants"
+import { findNearestMessageWithFields, MESSAGE_STORAGE } from "../../features/hook-message-injector"
+import { log } from "../../shared/logger"
+
+export * from "./constants"
+
+function isAllowedFile(filePath: string): boolean {
+  const hasAllowedExtension = ALLOWED_EXTENSIONS.some(ext => filePath.endsWith(ext))
+  const isInAllowedPath = filePath.includes(ALLOWED_PATH_PREFIX)
+  return hasAllowedExtension && isInAllowedPath
+}
+
+function getMessageDir(sessionID: string): string | null {
+  if (!existsSync(MESSAGE_STORAGE)) return null
+
+  const directPath = join(MESSAGE_STORAGE, sessionID)
+  if (existsSync(directPath)) return directPath
+
+  for (const dir of readdirSync(MESSAGE_STORAGE)) {
+    const sessionPath = join(MESSAGE_STORAGE, dir, sessionID)
+    if (existsSync(sessionPath)) return sessionPath
+  }
+
+  return null
+}
+
+const TASK_TOOLS = ["sisyphus_task", "task", "call_omo_agent"]
+
+function getAgentFromSession(sessionID: string): string | undefined {
+  const messageDir = getMessageDir(sessionID)
+  if (!messageDir) return undefined
+  return findNearestMessageWithFields(messageDir)?.agent
+}
+
+export function createPrometheusMdOnlyHook(_ctx: PluginInput) {
+  return {
+    "tool.execute.before": async (
+      input: { tool: string; sessionID: string; callID: string },
+      output: { args: Record<string, unknown>; message?: string }
+    ): Promise<void> => {
+      const agentName = getAgentFromSession(input.sessionID)
+      
+      if (!agentName || !PROMETHEUS_AGENTS.includes(agentName)) {
+        return
+      }
+
+      const toolName = input.tool
+
+      // Inject read-only warning for task tools called by Prometheus
+      if (TASK_TOOLS.includes(toolName)) {
+        const prompt = output.args.prompt as string | undefined
+        if (prompt && !prompt.includes("[SYSTEM DIRECTIVE - READ-ONLY PLANNING CONSULTATION]")) {
+          output.args.prompt = prompt + PLANNING_CONSULT_WARNING
+          log(`[${HOOK_NAME}] Injected read-only planning warning to ${toolName}`, {
+            sessionID: input.sessionID,
+            tool: toolName,
+            agent: agentName,
+          })
+        }
+        return
+      }
+
+      if (!BLOCKED_TOOLS.includes(toolName)) {
+        return
+      }
+
+      const filePath = (output.args.filePath ?? output.args.path ?? output.args.file) as string | undefined
+      if (!filePath) {
+        return
+      }
+
+      if (!isAllowedFile(filePath)) {
+        log(`[${HOOK_NAME}] Blocked: Prometheus can only write to .sisyphus/*.md`, {
+          sessionID: input.sessionID,
+          tool: toolName,
+          filePath,
+          agent: agentName,
+        })
+        throw new Error(
+          `[${HOOK_NAME}] Prometheus (Planner) can only write/edit .md files inside .sisyphus/ directory. ` +
+          `Attempted to modify: ${filePath}. ` +
+          `Prometheus is a READ-ONLY planner. Use /start-work to execute the plan.`
+        )
+      }
+
+      log(`[${HOOK_NAME}] Allowed: .sisyphus/*.md write permitted`, {
+        sessionID: input.sessionID,
+        tool: toolName,
+        filePath,
+        agent: agentName,
+      })
+    },
+  }
+}

src/hooks/ralph-loop/index.test.ts

@@ -591,6 +591,73 @@ describe("ralph-loop", () => {
       expect(hook.getState()).toBeNull()
     })
 
+    test("should allow starting new loop while previous loop is active (different session)", async () => {
+      // #given - active loop in session A
+      const hook = createRalphLoopHook(createMockPluginInput())
+      hook.startLoop("session-A", "First task", { maxIterations: 10 })
+      expect(hook.getState()?.session_id).toBe("session-A")
+      expect(hook.getState()?.prompt).toBe("First task")
+
+      // #when - start new loop in session B (without completing A)
+      hook.startLoop("session-B", "Second task", { maxIterations: 20 })
+
+      // #then - state should be overwritten with session B's loop
+      expect(hook.getState()?.session_id).toBe("session-B")
+      expect(hook.getState()?.prompt).toBe("Second task")
+      expect(hook.getState()?.max_iterations).toBe(20)
+      expect(hook.getState()?.iteration).toBe(1)
+
+      // #when - session B goes idle
+      await hook.event({
+        event: { type: "session.idle", properties: { sessionID: "session-B" } },
+      })
+
+      // #then - continuation should be injected for session B
+      expect(promptCalls.length).toBe(1)
+      expect(promptCalls[0].sessionID).toBe("session-B")
+      expect(promptCalls[0].text).toContain("Second task")
+      expect(promptCalls[0].text).toContain("2/20")
+
+      // #then - iteration incremented
+      expect(hook.getState()?.iteration).toBe(2)
+    })
+
+    test("should allow starting new loop in same session (restart)", async () => {
+      // #given - active loop in session A at iteration 5
+      const hook = createRalphLoopHook(createMockPluginInput())
+      hook.startLoop("session-A", "First task", { maxIterations: 10 })
+      
+      // Simulate some iterations
+      await hook.event({
+        event: { type: "session.idle", properties: { sessionID: "session-A" } },
+      })
+      await hook.event({
+        event: { type: "session.idle", properties: { sessionID: "session-A" } },
+      })
+      expect(hook.getState()?.iteration).toBe(3)
+      expect(promptCalls.length).toBe(2)
+
+      // #when - start NEW loop in same session (restart)
+      hook.startLoop("session-A", "Restarted task", { maxIterations: 50 })
+
+      // #then - state should be reset to iteration 1 with new prompt
+      expect(hook.getState()?.session_id).toBe("session-A")
+      expect(hook.getState()?.prompt).toBe("Restarted task")
+      expect(hook.getState()?.max_iterations).toBe(50)
+      expect(hook.getState()?.iteration).toBe(1)
+
+      // #when - session goes idle
+      promptCalls = [] // Reset to check new continuation
+      await hook.event({
+        event: { type: "session.idle", properties: { sessionID: "session-A" } },
+      })
+
+      // #then - continuation should use new task
+      expect(promptCalls.length).toBe(1)
+      expect(promptCalls[0].text).toContain("Restarted task")
+      expect(promptCalls[0].text).toContain("2/50")
+    })
+
     test("should check transcript BEFORE API to optimize performance", async () => {
       // #given - transcript has completion promise
       const transcriptPath = join(TEST_DIR, "transcript.jsonl")

