release: v3.10.0

Hashline edit tool and companion hooks now require explicit opt-in
via `"hashline_edit": true` in config. Previously enabled by default.

- tool-registry: hashline edit tool not registered unless opted in
- create-tool-guard-hooks: hashline-read-enhancer disabled by default
- Updated config schema comment and documentation
- Added TDD tests for default behavior

.github/workflows/publish-platform.yml

@@ -35,15 +35,15 @@ jobs:
   # - Uploads compressed artifacts for the publish job
   # =============================================================================
   build:
-    runs-on: ${{ matrix.platform == 'windows-x64' && 'windows-latest' || 'ubuntu-latest' }}
+    runs-on: ${{ startsWith(matrix.platform, 'windows-') && 'windows-latest' || 'ubuntu-latest' }}
     defaults:
       run:
         shell: bash
     strategy:
       fail-fast: false
-      max-parallel: 7
+      max-parallel: 11
       matrix:
-        platform: [darwin-arm64, darwin-x64, linux-x64, linux-arm64, linux-x64-musl, linux-arm64-musl, windows-x64]
+        platform: [darwin-arm64, darwin-x64, darwin-x64-baseline, linux-x64, linux-x64-baseline, linux-arm64, linux-x64-musl, linux-x64-musl-baseline, linux-arm64-musl, windows-x64, windows-x64-baseline]
     steps:
       - uses: actions/checkout@v4
 
@@ -82,6 +82,52 @@ jobs:
           cd packages/${{ matrix.platform }}
           jq --arg v "$VERSION" '.version = $v' package.json > tmp.json && mv tmp.json package.json
 
+      - name: Pre-download baseline compile target
+        if: steps.check.outputs.skip != 'true' && endsWith(matrix.platform, '-baseline')
+        shell: bash
+        run: |
+          BUN_VERSION=$(bun --version)
+          PLATFORM="${{ matrix.platform }}"
+          PKG_NAME="bun-${PLATFORM}"
+          CACHE_DIR=$(bun pm cache)
+          CACHE_DEST="${CACHE_DIR}/${PKG_NAME}-v${BUN_VERSION}"
+          
+          if [[ -f "$CACHE_DEST" ]]; then
+            echo "✓ Compile target already cached at ${CACHE_DEST}"
+            exit 0
+          fi
+          
+          echo "Pre-downloading ${PKG_NAME} v${BUN_VERSION} to ${CACHE_DEST}"
+          TARBALL_URL="https://registry.npmjs.org/@oven/bun-${PLATFORM}/-/bun-${PLATFORM}-${BUN_VERSION}.tgz"
+          echo "URL: ${TARBALL_URL}"
+          
+          mkdir -p "$(dirname "$CACHE_DEST")"
+          TMP_DIR=$(mktemp -d)
+          
+          # Download and extract the bun binary from npm tarball
+          curl -fsSL --retry 5 --retry-delay 5 "${TARBALL_URL}" | tar -xzf - -C "${TMP_DIR}"
+          
+          if [[ "$PLATFORM" == windows-* ]]; then
+            BIN_NAME="bun.exe"
+          else
+            BIN_NAME="bun"
+          fi
+          
+          # npm tarball has package/bin/bun structure
+          if [[ -f "${TMP_DIR}/package/bin/${BIN_NAME}" ]]; then
+            cp "${TMP_DIR}/package/bin/${BIN_NAME}" "${CACHE_DEST}"
+          elif [[ -f "${TMP_DIR}/package/${BIN_NAME}" ]]; then
+            cp "${TMP_DIR}/package/${BIN_NAME}" "${CACHE_DEST}"
+          else
+            echo "Could not find ${BIN_NAME} in tarball, listing contents:"
+            find "${TMP_DIR}" -type f
+            exit 1
+          fi
+          
+          chmod +x "${CACHE_DEST}" 2>/dev/null || true
+          echo "✓ Pre-downloaded to ${CACHE_DEST}"
+          ls -lh "${CACHE_DEST}"
+
       - name: Build binary
         if: steps.check.outputs.skip != 'true'
         uses: nick-fields/retry@v3
@@ -95,14 +141,18 @@ jobs:
             case "$PLATFORM" in
               darwin-arm64) TARGET="bun-darwin-arm64" ;;
               darwin-x64) TARGET="bun-darwin-x64" ;;
+              darwin-x64-baseline) TARGET="bun-darwin-x64-baseline" ;;
               linux-x64) TARGET="bun-linux-x64" ;;
+              linux-x64-baseline) TARGET="bun-linux-x64-baseline" ;;
               linux-arm64) TARGET="bun-linux-arm64" ;;
               linux-x64-musl) TARGET="bun-linux-x64-musl" ;;
+              linux-x64-musl-baseline) TARGET="bun-linux-x64-musl-baseline" ;;
               linux-arm64-musl) TARGET="bun-linux-arm64-musl" ;;
               windows-x64) TARGET="bun-windows-x64" ;;
+              windows-x64-baseline) TARGET="bun-windows-x64-baseline" ;;
             esac
             