src/hooks/sisyphus-orchestrator/index.test.ts

@@ -0,0 +1,829 @@
+import { describe, expect, test, beforeEach, afterEach, mock } from "bun:test"
+import { existsSync, mkdirSync, rmSync, writeFileSync } from "node:fs"
+import { join } from "node:path"
+import { tmpdir } from "node:os"
+import { createSisyphusOrchestratorHook } from "./index"
+import {
+  writeBoulderState,
+  clearBoulderState,
+  readBoulderState,
+} from "../../features/boulder-state"
+import type { BoulderState } from "../../features/boulder-state"
+
+import { MESSAGE_STORAGE } from "../../features/hook-message-injector"
+
+describe("sisyphus-orchestrator hook", () => {
+  const TEST_DIR = join(tmpdir(), "sisyphus-orchestrator-test-" + Date.now())
+  const SISYPHUS_DIR = join(TEST_DIR, ".sisyphus")
+
+  function createMockPluginInput(overrides?: { promptMock?: ReturnType<typeof mock> }) {
+    const promptMock = overrides?.promptMock ?? mock(() => Promise.resolve())
+    return {
+      directory: TEST_DIR,
+      client: {
+        session: {
+          prompt: promptMock,
+        },
+      },
+      _promptMock: promptMock,
+    } as unknown as Parameters<typeof createSisyphusOrchestratorHook>[0] & { _promptMock: ReturnType<typeof mock> }
+  }
+
+  function setupMessageStorage(sessionID: string, agent: string): void {
+    const messageDir = join(MESSAGE_STORAGE, sessionID)
+    if (!existsSync(messageDir)) {
+      mkdirSync(messageDir, { recursive: true })
+    }
+    const messageData = {
+      agent,
+      model: { providerID: "anthropic", modelID: "claude-opus-4-5" },
+    }
+    writeFileSync(join(messageDir, "msg_test001.json"), JSON.stringify(messageData))
+  }
+
+  function cleanupMessageStorage(sessionID: string): void {
+    const messageDir = join(MESSAGE_STORAGE, sessionID)
+    if (existsSync(messageDir)) {
+      rmSync(messageDir, { recursive: true, force: true })
+    }
+  }
+
+  beforeEach(() => {
+    if (!existsSync(TEST_DIR)) {
+      mkdirSync(TEST_DIR, { recursive: true })
+    }
+    if (!existsSync(SISYPHUS_DIR)) {
+      mkdirSync(SISYPHUS_DIR, { recursive: true })
+    }
+    clearBoulderState(TEST_DIR)
+  })
+
+  afterEach(() => {
+    clearBoulderState(TEST_DIR)
+    if (existsSync(TEST_DIR)) {
+      rmSync(TEST_DIR, { recursive: true, force: true })
+    }
+  })
+
+  describe("tool.execute.after handler", () => {
+    test("should ignore non-sisyphus_task tools", async () => {
+      // #given - hook and non-sisyphus_task tool
+      const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+      const output = {
+        title: "Test Tool",
+        output: "Original output",
+        metadata: {},
+      }
+
+      // #when
+      await hook["tool.execute.after"](
+        { tool: "other_tool", sessionID: "session-123" },
+        output
+      )
+
+      // #then - output unchanged
+      expect(output.output).toBe("Original output")
+    })
+
+    test("should not transform when caller is not orchestrator-sisyphus", async () => {
+      // #given - boulder state exists but caller agent in message storage is not orchestrator
+      const sessionID = "session-non-orchestrator-test"
+      setupMessageStorage(sessionID, "other-agent")
+      
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1"],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+      const output = {
+        title: "Sisyphus Task",
+        output: "Task completed successfully",
+        metadata: {},
+      }
+
+      // #when
+      await hook["tool.execute.after"](
+        { tool: "sisyphus_task", sessionID },
+        output
+      )
+
+      // #then - output unchanged because caller is not orchestrator
+      expect(output.output).toBe("Task completed successfully")
+      
+      cleanupMessageStorage(sessionID)
+    })
+
+    test("should append standalone verification when no boulder state but caller is orchestrator", async () => {
+      // #given - no boulder state, but caller is orchestrator
+      const sessionID = "session-no-boulder-test"
+      setupMessageStorage(sessionID, "orchestrator-sisyphus")
+      
+      const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+      const output = {
+        title: "Sisyphus Task",
+        output: "Task completed successfully",
+        metadata: {},
+      }
+
+      // #when
+      await hook["tool.execute.after"](
+        { tool: "sisyphus_task", sessionID },
+        output
+      )
+
+      // #then - standalone verification reminder appended
+      expect(output.output).toContain("Task completed successfully")
+      expect(output.output).toContain("MANDATORY VERIFICATION")
+      expect(output.output).toContain("sisyphus_task(resume=")
+      
+      cleanupMessageStorage(sessionID)
+    })
+
+    test("should transform output when caller is orchestrator-sisyphus with boulder state", async () => {
+      // #given - orchestrator-sisyphus caller with boulder state
+      const sessionID = "session-transform-test"
+      setupMessageStorage(sessionID, "orchestrator-sisyphus")
+      
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1\n- [x] Task 2")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1"],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+      const output = {
+        title: "Sisyphus Task",
+        output: "Task completed successfully",
+        metadata: {},
+      }
+
+      // #when
+      await hook["tool.execute.after"](
+        { tool: "sisyphus_task", sessionID },
+        output
+      )
+
+      // #then - output should be transformed (original output replaced)
+      expect(output.output).not.toContain("Task completed successfully")
+      expect(output.output).toContain("SUBAGENT WORK COMPLETED")
+      expect(output.output).toContain("test-plan")
+      expect(output.output).toContain("SUBAGENTS LIE")
+      expect(output.output).toContain("sisyphus_task(resume=")
+      
+      cleanupMessageStorage(sessionID)
+    })
+
+    test("should still transform when plan is complete (shows progress)", async () => {
+      // #given - boulder state with complete plan, orchestrator caller
+      const sessionID = "session-complete-plan-test"
+      setupMessageStorage(sessionID, "orchestrator-sisyphus")
+      
+      const planPath = join(TEST_DIR, "complete-plan.md")
+      writeFileSync(planPath, "# Plan\n- [x] Task 1\n- [x] Task 2")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1"],
+        plan_name: "complete-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+      const output = {
+        title: "Sisyphus Task",
+        output: "Original output",
+        metadata: {},
+      }
+
+      // #when
+      await hook["tool.execute.after"](
+        { tool: "sisyphus_task", sessionID },
+        output
+      )
+
+      // #then - output transformed even when complete (shows 2/2 done)
+      expect(output.output).toContain("SUBAGENT WORK COMPLETED")
+      expect(output.output).toContain("2/2 done")
+      expect(output.output).toContain("0 left")
+      
+      cleanupMessageStorage(sessionID)
+    })
+
+    test("should append session ID to boulder state if not present", async () => {
+      // #given - boulder state without session-append-test, orchestrator caller
+      const sessionID = "session-append-test"
+      setupMessageStorage(sessionID, "orchestrator-sisyphus")
+      
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1"],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+      const output = {
+        title: "Sisyphus Task",
+        output: "Task output",
+        metadata: {},
+      }
+
+      // #when
+      await hook["tool.execute.after"](
+        { tool: "sisyphus_task", sessionID },
+        output
+      )
+
+      // #then - sessionID should be appended
+      const updatedState = readBoulderState(TEST_DIR)
+      expect(updatedState?.session_ids).toContain(sessionID)
+      
+      cleanupMessageStorage(sessionID)
+    })
+
+    test("should not duplicate existing session ID", async () => {
+      // #given - boulder state already has session-dup-test, orchestrator caller
+      const sessionID = "session-dup-test"
+      setupMessageStorage(sessionID, "orchestrator-sisyphus")
+      
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: [sessionID],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+      const output = {
+        title: "Sisyphus Task",
+        output: "Task output",
+        metadata: {},
+      }
+
+      // #when
+      await hook["tool.execute.after"](
+        { tool: "sisyphus_task", sessionID },
+        output
+      )
+
+      // #then - should still have only one sessionID
+      const updatedState = readBoulderState(TEST_DIR)
+      const count = updatedState?.session_ids.filter((id) => id === sessionID).length
+      expect(count).toBe(1)
+      
+      cleanupMessageStorage(sessionID)
+    })
+
+    test("should include boulder.json path and notepad path in transformed output", async () => {
+      // #given - boulder state, orchestrator caller
+      const sessionID = "session-path-test"
+      setupMessageStorage(sessionID, "orchestrator-sisyphus")
+      
+      const planPath = join(TEST_DIR, "my-feature.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1\n- [ ] Task 2\n- [x] Task 3")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1"],
+        plan_name: "my-feature",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+      const output = {
+        title: "Sisyphus Task",
+        output: "Task completed",
+        metadata: {},
+      }
+
+      // #when
+      await hook["tool.execute.after"](
+        { tool: "sisyphus_task", sessionID },
+        output
+      )
+
+      // #then - output should contain plan name and progress
+      expect(output.output).toContain("my-feature")
+      expect(output.output).toContain("1/3 done")
+      expect(output.output).toContain("2 left")
+      
+      cleanupMessageStorage(sessionID)
+    })
+
+    test("should include resume and checkbox instructions in reminder", async () => {
+      // #given - boulder state, orchestrator caller
+      const sessionID = "session-resume-test"
+      setupMessageStorage(sessionID, "orchestrator-sisyphus")
+      
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1"],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+      const output = {
+        title: "Sisyphus Task",
+        output: "Task completed",
+        metadata: {},
+      }
+
+      // #when
+      await hook["tool.execute.after"](
+        { tool: "sisyphus_task", sessionID },
+        output
+      )
+
+      // #then - should include resume instructions and verification
+      expect(output.output).toContain("sisyphus_task(resume=")
+      expect(output.output).toContain("[x]")
+      expect(output.output).toContain("MANDATORY VERIFICATION")
+      
+      cleanupMessageStorage(sessionID)
+    })
+
+    describe("Write/Edit tool direct work reminder", () => {
+      const ORCHESTRATOR_SESSION = "orchestrator-write-test"
+
+      beforeEach(() => {
+        setupMessageStorage(ORCHESTRATOR_SESSION, "orchestrator-sisyphus")
+      })
+
+      afterEach(() => {
+        cleanupMessageStorage(ORCHESTRATOR_SESSION)
+      })
+
+      test("should append delegation reminder when orchestrator writes outside .sisyphus/", async () => {
+        // #given
+        const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+        const output = {
+          title: "Write",
+          output: "File written successfully",
+          metadata: { filePath: "/path/to/code.ts" },
+        }
+
+        // #when
+        await hook["tool.execute.after"](
+          { tool: "Write", sessionID: ORCHESTRATOR_SESSION },
+          output
+        )
+
+        // #then
+        expect(output.output).toContain("DELEGATION REQUIRED")
+        expect(output.output).toContain("ORCHESTRATOR, not an IMPLEMENTER")
+        expect(output.output).toContain("sisyphus_task")
+      })
+
+      test("should append delegation reminder when orchestrator edits outside .sisyphus/", async () => {
+        // #given
+        const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+        const output = {
+          title: "Edit",
+          output: "File edited successfully",
+          metadata: { filePath: "/src/components/button.tsx" },
+        }
+
+        // #when
+        await hook["tool.execute.after"](
+          { tool: "Edit", sessionID: ORCHESTRATOR_SESSION },
+          output
+        )
+
+        // #then
+        expect(output.output).toContain("DELEGATION REQUIRED")
+      })
+
+      test("should NOT append reminder when orchestrator writes inside .sisyphus/", async () => {
+        // #given
+        const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+        const originalOutput = "File written successfully"
+        const output = {
+          title: "Write",
+          output: originalOutput,
+          metadata: { filePath: "/project/.sisyphus/plans/work-plan.md" },
+        }
+
+        // #when
+        await hook["tool.execute.after"](
+          { tool: "Write", sessionID: ORCHESTRATOR_SESSION },
+          output
+        )
+
+        // #then
+        expect(output.output).toBe(originalOutput)
+        expect(output.output).not.toContain("DELEGATION REQUIRED")
+      })
+
+      test("should NOT append reminder when non-orchestrator writes outside .sisyphus/", async () => {
+        // #given
+        const nonOrchestratorSession = "non-orchestrator-session"
+        setupMessageStorage(nonOrchestratorSession, "Sisyphus-Junior")
+        
+        const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+        const originalOutput = "File written successfully"
+        const output = {
+          title: "Write",
+          output: originalOutput,
+          metadata: { filePath: "/path/to/code.ts" },
+        }
+
+        // #when
+        await hook["tool.execute.after"](
+          { tool: "Write", sessionID: nonOrchestratorSession },
+          output
+        )
+
+        // #then
+        expect(output.output).toBe(originalOutput)
+        expect(output.output).not.toContain("DELEGATION REQUIRED")
+        
+        cleanupMessageStorage(nonOrchestratorSession)
+      })
+
+      test("should NOT append reminder for read-only tools", async () => {
+        // #given
+        const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+        const originalOutput = "File content"
+        const output = {
+          title: "Read",
+          output: originalOutput,
+          metadata: { filePath: "/path/to/code.ts" },
+        }
+
+        // #when
+        await hook["tool.execute.after"](
+          { tool: "Read", sessionID: ORCHESTRATOR_SESSION },
+          output
+        )
+
+        // #then
+        expect(output.output).toBe(originalOutput)
+      })
+
+      test("should handle missing filePath gracefully", async () => {
+        // #given
+        const hook = createSisyphusOrchestratorHook(createMockPluginInput())
+        const originalOutput = "File written successfully"
+        const output = {
+          title: "Write",
+          output: originalOutput,
+          metadata: {},
+        }
+
+        // #when
+        await hook["tool.execute.after"](
+          { tool: "Write", sessionID: ORCHESTRATOR_SESSION },
+          output
+        )
+
+        // #then
+        expect(output.output).toBe(originalOutput)
+      })
+    })
+  })
+
+  describe("session.idle handler (boulder continuation)", () => {
+    const MAIN_SESSION_ID = "main-session-123"
+
+    beforeEach(() => {
+      mock.module("../../features/claude-code-session-state", () => ({
+        getMainSessionID: () => MAIN_SESSION_ID,
+        subagentSessions: new Set<string>(),
+      }))
+      setupMessageStorage(MAIN_SESSION_ID, "orchestrator-sisyphus")
+    })
+
+    afterEach(() => {
+      cleanupMessageStorage(MAIN_SESSION_ID)
+    })
+
+    test("should inject continuation when boulder has incomplete tasks", async () => {
+      // #given - boulder state with incomplete plan
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1\n- [x] Task 2\n- [ ] Task 3")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: [MAIN_SESSION_ID],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const mockInput = createMockPluginInput()
+      const hook = createSisyphusOrchestratorHook(mockInput)
+
+      // #when
+      await hook.handler({
+        event: {
+          type: "session.idle",
+          properties: { sessionID: MAIN_SESSION_ID },
+        },
+      })
+
+      // #then - should call prompt with continuation
+      expect(mockInput._promptMock).toHaveBeenCalled()
+      const callArgs = mockInput._promptMock.mock.calls[0][0]
+      expect(callArgs.path.id).toBe(MAIN_SESSION_ID)
+      expect(callArgs.body.parts[0].text).toContain("BOULDER CONTINUATION")
+      expect(callArgs.body.parts[0].text).toContain("2 remaining")
+    })
+
+    test("should not inject when no boulder state exists", async () => {
+      // #given - no boulder state
+      const mockInput = createMockPluginInput()
+      const hook = createSisyphusOrchestratorHook(mockInput)
+
+      // #when
+      await hook.handler({
+        event: {
+          type: "session.idle",
+          properties: { sessionID: MAIN_SESSION_ID },
+        },
+      })
+
+      // #then - should not call prompt
+      expect(mockInput._promptMock).not.toHaveBeenCalled()
+    })
+
+    test("should not inject when boulder plan is complete", async () => {
+      // #given - boulder state with complete plan
+      const planPath = join(TEST_DIR, "complete-plan.md")
+      writeFileSync(planPath, "# Plan\n- [x] Task 1\n- [x] Task 2")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: [MAIN_SESSION_ID],
+        plan_name: "complete-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const mockInput = createMockPluginInput()
+      const hook = createSisyphusOrchestratorHook(mockInput)
+
+      // #when
+      await hook.handler({
+        event: {
+          type: "session.idle",
+          properties: { sessionID: MAIN_SESSION_ID },
+        },
+      })
+
+      // #then - should not call prompt
+      expect(mockInput._promptMock).not.toHaveBeenCalled()
+    })
+
+    test("should skip when abort error occurred before idle", async () => {
+      // #given - boulder state with incomplete plan
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: [MAIN_SESSION_ID],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const mockInput = createMockPluginInput()
+      const hook = createSisyphusOrchestratorHook(mockInput)
+
+      // #when - send abort error then idle
+      await hook.handler({
+        event: {
+          type: "session.error",
+          properties: {
+            sessionID: MAIN_SESSION_ID,
+            error: { name: "AbortError", message: "aborted" },
+          },
+        },
+      })
+      await hook.handler({
+        event: {
+          type: "session.idle",
+          properties: { sessionID: MAIN_SESSION_ID },
+        },
+      })
+
+      // #then - should not call prompt
+      expect(mockInput._promptMock).not.toHaveBeenCalled()
+    })
+
+    test("should skip when background tasks are running", async () => {
+      // #given - boulder state with incomplete plan
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: [MAIN_SESSION_ID],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const mockBackgroundManager = {
+        getTasksByParentSession: () => [{ status: "running" }],
+      }
+
+      const mockInput = createMockPluginInput()
+      const hook = createSisyphusOrchestratorHook(mockInput, {
+        directory: TEST_DIR,
+        backgroundManager: mockBackgroundManager as any,
+      })
+
+      // #when
+      await hook.handler({
+        event: {
+          type: "session.idle",
+          properties: { sessionID: MAIN_SESSION_ID },
+        },
+      })
+
+      // #then - should not call prompt
+      expect(mockInput._promptMock).not.toHaveBeenCalled()
+    })
+
+    test("should clear abort state on message.updated", async () => {
+      // #given - boulder with incomplete plan
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: [MAIN_SESSION_ID],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const mockInput = createMockPluginInput()
+      const hook = createSisyphusOrchestratorHook(mockInput)
+
+      // #when - abort error, then message update, then idle
+      await hook.handler({
+        event: {
+          type: "session.error",
+          properties: {
+            sessionID: MAIN_SESSION_ID,
+            error: { name: "AbortError" },
+          },
+        },
+      })
+      await hook.handler({
+        event: {
+          type: "message.updated",
+          properties: { info: { sessionID: MAIN_SESSION_ID, role: "user" } },
+        },
+      })
+      await hook.handler({
+        event: {
+          type: "session.idle",
+          properties: { sessionID: MAIN_SESSION_ID },
+        },
+      })
+
+      // #then - should call prompt because abort state was cleared
+      expect(mockInput._promptMock).toHaveBeenCalled()
+    })
+
+    test("should include plan progress in continuation prompt", async () => {
+      // #given - boulder state with specific progress
+      const planPath = join(TEST_DIR, "progress-plan.md")
+      writeFileSync(planPath, "# Plan\n- [x] Task 1\n- [x] Task 2\n- [ ] Task 3\n- [ ] Task 4")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: [MAIN_SESSION_ID],
+        plan_name: "progress-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const mockInput = createMockPluginInput()
+      const hook = createSisyphusOrchestratorHook(mockInput)
+
+      // #when
+      await hook.handler({
+        event: {
+          type: "session.idle",
+          properties: { sessionID: MAIN_SESSION_ID },
+        },
+      })
+
+      // #then - should include progress
+      const callArgs = mockInput._promptMock.mock.calls[0][0]
+      expect(callArgs.body.parts[0].text).toContain("2/4 completed")
+      expect(callArgs.body.parts[0].text).toContain("2 remaining")
+    })
+
+    test("should not inject when last agent is not orchestrator-sisyphus", async () => {
+      // #given - boulder state with incomplete plan, but last agent is NOT orchestrator-sisyphus
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1\n- [ ] Task 2")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: [MAIN_SESSION_ID],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      // #given - last agent is NOT orchestrator-sisyphus
+      cleanupMessageStorage(MAIN_SESSION_ID)
+      setupMessageStorage(MAIN_SESSION_ID, "Sisyphus")
+
+      const mockInput = createMockPluginInput()
+      const hook = createSisyphusOrchestratorHook(mockInput)
+
+      // #when
+      await hook.handler({
+        event: {
+          type: "session.idle",
+          properties: { sessionID: MAIN_SESSION_ID },
+        },
+      })
+
+      // #then - should NOT call prompt because agent is not orchestrator-sisyphus
+      expect(mockInput._promptMock).not.toHaveBeenCalled()
+    })
+
+    test("should cleanup on session.deleted", async () => {
+      // #given - boulder state
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: [MAIN_SESSION_ID],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const mockInput = createMockPluginInput()
+      const hook = createSisyphusOrchestratorHook(mockInput)
+
+      // #when - create abort state then delete
+      await hook.handler({
+        event: {
+          type: "session.error",
+          properties: {
+            sessionID: MAIN_SESSION_ID,
+            error: { name: "AbortError" },
+          },
+        },
+      })
+      await hook.handler({
+        event: {
+          type: "session.deleted",
+          properties: { info: { id: MAIN_SESSION_ID } },
+        },
+      })
+
+      // Re-create boulder after deletion
+      writeBoulderState(TEST_DIR, state)
+
+      // Trigger idle - should inject because state was cleaned up
+      await hook.handler({
+        event: {
+          type: "session.idle",
+          properties: { sessionID: MAIN_SESSION_ID },
+        },
+      })
+
+      // #then - should call prompt because session state was cleaned
+      expect(mockInput._promptMock).toHaveBeenCalled()
+    })
+  })
+})

src/hooks/sisyphus-orchestrator/index.ts

@@ -0,0 +1,660 @@
+import type { PluginInput } from "@opencode-ai/plugin"
+import { execSync } from "node:child_process"
+import { existsSync, readdirSync } from "node:fs"
+import { join } from "node:path"
+import {
+  readBoulderState,
+  appendSessionId,
+  getPlanProgress,
+} from "../../features/boulder-state"
+import { getMainSessionID, subagentSessions } from "../../features/claude-code-session-state"
+import { findNearestMessageWithFields, MESSAGE_STORAGE } from "../../features/hook-message-injector"
+import { log } from "../../shared/logger"
+import type { BackgroundManager } from "../../features/background-agent"
+
+export const HOOK_NAME = "sisyphus-orchestrator"
+
+const ALLOWED_PATH_PREFIX = ".sisyphus/"
+const WRITE_EDIT_TOOLS = ["Write", "Edit", "write", "edit"]
+
+const DIRECT_WORK_REMINDER = `
+
+---
+
+[SYSTEM REMINDER - DELEGATION REQUIRED]
+
+You just performed direct file modifications outside \`.sisyphus/\`.
+
+**You are an ORCHESTRATOR, not an IMPLEMENTER.**
+
+As an orchestrator, you should:
+- **DELEGATE** implementation work to subagents via \`sisyphus_task\`
+- **VERIFY** the work done by subagents
+- **COORDINATE** multiple tasks and ensure completion
+
+You should NOT:
+- Write code directly (except for \`.sisyphus/\` files like plans and notepads)
+- Make direct file edits outside \`.sisyphus/\`
+- Implement features yourself
+
+**If you need to make changes:**
+1. Use \`sisyphus_task\` to delegate to an appropriate subagent
+2. Provide clear instructions in the prompt
+3. Verify the subagent's work after completion
+
+---
+`
+
+const BOULDER_CONTINUATION_PROMPT = `[SYSTEM REMINDER - BOULDER CONTINUATION]
+
+You have an active work plan with incomplete tasks. Continue working.
+
+RULES:
+- Proceed without asking for permission
+- Mark each checkbox [x] in the plan file when done
+- Use the notepad at .sisyphus/notepads/{PLAN_NAME}/ to record learnings
+- Do not stop until all tasks are complete
+- If blocked, document the blocker and move to the next task`
+
+const VERIFICATION_REMINDER = `**MANDATORY VERIFICATION - SUBAGENTS LIE**
+
+Subagents FREQUENTLY claim completion when:
+- Tests are actually FAILING
+- Code has type/lint ERRORS
+- Implementation is INCOMPLETE
+- Patterns were NOT followed
+
+**YOU MUST VERIFY EVERYTHING YOURSELF:**
+
+1. Run \`lsp_diagnostics\` on changed files - Must be CLEAN
+2. Run tests yourself - Must PASS (not "agent said it passed")
+3. Read the actual code - Must match requirements
+4. Check build/typecheck - Must succeed
+
+DO NOT TRUST THE AGENT'S SELF-REPORT.
+VERIFY EACH CLAIM WITH YOUR OWN TOOL CALLS.
+
+**HANDS-ON QA REQUIRED (after ALL tasks complete):**
+
+| Deliverable Type | Verification Tool | Action |
+|------------------|-------------------|--------|
+| **Frontend/UI** | \`/playwright\` skill | Navigate, interact, screenshot evidence |
+| **TUI/CLI** | \`interactive_bash\` (tmux) | Run interactively, verify output |
+| **API/Backend** | \`bash\` with curl | Send requests, verify responses |
+
+Static analysis CANNOT catch: visual bugs, animation issues, user flow breakages, integration problems.
+**FAILURE TO DO HANDS-ON QA = INCOMPLETE WORK.**`
+
+const ORCHESTRATOR_DELEGATION_REQUIRED = `
+
+---
+
+⚠️⚠️⚠️ [CRITICAL SYSTEM DIRECTIVE - DELEGATION REQUIRED] ⚠️⚠️⚠️
+
+**STOP. YOU ARE VIOLATING ORCHESTRATOR PROTOCOL.**
+
+You (orchestrator-sisyphus) are attempting to directly modify a file outside \`.sisyphus/\`.
+
+**Path attempted:** $FILE_PATH
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+🚫 **THIS IS FORBIDDEN** (except for VERIFICATION purposes)
+
+As an ORCHESTRATOR, you MUST:
+1. **DELEGATE** all implementation work via \`sisyphus_task\`
+2. **VERIFY** the work done by subagents (reading files is OK)
+3. **COORDINATE** - you orchestrate, you don't implement
+
+**ALLOWED direct file operations:**
+- Files inside \`.sisyphus/\` (plans, notepads, drafts)
+- Reading files for verification
+- Running diagnostics/tests
+
+**FORBIDDEN direct file operations:**
+- Writing/editing source code
+- Creating new files outside \`.sisyphus/\`
+- Any implementation work
+
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+
+**IF THIS IS FOR VERIFICATION:**
+Proceed if you are verifying subagent work by making a small fix.
+But for any substantial changes, USE \`sisyphus_task\`.
+
+**CORRECT APPROACH:**
+\`\`\`
+sisyphus_task(
+  category="...",
+  prompt="[specific single task with clear acceptance criteria]"
+)
+\`\`\`
+
+⚠️⚠️⚠️ DELEGATE. DON'T IMPLEMENT. ⚠️⚠️⚠️
+
+---
+`
+
+const SINGLE_TASK_DIRECTIVE = `
+
+[SYSTEM DIRECTIVE - SINGLE TASK ONLY]
+
+**STOP. READ THIS BEFORE PROCEEDING.**
+
+If you were NOT given **exactly ONE atomic task**, you MUST:
+1. **IMMEDIATELY REFUSE** this request
+2. **DEMAND** the orchestrator provide a single, specific task
+
+**Your response if multiple tasks detected:**
+> "I refuse to proceed. You provided multiple tasks. An orchestrator's impatience destroys work quality.
+> 
+> PROVIDE EXACTLY ONE TASK. One file. One change. One verification.
+> 
+> Your rushing will cause: incomplete work, missed edge cases, broken tests, wasted context."
+
+**WARNING TO ORCHESTRATOR:**
+- Your hasty batching RUINS deliverables
+- Each task needs FULL attention and PROPER verification  
+- Batch delegation = sloppy work = rework = wasted tokens
+
+**REFUSE multi-task requests. DEMAND single-task clarity.**
+`
+
+function buildVerificationReminder(sessionId: string): string {
+  return `${VERIFICATION_REMINDER}
+
+---
+
+**If ANY verification fails, use this immediately:**
+\`\`\`
+sisyphus_task(resume="${sessionId}", prompt="fix: [describe the specific failure]")
+\`\`\``
+}
+
+function buildOrchestratorReminder(planName: string, progress: { total: number; completed: number }, sessionId: string): string {
+  const remaining = progress.total - progress.completed
+  return `
+---
+
+**State:** Plan: ${planName} | ${progress.completed}/${progress.total} done, ${remaining} left
+
+---
+
+${buildVerificationReminder(sessionId)}
+
+ALL pass? → commit atomic unit, mark \`[x]\`, next task.`
+}
+
+function buildStandaloneVerificationReminder(sessionId: string): string {
+  return `
+---
+
+${buildVerificationReminder(sessionId)}`
+}
+
+function extractSessionIdFromOutput(output: string): string {
+  const match = output.match(/Session ID:\s*(ses_[a-zA-Z0-9]+)/)
+  return match?.[1] ?? "<session_id>"
+}
+
+interface GitFileStat {
+  path: string
+  added: number
+  removed: number
+  status: "modified" | "added" | "deleted"
+}
+
+function getGitDiffStats(directory: string): GitFileStat[] {
+  try {
+    const output = execSync("git diff --numstat HEAD", {
+      cwd: directory,
+      encoding: "utf-8",
+      timeout: 5000,
+    }).trim()
+
+    if (!output) return []
+
+    const statusOutput = execSync("git status --porcelain", {
+      cwd: directory,
+      encoding: "utf-8",
+      timeout: 5000,
+    }).trim()
+
+    const statusMap = new Map<string, "modified" | "added" | "deleted">()
+    for (const line of statusOutput.split("\n")) {
+      if (!line) continue
+      const status = line.substring(0, 2).trim()
+      const filePath = line.substring(3)
+      if (status === "A" || status === "??") {
+        statusMap.set(filePath, "added")
+      } else if (status === "D") {
+        statusMap.set(filePath, "deleted")
+      } else {
+        statusMap.set(filePath, "modified")
+      }
+    }
+
+    const stats: GitFileStat[] = []
+    for (const line of output.split("\n")) {
+      const parts = line.split("\t")
+      if (parts.length < 3) continue
+
+      const [addedStr, removedStr, path] = parts
+      const added = addedStr === "-" ? 0 : parseInt(addedStr, 10)
+      const removed = removedStr === "-" ? 0 : parseInt(removedStr, 10)
+
+      stats.push({
+        path,
+        added,
+        removed,
+        status: statusMap.get(path) ?? "modified",
+      })
+    }
+
+    return stats
+  } catch {
+    return []
+  }
+}
+
+function formatFileChanges(stats: GitFileStat[], notepadPath?: string): string {
+  if (stats.length === 0) return "[FILE CHANGES SUMMARY]\nNo file changes detected.\n"
+
+  const modified = stats.filter((s) => s.status === "modified")
+  const added = stats.filter((s) => s.status === "added")
+  const deleted = stats.filter((s) => s.status === "deleted")
+
+  const lines: string[] = ["[FILE CHANGES SUMMARY]"]
+
+  if (modified.length > 0) {
+    lines.push("Modified files:")
+    for (const f of modified) {
+      lines.push(`  ${f.path}  (+${f.added}, -${f.removed})`)
+    }
+    lines.push("")
+  }
+
+  if (added.length > 0) {
+    lines.push("Created files:")
+    for (const f of added) {
+      lines.push(`  ${f.path}  (+${f.added})`)
+    }
+    lines.push("")
+  }
+
+  if (deleted.length > 0) {
+    lines.push("Deleted files:")
+    for (const f of deleted) {
+      lines.push(`  ${f.path}  (-${f.removed})`)
+    }
+    lines.push("")
+  }
+
+  if (notepadPath) {
+    const notepadStat = stats.find((s) => s.path.includes("notepad") || s.path.includes(".sisyphus"))
+    if (notepadStat) {
+      lines.push("[NOTEPAD UPDATED]")
+      lines.push(`  ${notepadStat.path}  (+${notepadStat.added})`)
+      lines.push("")
+    }
+  }
+
+  return lines.join("\n")
+}
+
+interface ToolExecuteAfterInput {
+  tool: string
+  sessionID?: string
+  callID?: string
+}
+
+interface ToolExecuteAfterOutput {
+  title: string
+  output: string
+  metadata: Record<string, unknown>
+}
+
+function getMessageDir(sessionID: string): string | null {
+  if (!existsSync(MESSAGE_STORAGE)) return null
+
+  const directPath = join(MESSAGE_STORAGE, sessionID)
+  if (existsSync(directPath)) return directPath
+
+  for (const dir of readdirSync(MESSAGE_STORAGE)) {
+    const sessionPath = join(MESSAGE_STORAGE, dir, sessionID)
+    if (existsSync(sessionPath)) return sessionPath
+  }
+
+  return null
+}
+
+function isCallerOrchestrator(sessionID?: string): boolean {
+  if (!sessionID) return false
+  const messageDir = getMessageDir(sessionID)
+  if (!messageDir) return false
+  const nearest = findNearestMessageWithFields(messageDir)
+  return nearest?.agent === "orchestrator-sisyphus"
+}
+
+interface SessionState {
+  lastEventWasAbortError?: boolean
+}
+
+export interface SisyphusOrchestratorHookOptions {
+  directory: string
+  backgroundManager?: BackgroundManager
+}
+
+function isAbortError(error: unknown): boolean {
+  if (!error) return false
+
+  if (typeof error === "object") {
+    const errObj = error as Record<string, unknown>
+    const name = errObj.name as string | undefined
+    const message = (errObj.message as string | undefined)?.toLowerCase() ?? ""
+
+    if (name === "MessageAbortedError" || name === "AbortError") return true
+    if (name === "DOMException" && message.includes("abort")) return true
+    if (message.includes("aborted") || message.includes("cancelled") || message.includes("interrupted")) return true
+  }
+
+  if (typeof error === "string") {
+    const lower = error.toLowerCase()
+    return lower.includes("abort") || lower.includes("cancel") || lower.includes("interrupt")
+  }
+
+  return false
+}
+
+export function createSisyphusOrchestratorHook(
+  ctx: PluginInput,
+  options?: SisyphusOrchestratorHookOptions
+) {
+  const backgroundManager = options?.backgroundManager
+  const sessions = new Map<string, SessionState>()
+  const pendingFilePaths = new Map<string, string>()
+
+  function getState(sessionID: string): SessionState {
+    let state = sessions.get(sessionID)
+    if (!state) {
+      state = {}
+      sessions.set(sessionID, state)
+    }
+    return state
+  }
+
+  async function injectContinuation(sessionID: string, planName: string, remaining: number, total: number): Promise<void> {
+    const hasRunningBgTasks = backgroundManager
+      ? backgroundManager.getTasksByParentSession(sessionID).some(t => t.status === "running")
+      : false
+
+    if (hasRunningBgTasks) {
+      log(`[${HOOK_NAME}] Skipped injection: background tasks running`, { sessionID })
+      return
+    }
+
+    const prompt = BOULDER_CONTINUATION_PROMPT
+      .replace(/{PLAN_NAME}/g, planName) +
+      `\n\n[Status: ${total - remaining}/${total} completed, ${remaining} remaining]`
+
+    try {
+      log(`[${HOOK_NAME}] Injecting boulder continuation`, { sessionID, planName, remaining })
+
+      await ctx.client.session.prompt({
+        path: { id: sessionID },
+        body: {
+          agent: "orchestrator-sisyphus",
+          parts: [{ type: "text", text: prompt }],
+        },
+        query: { directory: ctx.directory },
+      })
+
+      log(`[${HOOK_NAME}] Boulder continuation injected`, { sessionID })
+    } catch (err) {
+      log(`[${HOOK_NAME}] Boulder continuation failed`, { sessionID, error: String(err) })
+    }
+  }
+
+  return {
+    handler: async ({ event }: { event: { type: string; properties?: unknown } }): Promise<void> => {
+      const props = event.properties as Record<string, unknown> | undefined
+
+      if (event.type === "session.error") {
+        const sessionID = props?.sessionID as string | undefined
+        if (!sessionID) return
+
+        const state = getState(sessionID)
+        const isAbort = isAbortError(props?.error)
+        state.lastEventWasAbortError = isAbort
+
+        log(`[${HOOK_NAME}] session.error`, { sessionID, isAbort })
+        return
+      }
+
+      if (event.type === "session.idle") {
+        const sessionID = props?.sessionID as string | undefined
+        if (!sessionID) return
+
+        log(`[${HOOK_NAME}] session.idle`, { sessionID })
+
+        // Read boulder state FIRST to check if this session is part of an active boulder
+        const boulderState = readBoulderState(ctx.directory)
+        const isBoulderSession = boulderState?.session_ids.includes(sessionID) ?? false
+
+        const mainSessionID = getMainSessionID()
+        const isMainSession = sessionID === mainSessionID
+        const isBackgroundTaskSession = subagentSessions.has(sessionID)
+
+        // Allow continuation if: main session OR background task OR boulder session
+        if (mainSessionID && !isMainSession && !isBackgroundTaskSession && !isBoulderSession) {
+          log(`[${HOOK_NAME}] Skipped: not main, background task, or boulder session`, { sessionID })
+          return
+        }
+
+        const state = getState(sessionID)
+
+        if (state.lastEventWasAbortError) {
+          state.lastEventWasAbortError = false
+          log(`[${HOOK_NAME}] Skipped: abort error immediately before idle`, { sessionID })
+          return
+        }
+
+        const hasRunningBgTasks = backgroundManager
+          ? backgroundManager.getTasksByParentSession(sessionID).some(t => t.status === "running")
+          : false
+
+        if (hasRunningBgTasks) {
+          log(`[${HOOK_NAME}] Skipped: background tasks running`, { sessionID })
+          return
+        }
+
+
+        if (!boulderState) {
+          log(`[${HOOK_NAME}] No active boulder`, { sessionID })
+          return
+        }
+
+        if (!isCallerOrchestrator(sessionID)) {
+          log(`[${HOOK_NAME}] Skipped: last agent is not orchestrator-sisyphus`, { sessionID })
+          return
+        }
+
+        const progress = getPlanProgress(boulderState.active_plan)
+        if (progress.isComplete) {
+          log(`[${HOOK_NAME}] Boulder complete`, { sessionID, plan: boulderState.plan_name })
+          return
+        }
+
+        const remaining = progress.total - progress.completed
+        injectContinuation(sessionID, boulderState.plan_name, remaining, progress.total)
+        return
+      }
+
+      if (event.type === "message.updated") {
+        const info = props?.info as Record<string, unknown> | undefined
+        const sessionID = info?.sessionID as string | undefined
+
+        if (!sessionID) return
+
+        const state = sessions.get(sessionID)
+        if (state) {
+          state.lastEventWasAbortError = false
+        }
+        return
+      }
+
+      if (event.type === "message.part.updated") {
+        const info = props?.info as Record<string, unknown> | undefined
+        const sessionID = info?.sessionID as string | undefined
+        const role = info?.role as string | undefined
+
+        if (sessionID && role === "assistant") {
+          const state = sessions.get(sessionID)
+          if (state) {
+            state.lastEventWasAbortError = false
+          }
+        }
+        return
+      }
+
+      if (event.type === "tool.execute.before" || event.type === "tool.execute.after") {
+        const sessionID = props?.sessionID as string | undefined
+        if (sessionID) {
+          const state = sessions.get(sessionID)
+          if (state) {
+            state.lastEventWasAbortError = false
+          }
+        }
+        return
+      }
+
+      if (event.type === "session.deleted") {
+        const sessionInfo = props?.info as { id?: string } | undefined
+        if (sessionInfo?.id) {
+          sessions.delete(sessionInfo.id)
+          log(`[${HOOK_NAME}] Session deleted: cleaned up`, { sessionID: sessionInfo.id })
+        }
+        return
+      }
+    },
+
+    "tool.execute.before": async (
+      input: { tool: string; sessionID?: string; callID?: string },
+      output: { args: Record<string, unknown>; message?: string }
+    ): Promise<void> => {
+      if (!isCallerOrchestrator(input.sessionID)) {
+        return
+      }
+
+      // Check Write/Edit tools for orchestrator - inject strong warning
+      if (WRITE_EDIT_TOOLS.includes(input.tool)) {
+        const filePath = (output.args.filePath ?? output.args.path ?? output.args.file) as string | undefined
+        if (filePath && !filePath.includes(ALLOWED_PATH_PREFIX)) {
+          // Store filePath for use in tool.execute.after
+          if (input.callID) {
+            pendingFilePaths.set(input.callID, filePath)
+          }
+          const warning = ORCHESTRATOR_DELEGATION_REQUIRED.replace("$FILE_PATH", filePath)
+          output.message = (output.message || "") + warning
+          log(`[${HOOK_NAME}] Injected delegation warning for direct file modification`, {
+            sessionID: input.sessionID,
+            tool: input.tool,
+            filePath,
+          })
+        }
+        return
+      }
+
+      // Check sisyphus_task - inject single-task directive
+      if (input.tool === "sisyphus_task") {
+        const prompt = output.args.prompt as string | undefined
+        if (prompt && !prompt.includes("[SYSTEM DIRECTIVE - SINGLE TASK ONLY]")) {
+          output.args.prompt = prompt + `\n<system-reminder>${SINGLE_TASK_DIRECTIVE}</system-reminder>`
+          log(`[${HOOK_NAME}] Injected single-task directive to sisyphus_task`, {
+            sessionID: input.sessionID,
+          })
+        }
+      }
+    },
+
+    "tool.execute.after": async (
+      input: ToolExecuteAfterInput,
+      output: ToolExecuteAfterOutput
+    ): Promise<void> => {
+      if (!isCallerOrchestrator(input.sessionID)) {
+        return
+      }
+
+      if (WRITE_EDIT_TOOLS.includes(input.tool)) {
+        let filePath = input.callID ? pendingFilePaths.get(input.callID) : undefined
+        if (input.callID) {
+          pendingFilePaths.delete(input.callID)
+        }
+        if (!filePath) {
+          filePath = output.metadata?.filePath as string | undefined
+        }
+        if (filePath && !filePath.includes(ALLOWED_PATH_PREFIX)) {
+          output.output = (output.output || "") + DIRECT_WORK_REMINDER
+          log(`[${HOOK_NAME}] Direct work reminder appended`, {
+            sessionID: input.sessionID,
+            tool: input.tool,
+            filePath,
+          })
+        }
+        return
+      }
+
+      if (input.tool !== "sisyphus_task") {
+        return
+      }
+
+      const outputStr = output.output && typeof output.output === "string" ? output.output : ""
+      const isBackgroundLaunch = outputStr.includes("Background task launched") || outputStr.includes("Background task resumed")
+      
+      if (isBackgroundLaunch) {
+        return
+      }
+      
+      if (output.output && typeof output.output === "string") {
+        const gitStats = getGitDiffStats(ctx.directory)
+        const fileChanges = formatFileChanges(gitStats)
+        const subagentSessionId = extractSessionIdFromOutput(output.output)
+
+        const boulderState = readBoulderState(ctx.directory)
+
+        if (boulderState) {
+          const progress = getPlanProgress(boulderState.active_plan)
+
+          if (input.sessionID && !boulderState.session_ids.includes(input.sessionID)) {
+            appendSessionId(ctx.directory, input.sessionID)
+            log(`[${HOOK_NAME}] Appended session to boulder`, {
+              sessionID: input.sessionID,
+              plan: boulderState.plan_name,
+            })
+          }
+
+          output.output = `
+## SUBAGENT WORK COMPLETED
+
+${fileChanges}
+<system-reminder>
+${buildOrchestratorReminder(boulderState.plan_name, progress, subagentSessionId)}
+</system-reminder>`
+
+          log(`[${HOOK_NAME}] Output transformed for orchestrator mode (boulder)`, {
+            plan: boulderState.plan_name,
+            progress: `${progress.completed}/${progress.total}`,
+            fileCount: gitStats.length,
+          })
+        } else {
+          output.output += `\n<system-reminder>\n${buildStandaloneVerificationReminder(subagentSessionId)}\n</system-reminder>`
+
+          log(`[${HOOK_NAME}] Verification reminder appended for orchestrator`, {
+            sessionID: input.sessionID,
+            fileCount: gitStats.length,
+          })
+        }
+      }
+    },
+  }
+}