-            if [ "$PLATFORM" = "windows-x64" ]; then
+            if [[ "$PLATFORM" == windows-* ]]; then
               OUTPUT="packages/${PLATFORM}/bin/oh-my-opencode.exe"
             else
               OUTPUT="packages/${PLATFORM}/bin/oh-my-opencode"
@@ -119,7 +169,7 @@ jobs:
           PLATFORM="${{ matrix.platform }}"
           cd packages/${PLATFORM}
           
-          if [ "$PLATFORM" = "windows-x64" ]; then
+          if [[ "$PLATFORM" == windows-* ]]; then
             # Windows: use 7z (pre-installed on windows-latest)
             7z a -tzip ../../binary-${PLATFORM}.zip bin/ package.json
           else
@@ -150,12 +200,13 @@ jobs:
   # =============================================================================
   publish:
     needs: build
+    if: always() && !cancelled()
     runs-on: ubuntu-latest
     strategy:
       fail-fast: false
       max-parallel: 2
       matrix:
-        platform: [darwin-arm64, darwin-x64, linux-x64, linux-arm64, linux-x64-musl, linux-arm64-musl, windows-x64]
+        platform: [darwin-arm64, darwin-x64, darwin-x64-baseline, linux-x64, linux-x64-baseline, linux-arm64, linux-x64-musl, linux-x64-musl-baseline, linux-arm64-musl, windows-x64, windows-x64-baseline]
     steps:
       - name: Check if already published
         id: check
@@ -172,19 +223,21 @@ jobs:
           fi
 
       - name: Download artifact
+        id: download
         if: steps.check.outputs.skip != 'true'
+        continue-on-error: true
         uses: actions/download-artifact@v4
         with:
           name: binary-${{ matrix.platform }}
           path: .
 
       - name: Extract artifact
-        if: steps.check.outputs.skip != 'true'
+        if: steps.check.outputs.skip != 'true' && steps.download.outcome == 'success'
         run: |
           PLATFORM="${{ matrix.platform }}"
           mkdir -p packages/${PLATFORM}
           
-          if [ "$PLATFORM" = "windows-x64" ]; then
+          if [[ "$PLATFORM" == windows-* ]]; then
             unzip binary-${PLATFORM}.zip -d packages/${PLATFORM}/
           else
             tar -xzvf binary-${PLATFORM}.tar.gz -C packages/${PLATFORM}/
@@ -195,13 +248,13 @@ jobs:
           ls -la packages/${PLATFORM}/bin/
 
       - uses: actions/setup-node@v4
-        if: steps.check.outputs.skip != 'true'
+        if: steps.check.outputs.skip != 'true' && steps.download.outcome == 'success'
         with:
           node-version: "24"
           registry-url: "https://registry.npmjs.org"
 
       - name: Publish ${{ matrix.platform }}
-        if: steps.check.outputs.skip != 'true'
+        if: steps.check.outputs.skip != 'true' && steps.download.outcome == 'success'
         run: |
           cd packages/${{ matrix.platform }}
           

.github/workflows/publish.yml

@@ -189,7 +189,7 @@ jobs:
           VERSION="${{ steps.version.outputs.version }}"
           jq --arg v "$VERSION" '.version = $v' package.json > tmp.json && mv tmp.json package.json
           
-          for platform in darwin-arm64 darwin-x64 linux-x64 linux-arm64 linux-x64-musl linux-arm64-musl windows-x64; do
+          for platform in darwin-arm64 darwin-x64 darwin-x64-baseline linux-x64 linux-x64-baseline linux-arm64 linux-x64-musl linux-x64-musl-baseline linux-arm64-musl windows-x64 windows-x64-baseline; do
             jq --arg v "$VERSION" '.version = $v' "packages/${platform}/package.json" > tmp.json
             mv tmp.json "packages/${platform}/package.json"
           done