src/hooks/start-work/index.test.ts

@@ -0,0 +1,240 @@
+import { describe, expect, test, beforeEach, afterEach } from "bun:test"
+import { existsSync, mkdirSync, rmSync, writeFileSync } from "node:fs"
+import { join } from "node:path"
+import { tmpdir, homedir } from "node:os"
+import { createStartWorkHook } from "./index"
+import {
+  writeBoulderState,
+  clearBoulderState,
+} from "../../features/boulder-state"
+import type { BoulderState } from "../../features/boulder-state"
+
+describe("start-work hook", () => {
+  const TEST_DIR = join(tmpdir(), "start-work-test-" + Date.now())
+  const SISYPHUS_DIR = join(TEST_DIR, ".sisyphus")
+
+  function createMockPluginInput() {
+    return {
+      directory: TEST_DIR,
+      client: {},
+    } as Parameters<typeof createStartWorkHook>[0]
+  }
+
+  beforeEach(() => {
+    if (!existsSync(TEST_DIR)) {
+      mkdirSync(TEST_DIR, { recursive: true })
+    }
+    if (!existsSync(SISYPHUS_DIR)) {
+      mkdirSync(SISYPHUS_DIR, { recursive: true })
+    }
+    clearBoulderState(TEST_DIR)
+  })
+
+  afterEach(() => {
+    clearBoulderState(TEST_DIR)
+    if (existsSync(TEST_DIR)) {
+      rmSync(TEST_DIR, { recursive: true, force: true })
+    }
+  })
+
+  describe("chat.message handler", () => {
+    test("should ignore non-start-work commands", async () => {
+      // #given - hook and non-start-work message
+      const hook = createStartWorkHook(createMockPluginInput())
+      const output = {
+        parts: [{ type: "text", text: "Just a regular message" }],
+      }
+
+      // #when
+      await hook["chat.message"](
+        { sessionID: "session-123" },
+        output
+      )
+
+      // #then - output should be unchanged
+      expect(output.parts[0].text).toBe("Just a regular message")
+    })
+
+    test("should detect start-work command via session-context tag", async () => {
+      // #given - hook and start-work message
+      const hook = createStartWorkHook(createMockPluginInput())
+      const output = {
+        parts: [
+          {
+            type: "text",
+            text: "<session-context>Some context here</session-context>",
+          },
+        ],
+      }
+
+      // #when
+      await hook["chat.message"](
+        { sessionID: "session-123" },
+        output
+      )
+
+      // #then - output should be modified with context info
+      expect(output.parts[0].text).toContain("---")
+    })
+
+    test("should inject resume info when existing boulder state found", async () => {
+      // #given - existing boulder state with incomplete plan
+      const planPath = join(TEST_DIR, "test-plan.md")
+      writeFileSync(planPath, "# Plan\n- [ ] Task 1\n- [x] Task 2")
+
+      const state: BoulderState = {
+        active_plan: planPath,
+        started_at: "2026-01-02T10:00:00Z",
+        session_ids: ["session-1"],
+        plan_name: "test-plan",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      const hook = createStartWorkHook(createMockPluginInput())
+      const output = {
+        parts: [{ type: "text", text: "Start Sisyphus work session" }],
+      }
+
+      // #when
+      await hook["chat.message"](
+        { sessionID: "session-123" },
+        output
+      )
+
+      // #then - should show resuming status
+      expect(output.parts[0].text).toContain("RESUMING")
+      expect(output.parts[0].text).toContain("test-plan")
+    })
+
+    test("should replace $SESSION_ID placeholder", async () => {
+      // #given - hook and message with placeholder
+      const hook = createStartWorkHook(createMockPluginInput())
+      const output = {
+        parts: [
+          {
+            type: "text",
+            text: "Start Sisyphus work session\nSession: $SESSION_ID",
+          },
+        ],
+      }
+
+      // #when
+      await hook["chat.message"](
+        { sessionID: "ses-abc123" },
+        output
+      )
+
+      // #then - placeholder should be replaced
+      expect(output.parts[0].text).toContain("ses-abc123")
+      expect(output.parts[0].text).not.toContain("$SESSION_ID")
+    })
+
+    test("should replace $TIMESTAMP placeholder", async () => {
+      // #given - hook and message with placeholder
+      const hook = createStartWorkHook(createMockPluginInput())
+      const output = {
+        parts: [
+          {
+            type: "text",
+            text: "Start Sisyphus work session\nTime: $TIMESTAMP",
+          },
+        ],
+      }
+
+      // #when
+      await hook["chat.message"](
+        { sessionID: "session-123" },
+        output
+      )
+
+      // #then - placeholder should be replaced with ISO timestamp
+      expect(output.parts[0].text).not.toContain("$TIMESTAMP")
+      expect(output.parts[0].text).toMatch(/\d{4}-\d{2}-\d{2}T/)
+    })
+
+    test("should auto-select when only one incomplete plan among multiple plans", async () => {
+      // #given - multiple plans but only one incomplete
+      const plansDir = join(TEST_DIR, ".sisyphus", "plans")
+      mkdirSync(plansDir, { recursive: true })
+
+      // Plan 1: complete (all checked)
+      const plan1Path = join(plansDir, "plan-complete.md")
+      writeFileSync(plan1Path, "# Plan Complete\n- [x] Task 1\n- [x] Task 2")
+
+      // Plan 2: incomplete (has unchecked)
+      const plan2Path = join(plansDir, "plan-incomplete.md")
+      writeFileSync(plan2Path, "# Plan Incomplete\n- [ ] Task 1\n- [x] Task 2")
+
+      const hook = createStartWorkHook(createMockPluginInput())
+      const output = {
+        parts: [{ type: "text", text: "Start Sisyphus work session" }],
+      }
+
+      // #when
+      await hook["chat.message"](
+        { sessionID: "session-123" },
+        output
+      )
+
+      // #then - should auto-select the incomplete plan, not ask user
+      expect(output.parts[0].text).toContain("Auto-Selected Plan")
+      expect(output.parts[0].text).toContain("plan-incomplete")
+      expect(output.parts[0].text).not.toContain("Multiple Plans Found")
+    })
+
+    test("should wrap multiple plans message in system-reminder tag", async () => {
+      // #given - multiple incomplete plans
+      const plansDir = join(TEST_DIR, ".sisyphus", "plans")
+      mkdirSync(plansDir, { recursive: true })
+
+      const plan1Path = join(plansDir, "plan-a.md")
+      writeFileSync(plan1Path, "# Plan A\n- [ ] Task 1")
+
+      const plan2Path = join(plansDir, "plan-b.md")
+      writeFileSync(plan2Path, "# Plan B\n- [ ] Task 2")
+
+      const hook = createStartWorkHook(createMockPluginInput())
+      const output = {
+        parts: [{ type: "text", text: "Start Sisyphus work session" }],
+      }
+
+      // #when
+      await hook["chat.message"](
+        { sessionID: "session-123" },
+        output
+      )
+
+      // #then - should use system-reminder tag format
+      expect(output.parts[0].text).toContain("<system-reminder>")
+      expect(output.parts[0].text).toContain("</system-reminder>")
+      expect(output.parts[0].text).toContain("Multiple Plans Found")
+    })
+
+    test("should use 'ask user' prompt style for multiple plans", async () => {
+      // #given - multiple incomplete plans
+      const plansDir = join(TEST_DIR, ".sisyphus", "plans")
+      mkdirSync(plansDir, { recursive: true })
+
+      const plan1Path = join(plansDir, "plan-x.md")
+      writeFileSync(plan1Path, "# Plan X\n- [ ] Task 1")
+
+      const plan2Path = join(plansDir, "plan-y.md")
+      writeFileSync(plan2Path, "# Plan Y\n- [ ] Task 2")
+
+      const hook = createStartWorkHook(createMockPluginInput())
+      const output = {
+        parts: [{ type: "text", text: "Start Sisyphus work session" }],
+      }
+
+      // #when
+      await hook["chat.message"](
+        { sessionID: "session-123" },
+        output
+      )
+
+      // #then - should prompt agent to ask user, not ask directly
+      expect(output.parts[0].text).toContain("Ask the user")
+      expect(output.parts[0].text).not.toContain("Which plan would you like to work on?")
+    })
+  })
+})

src/hooks/start-work/index.ts

@@ -0,0 +1,153 @@
+import type { PluginInput } from "@opencode-ai/plugin"
+import {
+  readBoulderState,
+  writeBoulderState,
+  appendSessionId,
+  findPrometheusPlans,
+  getPlanProgress,
+  createBoulderState,
+  getPlanName,
+} from "../../features/boulder-state"
+import { log } from "../../shared/logger"
+
+export const HOOK_NAME = "start-work"
+
+interface StartWorkHookInput {
+  sessionID: string
+  messageID?: string
+}
+
+interface StartWorkHookOutput {
+  parts: Array<{ type: string; text?: string }>
+}
+
+export function createStartWorkHook(ctx: PluginInput) {
+  return {
+    "chat.message": async (
+      input: StartWorkHookInput,
+      output: StartWorkHookOutput
+    ): Promise<void> => {
+      const parts = output.parts
+      const promptText = parts
+        ?.filter((p) => p.type === "text" && p.text)
+        .map((p) => p.text)
+        .join("\n")
+        .trim() || ""
+
+      const isStartWorkCommand =
+        promptText.includes("Start Sisyphus work session") ||
+        promptText.includes("<session-context>")
+
+      if (!isStartWorkCommand) {
+        return
+      }
+
+      log(`[${HOOK_NAME}] Processing start-work command`, {
+        sessionID: input.sessionID,
+      })
+
+      const existingState = readBoulderState(ctx.directory)
+      const sessionId = input.sessionID
+      const timestamp = new Date().toISOString()
+
+      let contextInfo = ""
+
+      if (existingState) {
+        const progress = getPlanProgress(existingState.active_plan)
+        
+        if (!progress.isComplete) {
+          appendSessionId(ctx.directory, sessionId)
+          contextInfo = `
+## Active Work Session Found
+
+**Status**: RESUMING existing work
+**Plan**: ${existingState.plan_name}
+**Path**: ${existingState.active_plan}
+**Progress**: ${progress.completed}/${progress.total} tasks completed
+**Sessions**: ${existingState.session_ids.length + 1} (current session appended)
+**Started**: ${existingState.started_at}
+
+The current session (${sessionId}) has been added to session_ids.
+Read the plan file and continue from the first unchecked task.`
+        } else {
+          contextInfo = `
+## Previous Work Complete
+
+The previous plan (${existingState.plan_name}) has been completed.
+Looking for new plans...`
+        }
+      }
+
+      if (!existingState || getPlanProgress(existingState.active_plan).isComplete) {
+        const plans = findPrometheusPlans(ctx.directory)
+        const incompletePlans = plans.filter(p => !getPlanProgress(p).isComplete)
+        
+        if (plans.length === 0) {
+          contextInfo += `
+
+## No Plans Found
+
+No Prometheus plan files found at .sisyphus/plans/
+Use Prometheus to create a work plan first: /plan "your task"`
+        } else if (incompletePlans.length === 0) {
+          contextInfo += `
+
+## All Plans Complete
+
+All ${plans.length} plan(s) are complete. Create a new plan with: /plan "your task"`
+        } else if (incompletePlans.length === 1) {
+          const planPath = incompletePlans[0]
+          const progress = getPlanProgress(planPath)
+          const newState = createBoulderState(planPath, sessionId)
+          writeBoulderState(ctx.directory, newState)
+
+          contextInfo += `
+
+## Auto-Selected Plan
+
+**Plan**: ${getPlanName(planPath)}
+**Path**: ${planPath}
+**Progress**: ${progress.completed}/${progress.total} tasks
+**Session ID**: ${sessionId}
+**Started**: ${timestamp}
+
+boulder.json has been created. Read the plan and begin execution.`
+        } else {
+          const planList = incompletePlans.map((p, i) => {
+            const progress = getPlanProgress(p)
+            const stat = require("node:fs").statSync(p)
+            const modified = new Date(stat.mtimeMs).toISOString()
+            return `${i + 1}. [${getPlanName(p)}] - Modified: ${modified} - Progress: ${progress.completed}/${progress.total}`
+          }).join("\n")
+
+          contextInfo += `
+
+<system-reminder>
+## Multiple Plans Found
+
+Current Time: ${timestamp}
+Session ID: ${sessionId}
+
+${planList}
+
+Ask the user which plan to work on. Present the options above and wait for their response.
+</system-reminder>`
+        }
+      }
+
+      const idx = output.parts.findIndex((p) => p.type === "text" && p.text)
+      if (idx >= 0 && output.parts[idx].text) {
+        output.parts[idx].text = output.parts[idx].text
+          .replace(/\$SESSION_ID/g, sessionId)
+          .replace(/\$TIMESTAMP/g, timestamp)
+        
+        output.parts[idx].text += `\n\n---\n${contextInfo}`
+      }
+
+      log(`[${HOOK_NAME}] Context injected`, {
+        sessionID: input.sessionID,
+        hasExistingState: !!existingState,
+      })
+    },
+  }
+}

src/hooks/task-resume-info/index.ts

@@ -0,0 +1,36 @@
+const TARGET_TOOLS = ["task", "Task", "call_omo_agent", "sisyphus_task"]
+
+const SESSION_ID_PATTERNS = [
+  /Session ID: (ses_[a-zA-Z0-9_-]+)/,
+  /session_id: (ses_[a-zA-Z0-9_-]+)/,
+  /<task_metadata>\s*session_id: (ses_[a-zA-Z0-9_-]+)/,
+  /sessionId: (ses_[a-zA-Z0-9_-]+)/,
+]
+
+function extractSessionId(output: string): string | null {
+  for (const pattern of SESSION_ID_PATTERNS) {
+    const match = output.match(pattern)
+    if (match) return match[1]
+  }
+  return null
+}
+
+export function createTaskResumeInfoHook() {
+  const toolExecuteAfter = async (
+    input: { tool: string; sessionID: string; callID: string },
+    output: { title: string; output: string; metadata: unknown }
+  ) => {
+    if (!TARGET_TOOLS.includes(input.tool)) return
+    if (output.output.startsWith("Error:") || output.output.startsWith("Failed")) return
+    if (output.output.includes("\nto resume:")) return
+
+    const sessionId = extractSessionId(output.output)
+    if (!sessionId) return
+
+    output.output = output.output.trimEnd() + `\n\nto resume: sisyphus_task(resume="${sessionId}", prompt="...")`
+  }
+
+  return {
+    "tool.execute.after": toolExecuteAfter,
+  }
+}

src/hooks/todo-continuation-enforcer.test.ts

@@ -349,6 +349,25 @@ describe("todo-continuation-enforcer", () => {
     expect(promptCalls).toHaveLength(0)
   })
 
+  test("should accept skipAgents option without error", async () => {
+    // #given - session with skipAgents configured for Prometheus
+    const sessionID = "main-prometheus-option"
+    setMainSession(sessionID)
+
+    // #when - create hook with skipAgents option (should not throw)
+    const hook = createTodoContinuationEnforcer(createMockPluginInput(), {
+      skipAgents: ["Prometheus (Planner)", "custom-agent"],
+    })
+
+    // #then - handler works without error
+    await hook.handler({
+      event: { type: "session.idle", properties: { sessionID } },
+    })
+
+    await new Promise(r => setTimeout(r, 100))
+    expect(toastCalls.length).toBeGreaterThanOrEqual(1)
+  })
+
   test("should show countdown toast updates", async () => {
     // #given - session with incomplete todos
     const sessionID = "main-toast"

src/hooks/todo-continuation-enforcer.ts

@@ -11,8 +11,11 @@ import { log } from "../shared/logger"
 
 const HOOK_NAME = "todo-continuation-enforcer"
 
+const DEFAULT_SKIP_AGENTS = ["Prometheus (Planner)"]
+
 export interface TodoContinuationEnforcerOptions {
   backgroundManager?: BackgroundManager
+  skipAgents?: string[]
 }
 
 export interface TodoContinuationEnforcer {
@@ -89,7 +92,7 @@ export function createTodoContinuationEnforcer(
   ctx: PluginInput,
   options: TodoContinuationEnforcerOptions = {}
 ): TodoContinuationEnforcer {
-  const { backgroundManager } = options
+  const { backgroundManager, skipAgents = DEFAULT_SKIP_AGENTS } = options
   const sessions = new Map<string, SessionState>()
 
   function getState(sessionID: string): SessionState {
@@ -184,34 +187,32 @@ export function createTodoContinuationEnforcer(
     const messageDir = getMessageDir(sessionID)
     const prevMessage = messageDir ? findNearestMessageWithFields(messageDir) : null
 
-    const hasWritePermission = !prevMessage?.tools ||
-      (prevMessage.tools.write !== false && prevMessage.tools.edit !== false)
+    const agentName = prevMessage?.agent
+    if (agentName && skipAgents.includes(agentName)) {
+      log(`[${HOOK_NAME}] Skipped: agent in skipAgents list`, { sessionID, agent: agentName })
+      return
+    }
 
+    const editPermission = prevMessage?.tools?.edit
+    const writePermission = prevMessage?.tools?.write
+    const hasWritePermission = !prevMessage?.tools ||
+      ((editPermission !== false && editPermission !== "deny") &&
+       (writePermission !== false && writePermission !== "deny"))
     if (!hasWritePermission) {
       log(`[${HOOK_NAME}] Skipped: agent lacks write permission`, { sessionID, agent: prevMessage?.agent })
       return
     }
 
-    const agentName = prevMessage?.agent?.toLowerCase() ?? ""
-    if (agentName === "plan" || agentName === "planner-sisyphus") {
-      log(`[${HOOK_NAME}] Skipped: plan mode agent`, { sessionID, agent: prevMessage?.agent })
-      return
-    }
-
     const prompt = `${CONTINUATION_PROMPT}\n\n[Status: ${todos.length - freshIncompleteCount}/${todos.length} completed, ${freshIncompleteCount} remaining]`
 
-    const modelField = prevMessage?.model?.providerID && prevMessage?.model?.modelID
-      ? { providerID: prevMessage.model.providerID, modelID: prevMessage.model.modelID }
-      : undefined
-
     try {
-      log(`[${HOOK_NAME}] Injecting continuation`, { sessionID, agent: prevMessage?.agent, model: modelField, incompleteCount: freshIncompleteCount })
+      log(`[${HOOK_NAME}] Injecting continuation`, { sessionID, agent: prevMessage?.agent, incompleteCount: freshIncompleteCount })
 
+      // Don't pass model - let OpenCode use session's existing lastModel
       await ctx.client.session.prompt({
         path: { id: sessionID },
         body: {
           agent: prevMessage?.agent,
-          model: modelField,
           parts: [{ type: "text", text: prompt }],
         },
         query: { directory: ctx.directory },
@@ -324,6 +325,28 @@ export function createTodoContinuationEnforcer(
         return
       }
 
+      let agentName: string | undefined
+      try {
+        const messagesResp = await ctx.client.session.messages({
+          path: { id: sessionID },
+        })
+        const messages = (messagesResp.data ?? []) as Array<{ info?: { agent?: string } }>
+        for (let i = messages.length - 1; i >= 0; i--) {
+          if (messages[i].info?.agent) {
+            agentName = messages[i].info?.agent
+            break
+          }
+        }
+      } catch (err) {
+        log(`[${HOOK_NAME}] Failed to fetch messages for agent check`, { sessionID, error: String(err) })
+      }
+
+      log(`[${HOOK_NAME}] Agent check`, { sessionID, agentName, skipAgents })
+      if (agentName && skipAgents.includes(agentName)) {
+        log(`[${HOOK_NAME}] Skipped: agent in skipAgents list`, { sessionID, agent: agentName })
+        return
+      }
+
       startCountdown(sessionID, incompleteCount, todos.length)
       return
     }

src/index.ts

@@ -26,6 +26,10 @@ import {
   createRalphLoopHook,
   createAutoSlashCommandHook,
   createEditErrorRecoveryHook,
+  createTaskResumeInfoHook,
+  createStartWorkHook,
+  createSisyphusOrchestratorHook,
+  createPrometheusMdOnlyHook,
 } from "./hooks";
 import {
   contextCollector,
@@ -56,11 +60,13 @@ import {
   createSlashcommandTool,
   discoverCommandsSync,
   sessionExists,
+  createSisyphusTask,
   interactive_bash,
   startTmuxCheck,
 } from "./tools";
 import { BackgroundManager } from "./features/background-agent";
 import { SkillMcpManager } from "./features/skill-mcp-manager";
+import { initTaskToastManager } from "./features/task-toast-manager";
 import { type HookName } from "./config";
 import { log, detectExternalNotificationPlugin, getNotificationConflictWarning } from "./shared";
 import { loadPluginConfig } from "./plugin-config";
@@ -120,10 +126,14 @@ const OhMyOpenCodePlugin: Plugin = async (ctx) => {
     ? createEmptyTaskResponseDetectorHook(ctx)
     : null;
   const thinkMode = isHookEnabled("think-mode") ? createThinkModeHook() : null;
-  const claudeCodeHooks = createClaudeCodeHooksHook(ctx, {
-    disabledHooks: (pluginConfig.claude_code?.hooks ?? true) ? undefined : true,
-    keywordDetectorDisabled: !isHookEnabled("keyword-detector"),
-  });
+  const claudeCodeHooks = createClaudeCodeHooksHook(
+    ctx,
+    {
+      disabledHooks: (pluginConfig.claude_code?.hooks ?? true) ? undefined : true,
+      keywordDetectorDisabled: !isHookEnabled("keyword-detector"),
+    },
+    contextCollector
+  );
   const anthropicContextWindowLimitRecovery = isHookEnabled(
     "anthropic-context-window-limit-recovery"
   )
@@ -186,8 +196,24 @@ const OhMyOpenCodePlugin: Plugin = async (ctx) => {
     ? createEditErrorRecoveryHook(ctx)
     : null;
 