AGENTS.md

@@ -122,7 +122,7 @@ bunx oh-my-opencode run     # Non-interactive session
 | Workflow | Trigger | Purpose |
 |----------|---------|---------|
 | ci.yml | push/PR | Tests (split: mock-heavy isolated + batch), typecheck, build, schema auto-commit |
-| publish.yml | manual | Version bump, npm publish, platform binaries, GitHub release, merge to master |
+| publish.yml | manual | Version bump, npm publish, platform binaries, GitHub release, merge to dev |
 | publish-platform.yml | called | 11 platform binaries via bun compile (darwin/linux/windows) |
 | sisyphus-agent.yml | @mention | AI agent handles issues/PRs |
 

README.ja.md

@@ -44,7 +44,7 @@
 [![GitHub Forks](https://img.shields.io/github/forks/code-yeongyu/oh-my-opencode?color=8ae8ff&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/network/members)
 [![GitHub Stars](https://img.shields.io/github/stars/code-yeongyu/oh-my-opencode?color=ffcb47&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/stargazers)
 [![GitHub Issues](https://img.shields.io/github/issues/code-yeongyu/oh-my-opencode?color=ff80eb&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/issues)
-[![License](https://img.shields.io/badge/license-SUL--1.0-white?labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/blob/master/LICENSE.md)
+[![License](https://img.shields.io/badge/license-SUL--1.0-white?labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/blob/dev/LICENSE.md)
 [![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/code-yeongyu/oh-my-opencode)
 
 [English](README.md) | [한국어](README.ko.md) | [日本語](README.ja.md) | [简体中文](README.zh-cn.md)
@@ -96,7 +96,7 @@ OmOをインストールして、`ultrawork`とタイプしてください。狂
 
 ```
 Install and configure oh-my-opencode by following the instructions here:
-https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 ```
 
 もしくは[インストールガイド](docs/guide/installation.md)を直接読んでもいいですが、マジでエージェントにやらせてください。人間は設定で必ずタイポします。
@@ -106,7 +106,7 @@ https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/
 インストールガイドを取得して、それに従ってください：
 
 ```bash
-curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 ```
 
 ---

README.ko.md

@@ -44,7 +44,7 @@
 [![GitHub Forks](https://img.shields.io/github/forks/code-yeongyu/oh-my-opencode?color=8ae8ff&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/network/members)
 [![GitHub Stars](https://img.shields.io/github/stars/code-yeongyu/oh-my-opencode?color=ffcb47&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/stargazers)
 [![GitHub Issues](https://img.shields.io/github/issues/code-yeongyu/oh-my-opencode?color=ff80eb&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/issues)
-[![License](https://img.shields.io/badge/license-SUL--1.0-white?labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/blob/master/LICENSE.md)
+[![License](https://img.shields.io/badge/license-SUL--1.0-white?labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/blob/dev/LICENSE.md)
 [![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/code-yeongyu/oh-my-opencode)
 
 [English](README.md) | [한국어](README.ko.md) | [日本語](README.ja.md) | [简体中文](README.zh-cn.md)
@@ -95,7 +95,7 @@ OmO 설치하고. `ultrawork` 치세요. 끝.
 
 ```
 Install and configure oh-my-opencode by following the instructions here:
-https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 ```
 
 아니면 [설치 가이드](docs/guide/installation.md)를 직접 읽으셔도 되지만, 진심으로 그냥 에이전트한테 시키세요. 사람은 설정하다 꼭 오타 냅니다.
@@ -105,7 +105,7 @@ https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/
 설치 가이드를 가져와서 따라 하세요:
 
 ```bash
-curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 ```
 
 ---

README.md

@@ -47,7 +47,7 @@
 [![GitHub Forks](https://img.shields.io/github/forks/code-yeongyu/oh-my-opencode?color=8ae8ff&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/network/members)
 [![GitHub Stars](https://img.shields.io/github/stars/code-yeongyu/oh-my-opencode?color=ffcb47&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/stargazers)
 [![GitHub Issues](https://img.shields.io/github/issues/code-yeongyu/oh-my-opencode?color=ff80eb&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/issues)
-[![License](https://img.shields.io/badge/license-SUL--1.0-white?labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/blob/master/LICENSE.md)
+[![License](https://img.shields.io/badge/license-SUL--1.0-white?labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/blob/dev/LICENSE.md)
 [![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/code-yeongyu/oh-my-opencode)
 