+  const startWork = isHookEnabled("start-work")
+    ? createStartWorkHook(ctx)
+    : null;
+
+  const sisyphusOrchestrator = isHookEnabled("sisyphus-orchestrator")
+    ? createSisyphusOrchestratorHook(ctx)
+    : null;
+
+  const prometheusMdOnly = isHookEnabled("prometheus-md-only")
+    ? createPrometheusMdOnlyHook(ctx)
+    : null;
+
+  const taskResumeInfo = createTaskResumeInfoHook();
+
   const backgroundManager = new BackgroundManager(ctx);
 
+  initTaskToastManager(ctx.client);
+
   const todoContinuationEnforcer = isHookEnabled("todo-continuation-enforcer")
     ? createTodoContinuationEnforcer(ctx, { backgroundManager })
     : null;
@@ -206,6 +232,11 @@ const OhMyOpenCodePlugin: Plugin = async (ctx) => {
 
   const callOmoAgent = createCallOmoAgent(ctx, backgroundManager);
   const lookAt = createLookAt(ctx);
+  const sisyphusTask = createSisyphusTask({
+    manager: backgroundManager,
+    client: ctx.client,
+    userCategories: pluginConfig.categories,
+  });
   const disabledSkills = new Set(pluginConfig.disabled_skills ?? []);
   const systemMcpNames = getSystemMcpServerNames();
   const builtinSkills = createBuiltinSkills().filter((skill) => {
@@ -273,6 +304,7 @@ const OhMyOpenCodePlugin: Plugin = async (ctx) => {
       ...backgroundTools,
       call_omo_agent: callOmoAgent,
       look_at: lookAt,
+      sisyphus_task: sisyphusTask,
       skill: skillTool,
       skill_mcp: skillMcpTool,
       slashcommand: slashcommandTool,
@@ -284,6 +316,7 @@ const OhMyOpenCodePlugin: Plugin = async (ctx) => {
       await keywordDetector?.["chat.message"]?.(input, output);
       await contextInjector["chat.message"]?.(input, output);
       await autoSlashCommand?.["chat.message"]?.(input, output);
+      await startWork?.["chat.message"]?.(input, output);
 
       if (ralphLoop) {
         const parts = (
@@ -373,6 +406,7 @@ const OhMyOpenCodePlugin: Plugin = async (ctx) => {
       await agentUsageReminder?.event(input);
       await interactiveBashSession?.event(input);
       await ralphLoop?.event(input);
+      await sisyphusOrchestrator?.handler(input);
 
       const { event } = input;
       const props = event.properties as Record<string, unknown> | undefined;
@@ -430,6 +464,7 @@ const OhMyOpenCodePlugin: Plugin = async (ctx) => {
       await directoryAgentsInjector?.["tool.execute.before"]?.(input, output);
       await directoryReadmeInjector?.["tool.execute.before"]?.(input, output);
       await rulesInjector?.["tool.execute.before"]?.(input, output);
+      await prometheusMdOnly?.["tool.execute.before"]?.(input, output);
 
       if (input.tool === "task") {
         const args = output.args as Record<string, unknown>;
@@ -440,7 +475,7 @@ const OhMyOpenCodePlugin: Plugin = async (ctx) => {
 
         args.tools = {
           ...(args.tools as Record<string, boolean> | undefined),
-          background_task: false,
+          sisyphus_task: false,
           ...(isExploreOrLibrarian ? { call_omo_agent: false } : {}),
         };
       }
@@ -488,6 +523,8 @@ const OhMyOpenCodePlugin: Plugin = async (ctx) => {
       await agentUsageReminder?.["tool.execute.after"](input, output);
       await interactiveBashSession?.["tool.execute.after"](input, output);
       await editErrorRecovery?.["tool.execute.after"](input, output);
+      await sisyphusOrchestrator?.["tool.execute.after"]?.(input, output);
+      await taskResumeInfo["tool.execute.after"](input, output);
     },
   };
 };

src/plugin-handlers/config-handler.ts

@@ -1,4 +1,5 @@
 import { createBuiltinAgents } from "../agents";
+import { createSisyphusJuniorAgent } from "../agents/sisyphus-junior";
 import {
   loadUserCommands,
   loadProjectCommands,
@@ -22,7 +23,7 @@ import { createBuiltinMcps } from "../mcp";
 import type { OhMyOpenCodeConfig } from "../config";
 import { log } from "../shared";
 import { migrateAgentConfig } from "../shared/permission-compat";
-import { PLAN_SYSTEM_PROMPT, PLAN_PERMISSION } from "../agents/plan-prompt";
+import { PROMETHEUS_SYSTEM_PROMPT, PROMETHEUS_PERMISSION } from "../agents/prometheus-prompt";
 import type { ModelCacheState } from "../plugin-state";
 
 export interface ConfigHandlerDeps {
@@ -131,6 +132,7 @@ export function createConfigHandler(deps: ConfigHandlerDeps) {
       explore?: { tools?: Record<string, unknown> };
       librarian?: { tools?: Record<string, unknown> };
       "multimodal-looker"?: { tools?: Record<string, unknown> };
+      "orchestrator-sisyphus"?: { tools?: Record<string, unknown> };
     };
     const configAgent = config.agent as AgentConfig | undefined;
 
@@ -141,6 +143,11 @@ export function createConfigHandler(deps: ConfigHandlerDeps) {
         Sisyphus: builtinAgents.Sisyphus,
       };
 
+      agentConfig["Sisyphus-Junior"] = createSisyphusJuniorAgent({
+        model: "anthropic/claude-sonnet-4-5",
+        temperature: 0.1,
+      });
+
       if (builderEnabled) {
         const { name: _buildName, ...buildConfigWithoutName } =
           configAgent?.build ?? {};
@@ -165,21 +172,21 @@ export function createConfigHandler(deps: ConfigHandlerDeps) {
         const migratedPlanConfig = migrateAgentConfig(
           planConfigWithoutName as Record<string, unknown>
         );
-        const plannerSisyphusOverride =
-          pluginConfig.agents?.["Planner-Sisyphus"];
+        const prometheusOverride =
+          pluginConfig.agents?.["Prometheus (Planner)"];
         const defaultModel = config.model as string | undefined;
-        const plannerSisyphusBase = {
-          model: (migratedPlanConfig as Record<string, unknown>).model ?? defaultModel,
+        const prometheusBase = {
+          model: defaultModel ?? "anthropic/claude-opus-4-5",
           mode: "primary" as const,
-          prompt: PLAN_SYSTEM_PROMPT,
-          permission: PLAN_PERMISSION,
-          description: `${configAgent?.plan?.description ?? "Plan agent"} (OhMyOpenCode version)`,
-          color: (configAgent?.plan?.color as string) ?? "#6495ED",
+          prompt: PROMETHEUS_SYSTEM_PROMPT,
+          permission: PROMETHEUS_PERMISSION,
+          description: `${configAgent?.plan?.description ?? "Plan agent"} (Prometheus - OhMyOpenCode)`,
+          color: (configAgent?.plan?.color as string) ?? "#FF6347",
         };
 
-        agentConfig["Planner-Sisyphus"] = plannerSisyphusOverride
-          ? { ...plannerSisyphusBase, ...plannerSisyphusOverride }
-          : plannerSisyphusBase;
+        agentConfig["Prometheus (Planner)"] = prometheusOverride
+          ? { ...prometheusBase, ...prometheusOverride }
+          : prometheusBase;
       }
 
     const filteredConfigAgents = configAgent
@@ -255,6 +262,13 @@ export function createConfigHandler(deps: ConfigHandlerDeps) {
         look_at: false,
       };
     }
+    if (agentResult["orchestrator-sisyphus"]) {
+      agentResult["orchestrator-sisyphus"].tools = {
+        ...agentResult["orchestrator-sisyphus"].tools,
+        task: false,
+        call_omo_agent: false,
+      };
+    }
 
     config.permission = {
       ...(config.permission as Record<string, unknown>),

src/shared/migration.test.ts

@@ -1,10 +1,14 @@
-import { describe, test, expect } from "bun:test"
+import { describe, test, expect, afterEach } from "bun:test"
+import * as fs from "fs"
+import * as path from "path"
 import {
   AGENT_NAME_MAP,
   HOOK_NAME_MAP,
   migrateAgentNames,
   migrateHookNames,
   migrateConfigFile,
+  migrateAgentConfigToCategory,
+  shouldDeleteAgentConfig,
 } from "./migration"
 
 describe("migrateAgentNames", () => {
@@ -19,10 +23,10 @@ describe("migrateAgentNames", () => {
     // #when: Migrate agent names
     const { migrated, changed } = migrateAgentNames(agents)
 
-    // #then: Legacy names should be migrated to Sisyphus
+    // #then: Legacy names should be migrated to Sisyphus/Prometheus
     expect(changed).toBe(true)
     expect(migrated["Sisyphus"]).toEqual({ temperature: 0.5 })
-    expect(migrated["Planner-Sisyphus"]).toEqual({ prompt: "custom prompt" })
+    expect(migrated["Prometheus (Planner)"]).toEqual({ prompt: "custom prompt" })
     expect(migrated["omo"]).toBeUndefined()
     expect(migrated["OmO"]).toBeUndefined()
     expect(migrated["OmO-Plan"]).toBeUndefined()
@@ -50,7 +54,7 @@ describe("migrateAgentNames", () => {
     // #given: Config with mixed case agent names
     const agents = {
       SISYPHUS: { model: "test" },
-      "PLANNER-SISYPHUS": { prompt: "test" },
+      "planner-sisyphus": { prompt: "test" },
     }
 
     // #when: Migrate agent names
@@ -58,7 +62,7 @@ describe("migrateAgentNames", () => {
 
     // #then: Case-insensitive lookup should migrate correctly
     expect(migrated["Sisyphus"]).toEqual({ model: "test" })
-    expect(migrated["Planner-Sisyphus"]).toEqual({ prompt: "test" })
+    expect(migrated["Prometheus (Planner)"]).toEqual({ prompt: "test" })
   })
 
   test("passes through unknown agent names unchanged", () => {
@@ -220,7 +224,7 @@ describe("migrateConfigFile", () => {
     expect(rawConfig.omo_agent).toBeUndefined()
     const agents = rawConfig.agents as Record<string, unknown>
     expect(agents["Sisyphus"]).toBeDefined()
-    expect(agents["Planner-Sisyphus"]).toBeDefined()
+    expect(agents["Prometheus (Planner)"]).toBeDefined()
     expect(rawConfig.disabled_hooks).toContain("anthropic-context-window-limit-recovery")
   })
 })
@@ -231,13 +235,404 @@ describe("migration maps", () => {
     // #then: Should contain all legacy → current mappings
     expect(AGENT_NAME_MAP["omo"]).toBe("Sisyphus")
     expect(AGENT_NAME_MAP["OmO"]).toBe("Sisyphus")
-    expect(AGENT_NAME_MAP["OmO-Plan"]).toBe("Planner-Sisyphus")
-    expect(AGENT_NAME_MAP["omo-plan"]).toBe("Planner-Sisyphus")
+    expect(AGENT_NAME_MAP["OmO-Plan"]).toBe("Prometheus (Planner)")
+    expect(AGENT_NAME_MAP["omo-plan"]).toBe("Prometheus (Planner)")
+    expect(AGENT_NAME_MAP["Planner-Sisyphus"]).toBe("Prometheus (Planner)")
+    expect(AGENT_NAME_MAP["plan-consultant"]).toBe("Metis (Plan Consultant)")
   })
 
   test("HOOK_NAME_MAP contains anthropic-auto-compact migration", () => {
     // #given/#when: Check HOOK_NAME_MAP
-    // #then: Should contain the legacy hook name mapping
+    // #then: Should contain be legacy hook name mapping
     expect(HOOK_NAME_MAP["anthropic-auto-compact"]).toBe("anthropic-context-window-limit-recovery")
   })
 })
+
+describe("migrateAgentConfigToCategory", () => {
+  test("migrates model to category when mapping exists", () => {
+    // #given: Config with a model that has a category mapping
+    const config = {
+      model: "google/gemini-3-pro-preview",
+      temperature: 0.5,
+      top_p: 0.9,
+    }
+
+    // #when: Migrate agent config to category
+    const { migrated, changed } = migrateAgentConfigToCategory(config)
+
+    // #then: Model should be replaced with category
+    expect(changed).toBe(true)
+    expect(migrated.category).toBe("visual-engineering")
+    expect(migrated.model).toBeUndefined()
+    expect(migrated.temperature).toBe(0.5)
+    expect(migrated.top_p).toBe(0.9)
+  })
+
+  test("does not migrate when model is not in map", () => {
+    // #given: Config with a model that has no mapping
+    const config = {
+      model: "custom/model",
+      temperature: 0.5,
+    }
+
+    // #when: Migrate agent config to category
+    const { migrated, changed } = migrateAgentConfigToCategory(config)
+
+    // #then: Config should remain unchanged
+    expect(changed).toBe(false)
+    expect(migrated).toEqual(config)
+  })
+
+  test("does not migrate when model is not a string", () => {
+    // #given: Config with non-string model
+    const config = {
+      model: { name: "test" },
+      temperature: 0.5,
+    }
+
+    // #when: Migrate agent config to category
+    const { migrated, changed } = migrateAgentConfigToCategory(config)
+
+    // #then: Config should remain unchanged
+    expect(changed).toBe(false)
+    expect(migrated).toEqual(config)
+  })
+
+  test("handles all mapped models correctly", () => {
+    // #given: Configs for each mapped model
+    const configs = [
+      { model: "google/gemini-3-pro-preview" },
+      { model: "openai/gpt-5.2" },
+      { model: "anthropic/claude-haiku-4-5" },
+      { model: "anthropic/claude-opus-4-5" },
+      { model: "anthropic/claude-sonnet-4-5" },
+    ]
+
+    const expectedCategories = ["visual-engineering", "ultrabrain", "quick", "most-capable", "general"]
+
+    // #when: Migrate each config
+    const results = configs.map(migrateAgentConfigToCategory)
+
+    // #then: Each model should map to correct category
+    results.forEach((result, index) => {
+      expect(result.changed).toBe(true)
+      expect(result.migrated.category).toBe(expectedCategories[index])
+      expect(result.migrated.model).toBeUndefined()
+    })
+  })
+
+  test("preserves non-model fields during migration", () => {
+    // #given: Config with multiple fields
+    const config = {
+      model: "openai/gpt-5.2",
+      temperature: 0.1,
+      top_p: 0.95,
+      maxTokens: 4096,
+      prompt_append: "custom instruction",
+    }
+
+    // #when: Migrate agent config to category
+    const { migrated } = migrateAgentConfigToCategory(config)
+
+    // #then: All non-model fields should be preserved
+    expect(migrated.category).toBe("ultrabrain")
+    expect(migrated.temperature).toBe(0.1)
+    expect(migrated.top_p).toBe(0.95)
+    expect(migrated.maxTokens).toBe(4096)
+    expect(migrated.prompt_append).toBe("custom instruction")
+  })
+})
+
+describe("shouldDeleteAgentConfig", () => {
+  test("returns true when config only has category field", () => {
+    // #given: Config with only category field (no overrides)
+    const config = { category: "visual-engineering" }
+
+    // #when: Check if config should be deleted
+    const shouldDelete = shouldDeleteAgentConfig(config, "visual-engineering")
+
+    // #then: Should return true (matches category defaults)
+    expect(shouldDelete).toBe(true)
+  })
+
+  test("returns false when category does not exist", () => {
+    // #given: Config with unknown category
+    const config = { category: "unknown" }
+
+    // #when: Check if config should be deleted
+    const shouldDelete = shouldDeleteAgentConfig(config, "unknown")
+
+    // #then: Should return false (category not found)
+    expect(shouldDelete).toBe(false)
+  })
+
+  test("returns true when all fields match category defaults", () => {
+    // #given: Config with fields matching category defaults
+    const config = {
+      category: "visual-engineering",
+      model: "google/gemini-3-pro-preview",
+      temperature: 0.7,
+    }
+
+    // #when: Check if config should be deleted
+    const shouldDelete = shouldDeleteAgentConfig(config, "visual-engineering")
+
+    // #then: Should return true (all fields match defaults)
+    expect(shouldDelete).toBe(true)
+  })
+
+  test("returns false when fields differ from category defaults", () => {
+    // #given: Config with custom temperature override
+    const config = {
+      category: "visual-engineering",
+      temperature: 0.9, // Different from default (0.7)
+    }
+
+    // #when: Check if config should be deleted
+    const shouldDelete = shouldDeleteAgentConfig(config, "visual-engineering")
+
+    // #then: Should return false (has custom override)
+    expect(shouldDelete).toBe(false)
+  })
+
+  test("handles different categories with their defaults", () => {
+    // #given: Configs for different categories
+    const configs = [
+      { category: "ultrabrain", temperature: 0.1 },
+      { category: "quick", temperature: 0.3 },
+      { category: "most-capable", temperature: 0.1 },
+      { category: "general", temperature: 0.3 },
+    ]
+
+    // #when: Check each config
+    const results = configs.map((config) => shouldDeleteAgentConfig(config, config.category as string))
+
+    // #then: All should be true (all match defaults)
+    results.forEach((result) => {
+      expect(result).toBe(true)
+    })
+  })
+
+  test("returns false when additional fields are present", () => {
+    // #given: Config with extra fields
+    const config = {
+      category: "visual-engineering",
+      temperature: 0.7,
+      custom_field: "value", // Extra field not in defaults
+    }
+
+    // #when: Check if config should be deleted
+    const shouldDelete = shouldDeleteAgentConfig(config, "visual-engineering")
+
+    // #then: Should return false (has extra field)
+    expect(shouldDelete).toBe(false)
+  })
+
+  test("handles complex config with multiple overrides", () => {
+    // #given: Config with multiple custom overrides
+    const config = {
+      category: "visual-engineering",
+      temperature: 0.5, // Different from default
+      top_p: 0.8, // Different from default
+      prompt_append: "custom prompt", // Custom field
+    }
+
+    // #when: Check if config should be deleted
+    const shouldDelete = shouldDeleteAgentConfig(config, "visual-engineering")
+
+    // #then: Should return false (has overrides)
+    expect(shouldDelete).toBe(false)
+  })
+})
+
+describe("migrateConfigFile with backup", () => {
+  const cleanupPaths: string[] = []
+
+  afterEach(() => {
+    cleanupPaths.forEach((p) => {
+      try {
+        fs.unlinkSync(p)
+      } catch {
+      }
+    })
+  })
+
+  test("creates backup file with timestamp when migration needed", () => {
+    // #given: Config file path and config needing migration
+    const testConfigPath = "/tmp/test-config-migration.json"
+    const testConfigContent = globalThis.JSON.stringify({ agents: { oracle: { model: "openai/gpt-5.2" } } }, null, 2)
+    const rawConfig: Record<string, unknown> = {
+      agents: {
+        oracle: { model: "openai/gpt-5.2" },
+      },
+    }
+
+    fs.writeFileSync(testConfigPath, testConfigContent)
+    cleanupPaths.push(testConfigPath)
+
+    // #when: Migrate config file
+    const needsWrite = migrateConfigFile(testConfigPath, rawConfig)
+
+    // #then: Backup file should be created with timestamp
+    expect(needsWrite).toBe(true)
+
+    const dir = path.dirname(testConfigPath)
+    const basename = path.basename(testConfigPath)
+    const files = fs.readdirSync(dir)
+    const backupFiles = files.filter((f) => f.startsWith(`${basename}.bak.`))
+    expect(backupFiles.length).toBeGreaterThan(0)
+
+    const backupFile = backupFiles[0]
+    const backupPath = path.join(dir, backupFile)
+    cleanupPaths.push(backupPath)
+
+    expect(backupFile).toMatch(/test-config-migration\.json\.bak\.\d{4}-\d{2}-\d{2}T\d{2}-\d{2}-\d{2}/)
+
+    const backupContent = fs.readFileSync(backupPath, "utf-8")
+    expect(backupContent).toBe(testConfigContent)
+  })
+
+  test("deletes agent config when all fields match category defaults", () => {
+    // #given: Config with agent matching category defaults
+    const testConfigPath = "/tmp/test-config-delete.json"
+    const rawConfig: Record<string, unknown> = {
+      agents: {
+        oracle: {
+          model: "openai/gpt-5.2",
+          temperature: 0.1,
+        },
+      },
+    }
+
+    fs.writeFileSync(testConfigPath, globalThis.JSON.stringify({ agents: { oracle: { model: "openai/gpt-5.2" } } }, null, 2))
+    cleanupPaths.push(testConfigPath)
+
+    // #when: Migrate config file
+    const needsWrite = migrateConfigFile(testConfigPath, rawConfig)
+
+    // #then: Agent should be deleted (matches strategic category defaults)
+    expect(needsWrite).toBe(true)
+
+    const migratedConfig = JSON.parse(fs.readFileSync(testConfigPath, "utf-8"))
+    expect(migratedConfig.agents).toEqual({})
+
+    const dir = path.dirname(testConfigPath)
+    const basename = path.basename(testConfigPath)
+    const files = fs.readdirSync(dir)
+    const backupFiles = files.filter((f) => f.startsWith(`${basename}.bak.`))
+    backupFiles.forEach((f) => cleanupPaths.push(path.join(dir, f)))
+  })
+
+  test("keeps agent config with category when fields differ from defaults", () => {
+    // #given: Config with agent having custom temperature override
+    const testConfigPath = "/tmp/test-config-keep.json"
+    const rawConfig: Record<string, unknown> = {
+      agents: {
+        oracle: {
+          model: "openai/gpt-5.2",
+          temperature: 0.5,
+        },
+      },
+    }
+
+    fs.writeFileSync(testConfigPath, globalThis.JSON.stringify({ agents: { oracle: { model: "openai/gpt-5.2" } } }, null, 2))
+    cleanupPaths.push(testConfigPath)
+
+    // #when: Migrate config file
+    const needsWrite = migrateConfigFile(testConfigPath, rawConfig)
+
+    // #then: Agent should be kept with category and custom override
+    expect(needsWrite).toBe(true)
+
+    const migratedConfig = JSON.parse(fs.readFileSync(testConfigPath, "utf-8"))
+    const agents = migratedConfig.agents as Record<string, unknown>
+    expect(agents.oracle).toBeDefined()
+    expect((agents.oracle as Record<string, unknown>).category).toBe("ultrabrain")
+    expect((agents.oracle as Record<string, unknown>).temperature).toBe(0.5)
+    expect((agents.oracle as Record<string, unknown>).model).toBeUndefined()
+
+    const dir = path.dirname(testConfigPath)
+    const basename = path.basename(testConfigPath)
+    const files = fs.readdirSync(dir)
+    const backupFiles = files.filter((f) => f.startsWith(`${basename}.bak.`))
+    backupFiles.forEach((f) => cleanupPaths.push(path.join(dir, f)))
+  })
+
+  test("does not write when no migration needed", () => {
+    // #given: Config with no migrations needed
+    const testConfigPath = "/tmp/test-config-no-migration.json"
+    const rawConfig: Record<string, unknown> = {
+      agents: {
+        Sisyphus: { model: "test" },
+      },
+    }
+
+    fs.writeFileSync(testConfigPath, globalThis.JSON.stringify({ agents: { Sisyphus: { model: "test" } } }, null, 2))
+    cleanupPaths.push(testConfigPath)
+
+    // #when: Migrate config file
+    const needsWrite = migrateConfigFile(testConfigPath, rawConfig)
+
+    // #then: Should not write or create backup
+    expect(needsWrite).toBe(false)
+
+    const dir = path.dirname(testConfigPath)
+    const basename = path.basename(testConfigPath)
+    const files = fs.readdirSync(dir)
+    const backupFiles = files.filter((f) => f.startsWith(`${basename}.bak.`))
+    expect(backupFiles.length).toBe(0)
+  })
+
+  test("handles multiple agent migrations correctly", () => {
+    // #given: Config with multiple agents needing migration
+    const testConfigPath = "/tmp/test-config-multi-agent.json"
+    const rawConfig: Record<string, unknown> = {
+      agents: {
+        oracle: { model: "openai/gpt-5.2" },
+        librarian: { model: "anthropic/claude-sonnet-4-5" },
+        frontend: {
+          model: "google/gemini-3-pro-preview",
+          temperature: 0.9,
+        },
+      },
+    }
+
+    fs.writeFileSync(
+      testConfigPath,
+      globalThis.JSON.stringify(
+        {
+          agents: {
+            oracle: { model: "openai/gpt-5.2" },
+            librarian: { model: "anthropic/claude-sonnet-4-5" },
+            frontend: { model: "google/gemini-3-pro-preview" },
+          },
+        },
+        null,
+        2,
+      ),
+    )
+    cleanupPaths.push(testConfigPath)
+
+    // #when: Migrate config file
+    const needsWrite = migrateConfigFile(testConfigPath, rawConfig)
+
+    // #then: Should migrate correctly
+    expect(needsWrite).toBe(true)
+
+    const migratedConfig = JSON.parse(fs.readFileSync(testConfigPath, "utf-8"))
+    const agents = migratedConfig.agents as Record<string, unknown>
+
+    expect(agents.oracle).toBeUndefined()
+    expect(agents.librarian).toBeUndefined()
+
+    expect(agents.frontend).toBeDefined()
+    expect((agents.frontend as Record<string, unknown>).category).toBe("visual-engineering")
+    expect((agents.frontend as Record<string, unknown>).temperature).toBe(0.9)
+
+    const dir = path.dirname(testConfigPath)
+    const basename = path.basename(testConfigPath)
+    const files = fs.readdirSync(dir)
+    const backupFiles = files.filter((f) => f.startsWith(`${basename}.bak.`))
+    backupFiles.forEach((f) => cleanupPaths.push(path.join(dir, f)))
+  })
+})

src/shared/migration.ts

@@ -3,14 +3,16 @@ import { log } from "./logger"
 
 // Migration map: old keys → new keys (for backward compatibility)
 export const AGENT_NAME_MAP: Record<string, string> = {
-  // Legacy names (backward compatibility)
   omo: "Sisyphus",
   "OmO": "Sisyphus",
-  "OmO-Plan": "Planner-Sisyphus",
-  "omo-plan": "Planner-Sisyphus",
-  // Current names
   sisyphus: "Sisyphus",
-  "planner-sisyphus": "Planner-Sisyphus",
+  "OmO-Plan": "Prometheus (Planner)",
+  "omo-plan": "Prometheus (Planner)",
+  "Planner-Sisyphus": "Prometheus (Planner)",
+  "planner-sisyphus": "Prometheus (Planner)",
+  prometheus: "Prometheus (Planner)",
+  "plan-consultant": "Metis (Plan Consultant)",
+  metis: "Metis (Plan Consultant)",
   build: "build",
   oracle: "oracle",
   librarian: "librarian",
@@ -26,6 +28,15 @@ export const HOOK_NAME_MAP: Record<string, string> = {
   "anthropic-auto-compact": "anthropic-context-window-limit-recovery",
 }
 
+// Model to category mapping for auto-migration
+export const MODEL_TO_CATEGORY_MAP: Record<string, string> = {
+  "google/gemini-3-pro-preview": "visual-engineering",
+  "openai/gpt-5.2": "ultrabrain",
+  "anthropic/claude-haiku-4-5": "quick",
+  "anthropic/claude-opus-4-5": "most-capable",
+  "anthropic/claude-sonnet-4-5": "general",
+}
+
 export function migrateAgentNames(agents: Record<string, unknown>): { migrated: Record<string, unknown>; changed: boolean } {
   const migrated: Record<string, unknown> = {}
   let changed = false
@@ -56,6 +67,45 @@ export function migrateHookNames(hooks: string[]): { migrated: string[]; changed
   return { migrated, changed }
 }
 
+export function migrateAgentConfigToCategory(config: Record<string, unknown>): {
+  migrated: Record<string, unknown>
+  changed: boolean
+} {
+  const { model, ...rest } = config
+  if (typeof model !== "string") {
+    return { migrated: config, changed: false }
+  }
+
+  const category = MODEL_TO_CATEGORY_MAP[model]
+  if (!category) {
+    return { migrated: config, changed: false }
+  }
+
+  return {
+    migrated: { category, ...rest },
+    changed: true,
+  }
+}
+
+export function shouldDeleteAgentConfig(
+  config: Record<string, unknown>,
+  category: string
+): boolean {
+  const { DEFAULT_CATEGORIES } = require("../tools/sisyphus-task/constants")
+  const defaults = DEFAULT_CATEGORIES[category]
+  if (!defaults) return false
+
+  const keys = Object.keys(config).filter((k) => k !== "category")
+  if (keys.length === 0) return true
+
+  for (const key of keys) {
+    if (config[key] !== (defaults as Record<string, unknown>)[key]) {
+      return false
+    }
+  }
+  return true
+}
+
 export function migrateConfigFile(configPath: string, rawConfig: Record<string, unknown>): boolean {
   let needsWrite = false
 
@@ -67,6 +117,22 @@ export function migrateConfigFile(configPath: string, rawConfig: Record<string,
     }
   }
 
+  if (rawConfig.agents && typeof rawConfig.agents === "object") {
+    const agents = rawConfig.agents as Record<string, Record<string, unknown>>
+    for (const [name, config] of Object.entries(agents)) {
+      const { migrated, changed } = migrateAgentConfigToCategory(config)
+      if (changed) {
+        const category = migrated.category as string
+        if (shouldDeleteAgentConfig(migrated, category)) {
+          delete agents[name]
+        } else {
+          agents[name] = migrated
+        }
+        needsWrite = true
+      }
+    }
+  }
+
   if (rawConfig.omo_agent) {
     rawConfig.sisyphus_agent = rawConfig.omo_agent
     delete rawConfig.omo_agent
@@ -83,8 +149,12 @@ export function migrateConfigFile(configPath: string, rawConfig: Record<string,
 
   if (needsWrite) {
     try {
+      const timestamp = new Date().toISOString().replace(/[:.]/g, "-")
+      const backupPath = `${configPath}.bak.${timestamp}`
+      fs.copyFileSync(configPath, backupPath)
+
       fs.writeFileSync(configPath, JSON.stringify(rawConfig, null, 2) + "\n", "utf-8")
-      log(`Migrated config file: ${configPath}`)
+      log(`Migrated config file: ${configPath} (backup: ${backupPath})`)
     } catch (err) {
       log(`Failed to write migrated config to ${configPath}:`, err)
     }

src/tools/AGENTS.md

@@ -2,26 +2,35 @@
 
 ## OVERVIEW
 
-Custom tools: 11 LSP tools, AST-aware search/replace, file ops with timeouts, background task management, session navigation.
+Custom tools extending agent capabilities: LSP integration (11 tools), AST-aware code search/replace, file operations with timeouts, background task management.
 
 ## STRUCTURE
 
 ```
 tools/
 ├── ast-grep/           # AST-aware code search/replace (25 languages)
-│   ├── napi.ts         # @ast-grep/napi binding (preferred)
-│   └── cli.ts          # @ast-grep/cli fallback
+│   ├── cli.ts          # @ast-grep/cli subprocess
+│   ├── napi.ts         # @ast-grep/napi native binding (preferred)
+│   ├── constants.ts, types.ts, tools.ts, utils.ts
 ├── background-task/    # Async agent task management
 ├── call-omo-agent/     # Spawn explore/librarian agents
-├── glob/               # File pattern matching (60s timeout)
-├── grep/               # Content search (60s timeout)
+├── glob/               # File pattern matching (timeout-safe)
+├── grep/               # Content search (timeout-safe)
 ├── interactive-bash/   # Tmux session management
 ├── look-at/            # Multimodal analysis (PDF, images)
-├── lsp/                # 11 LSP tools (611 lines client.ts)
-│   ├── client.ts       # LSP connection lifecycle
+├── lsp/                # 11 LSP tools
+│   ├── client.ts       # LSP connection lifecycle (612 lines)
+│   ├── utils.ts        # LSP utilities (461 lines)
 │   ├── config.ts       # Server configurations
+│   ├── tools.ts        # Tool implementations (405 lines)
+│   └── types.ts
+├── session-manager/    # OpenCode session file management
+│   ├── constants.ts    # Storage paths, descriptions
+│   ├── types.ts        # Session data interfaces
+│   ├── storage.ts      # File I/O operations
+│   ├── utils.ts        # Formatting, filtering
 │   └── tools.ts        # Tool implementations
-├── session-manager/    # OpenCode session file ops
+├── sisyphus-task/      # Category-based task delegation (493 lines)
 ├── skill/              # Skill loading and execution
 ├── skill-mcp/          # Skill-embedded MCP invocation
 ├── slashcommand/       # Slash command execution
@@ -30,39 +39,47 @@ tools/
 
 ## TOOL CATEGORIES
 
-| Category | Tools |
-|----------|-------|
-| LSP | lsp_hover, lsp_goto_definition, lsp_find_references, lsp_document_symbols, lsp_workspace_symbols, lsp_diagnostics, lsp_servers, lsp_prepare_rename, lsp_rename, lsp_code_actions, lsp_code_action_resolve |
-| AST | ast_grep_search, ast_grep_replace |
-| File Search | grep, glob |
-| Session | session_list, session_read, session_search, session_info |
-| Background | background_task, background_output, background_cancel |
-| Multimodal | look_at |
-| Terminal | interactive_bash |
-| Skills | skill, skill_mcp |
-| Agents | call_omo_agent |
+| Category | Tools | Purpose |
+|----------|-------|---------|
+| LSP | lsp_hover, lsp_goto_definition, lsp_find_references, lsp_document_symbols, lsp_workspace_symbols, lsp_diagnostics, lsp_servers, lsp_prepare_rename, lsp_rename, lsp_code_actions, lsp_code_action_resolve | IDE-like code intelligence |
+| AST | ast_grep_search, ast_grep_replace | Pattern-based code search/replace |
+| File Search | grep, glob | Content and file pattern matching |
+| Session | session_list, session_read, session_search, session_info | OpenCode session file management |
+| Background | sisyphus_task, background_output, background_cancel | Async agent orchestration |
+| Multimodal | look_at | PDF/image analysis via Gemini |
+| Terminal | interactive_bash | Tmux session control |
+| Commands | slashcommand | Execute slash commands |
+| Skills | skill, skill_mcp | Load skills, invoke skill-embedded MCPs |
+| Agents | call_omo_agent | Spawn explore/librarian |
 
-## HOW TO ADD
+## HOW TO ADD A TOOL
 
-1. Create `src/tools/my-tool/`
-2. Files: `constants.ts`, `types.ts`, `tools.ts`, `index.ts`
+1. Create directory: `src/tools/my-tool/`
+2. Create files:
+   - `constants.ts`: `TOOL_NAME`, `TOOL_DESCRIPTION`
+   - `types.ts`: Parameter/result interfaces
+   - `tools.ts`: Tool implementation (returns OpenCode tool object)
+   - `index.ts`: Barrel export
+   - `utils.ts`: Helpers (optional)
 3. Add to `builtinTools` in `src/tools/index.ts`
 
 ## LSP SPECIFICS
 
-- Lazy init on first use, auto-shutdown on idle
-- Config priority: opencode.json > oh-my-opencode.json > defaults
-- Servers: typescript-language-server, pylsp, gopls, rust-analyzer
+- **Client lifecycle**: Lazy init on first use, auto-shutdown on idle
+- **Config priority**: opencode.json > oh-my-opencode.json > defaults
+- **Supported servers**: typescript-language-server, pylsp, gopls, rust-analyzer, etc.
+- **Custom servers**: Add via `lsp` config in oh-my-opencode.json
 
 ## AST-GREP SPECIFICS
 
-- Meta-variables: `$VAR` (single), `$$$` (multiple)
-- Pattern must be valid AST node, not fragment
-- Prefers napi binding for performance
+- **Meta-variables**: `$VAR` (single node), `$$$` (multiple nodes)
+- **Languages**: 25 supported (typescript, tsx, python, rust, go, etc.)
+- **Binding**: Prefers @ast-grep/napi (native), falls back to @ast-grep/cli
+- **Pattern must be valid AST**: `export async function $NAME($$$) { $$$ }` not fragments
 
-## ANTI-PATTERNS
+## ANTI-PATTERNS (TOOLS)
 
-- No timeout on file ops (always use, default 60s)
-- Sync file operations (use async/await)
-- Ignoring LSP errors (graceful handling required)
-- Raw subprocess for ast-grep (prefer napi)
+- **No timeout**: Always use timeout for file operations (default 60s)
+- **Blocking main thread**: Use async/await, never sync file ops
+- **Ignoring LSP errors**: Gracefully handle server not found/crashed
+- **Raw subprocess for ast-grep**: Prefer napi binding for performance

src/tools/background-task/index.ts

@@ -1,5 +1,4 @@
 export {
-  createBackgroundTask,
   createBackgroundOutput,
   createBackgroundCancel,
 } from "./tools"

src/tools/background-task/tools.ts

@@ -74,6 +74,7 @@ export function createBackgroundTask(manager: BackgroundManager): ToolDefinition
           parentSessionID: ctx.sessionID,
           parentMessageID: ctx.messageID,
           parentModel,
+          parentAgent: ctx.agent ?? prevMessage?.agent,
         })
 
         ctx.metadata?.({
@@ -175,8 +176,13 @@ async function formatTaskResult(task: BackgroundTask, client: OpencodeClient): P
   // Handle both SDK response structures: direct array or wrapped in .data
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
   const messages = ((messagesResult as any).data ?? messagesResult) as Array<{
-    info?: { role?: string }
-    parts?: Array<{ type?: string; text?: string }>
+    info?: { role?: string; time?: string }
+    parts?: Array<{ 
+      type?: string
+      text?: string
+      content?: string | Array<{ type: string; text?: string }>
+      name?: string
+    }>
   }>
 
   if (!Array.isArray(messages) || messages.length === 0) {
@@ -192,11 +198,13 @@ Session ID: ${task.sessionID}
 (No messages found)`
   }
 
-  const assistantMessages = messages.filter(
-    (m) => m.info?.role === "assistant"
+  // Include both assistant messages AND tool messages
+  // Tool results (grep, glob, bash output) come from role "tool"
+  const relevantMessages = messages.filter(
+    (m) => m.info?.role === "assistant" || m.info?.role === "tool"
   )
 
-  if (assistantMessages.length === 0) {
+  if (relevantMessages.length === 0) {
     return `Task Result
 
 Task ID: ${task.id}
@@ -206,17 +214,46 @@ Session ID: ${task.sessionID}
 
 ---
 
-(No assistant response found)`
+(No assistant or tool response found)`
   }
 
-  const lastMessage = assistantMessages[assistantMessages.length - 1]
-  const textParts = lastMessage?.parts?.filter(
-    (p) => p.type === "text"
-  ) ?? []
-  const textContent = textParts
-    .map((p) => p.text ?? "")
+  // Sort by time ascending (oldest first) to process messages in order
+  const sortedMessages = [...relevantMessages].sort((a, b) => {
+    const timeA = String((a as { info?: { time?: string } }).info?.time ?? "")
+    const timeB = String((b as { info?: { time?: string } }).info?.time ?? "")
+    return timeA.localeCompare(timeB)
+  })
+  
+  // Extract content from ALL messages, not just the last one
+  // Tool results may be in earlier messages while the final message is empty
+  const extractedContent: string[] = []
+  
+  for (const message of sortedMessages) {
+    for (const part of message.parts ?? []) {
+      // Handle both "text" and "reasoning" parts (thinking models use "reasoning")
+      if ((part.type === "text" || part.type === "reasoning") && part.text) {
+        extractedContent.push(part.text)
+      } else if (part.type === "tool_result") {
+        // Tool results contain the actual output from tool calls
+        const toolResult = part as { content?: string | Array<{ type: string; text?: string }> }
+        if (typeof toolResult.content === "string" && toolResult.content) {
+          extractedContent.push(toolResult.content)
+        } else if (Array.isArray(toolResult.content)) {
+          // Handle array of content blocks
+          for (const block of toolResult.content) {
+            // Handle both "text" and "reasoning" parts (thinking models use "reasoning")
+            if ((block.type === "text" || block.type === "reasoning") && block.text) {
+              extractedContent.push(block.text)
+            }
+          }
+        }
+      }
+    }
+  }
+  
+  const textContent = extractedContent
     .filter((text) => text.length > 0)
-    .join("\n")
+    .join("\n\n")
 
   const duration = formatDuration(task.startedAt, task.completedAt)
 

src/tools/call-omo-agent/constants.ts

@@ -4,4 +4,4 @@ export const CALL_OMO_AGENT_DESCRIPTION = `Spawn explore/librarian agent. run_in
 
 Available: {agents}
 
-Prompts MUST be in English. Use \`background_output\` for async results.`
+Pass \`resume=session_id\` to continue previous agent with full context. Prompts MUST be in English. Use \`background_output\` for async results.`

src/tools/call-omo-agent/tools.ts

@@ -142,7 +142,7 @@ async function executeSync(
         tools: {
           task: false,
           call_omo_agent: false,
-          background_task: false,
+          sisyphus_task: false,
         },
         parts: [{ type: "text", text: args.prompt }],
       },
@@ -170,23 +170,59 @@ async function executeSync(
   const messages = messagesResult.data
   log(`[call_omo_agent] Got ${messages.length} messages`)
 
+  // Include both assistant messages AND tool messages
+  // Tool results (grep, glob, bash output) come from role "tool"
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  const lastAssistantMessage = messages
-    .filter((m: any) => m.info.role === "assistant")
-    .sort((a: any, b: any) => (b.info.time?.created || 0) - (a.info.time?.created || 0))[0]
+  const relevantMessages = messages.filter(
+    (m: any) => m.info?.role === "assistant" || m.info?.role === "tool"
+  )
 
-  if (!lastAssistantMessage) {
-    log(`[call_omo_agent] No assistant message found`)
+  if (relevantMessages.length === 0) {
+    log(`[call_omo_agent] No assistant or tool messages found`)
     log(`[call_omo_agent] All messages:`, JSON.stringify(messages, null, 2))
-    return `Error: No assistant response found\n\n<task_metadata>\nsession_id: ${sessionID}\n</task_metadata>`
+    return `Error: No assistant or tool response found\n\n<task_metadata>\nsession_id: ${sessionID}\n</task_metadata>`
   }
 
-  log(`[call_omo_agent] Found assistant message with ${lastAssistantMessage.parts.length} parts`)
+  log(`[call_omo_agent] Found ${relevantMessages.length} relevant messages`)
 
+  // Sort by time ascending (oldest first) to process messages in order
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  const textParts = lastAssistantMessage.parts.filter((p: any) => p.type === "text")
-  // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  const responseText = textParts.map((p: any) => p.text).join("\n")
+  const sortedMessages = [...relevantMessages].sort((a: any, b: any) => {
+    const timeA = a.info?.time?.created ?? 0
+    const timeB = b.info?.time?.created ?? 0
+    return timeA - timeB
+  })
+
+  // Extract content from ALL messages, not just the last one
+  // Tool results may be in earlier messages while the final message is empty
+  const extractedContent: string[] = []
+
+  for (const message of sortedMessages) {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    for (const part of (message as any).parts ?? []) {
+      // Handle both "text" and "reasoning" parts (thinking models use "reasoning")
+      if ((part.type === "text" || part.type === "reasoning") && part.text) {
+        extractedContent.push(part.text)
+      } else if (part.type === "tool_result") {
+        // Tool results contain the actual output from tool calls
+        const toolResult = part as { content?: string | Array<{ type: string; text?: string }> }
+        if (typeof toolResult.content === "string" && toolResult.content) {
+          extractedContent.push(toolResult.content)
+        } else if (Array.isArray(toolResult.content)) {
+          // Handle array of content blocks
+          for (const block of toolResult.content) {
+            if ((block.type === "text" || block.type === "reasoning") && block.text) {
+              extractedContent.push(block.text)
+            }
+          }
+        }
+      }
+    }
+  }
+
+  const responseText = extractedContent
+    .filter((text) => text.length > 0)
+    .join("\n\n")
 
   log(`[call_omo_agent] Got response, length: ${responseText.length}`)