 [English](README.md) | [한국어](README.ko.md) | [日本語](README.ja.md) | [简体中文](README.zh-cn.md)
@@ -99,7 +99,7 @@ Copy and paste this prompt to your LLM agent (Claude Code, AmpCode, Cursor, etc.
 
 ```
 Install and configure oh-my-opencode by following the instructions here:
-https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 ```
 
 Or read the [Installation Guide](docs/guide/installation.md), but seriously, let an agent do it. Humans fat-finger configs.
@@ -109,7 +109,7 @@ Or read the [Installation Guide](docs/guide/installation.md), but seriously, let
 Fetch the installation guide and follow it:
 
 ```bash
-curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 ```
 
 ---

README.ru.md

@@ -0,0 +1,367 @@
+> [!WARNING] **Предупреждение о безопасности: сайт-имитатор**
+>
+> **ohmyopencode.com НЕ аффилирован с этим проектом.** Мы не управляем этим сайтом и не одобряем его.
+>
+> OhMyOpenCode — **бесплатный и открытый проект**. Не скачивайте установщики и не вводите платёжные данные на сторонних сайтах, которые называют себя «официальными».
+>
+> Поскольку сайт-имитатор находится за платным доступом, мы **не можем проверить, что именно он распространяет**. Относитесь к любым загрузкам с него как к **потенциально небезопасным**.
+>
+> ✅ Официальные загрузки: https://github.com/code-yeongyu/oh-my-opencode/releases
+
+> [!NOTE]
+>
+> [![Sisyphus Labs - Sisyphus is the agent that codes like your team.](./.github/assets/sisyphuslabs.png?v=2)](https://sisyphuslabs.ai)
+>
+> > **Мы создаём полноценную продуктовую версию Sisyphus, чтобы задать стандарты для frontier-агентов. <br />Присоединяйтесь к листу ожидания [здесь](https://sisyphuslabs.ai).**
+
+> [!TIP] Будьте с нами!
+>
+> | [](https://discord.gg/PUwSMR9XNk)   | Вступайте в наш [Discord](https://discord.gg/PUwSMR9XNk), чтобы общаться с контрибьюторами и пользователями `oh-my-opencode`. |
+> | ----------------------------------- | ------------------------------------------------------------ |
+> | [](https://x.com/justsisyphus)      | Новости и обновления `oh-my-opencode` раньше публиковались на моём аккаунте X. <br /> После ошибочной блокировки, [@justsisyphus](https://x.com/justsisyphus) публикует обновления вместо меня. |
+> | [](https://github.com/code-yeongyu) | Подпишитесь на [@code-yeongyu](https://github.com/code-yeongyu) на GitHub, чтобы следить за другими проектами. |
+
+<!-- <CENTERED SECTION FOR GITHUB DISPLAY> --> <div align="center">
+
+[![Oh My OpenCode](./.github/assets/hero.jpg)](https://github.com/code-yeongyu/oh-my-opencode#oh-my-opencode)
+
+[![Preview](./.github/assets/omo.png)](https://github.com/code-yeongyu/oh-my-opencode#oh-my-opencode)
+
+</div>
+
+> Anthropic [**заблокировал OpenCode из-за нас.**](https://x.com/thdxr/status/2010149530486911014) **Да, это правда.** Они хотят держать вас в замкнутой системе. Claude Code — красивая тюрьма, но всё равно тюрьма.
+>
+> Мы не делаем привязки. Мы работаем с любыми моделями. Claude / Kimi / GLM для оркестрации. GPT для рассуждений. Minimax для скорости. Gemini для творческих задач. Будущее — не в выборе одного победителя, а в оркестровке всех. Модели дешевеют каждый месяц. Умнеют каждый месяц. Ни один провайдер не будет доминировать. Мы строим под открытый рынок, а не под чьи-то огороженные сады.
+
+<div align="center">
+
+[![GitHub Release](https://img.shields.io/github/v/release/code-yeongyu/oh-my-opencode?color=369eff&labelColor=black&logo=github&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/releases) [![npm downloads](https://img.shields.io/npm/dt/oh-my-opencode?color=ff6b35&labelColor=black&style=flat-square)](https://www.npmjs.com/package/oh-my-opencode) [![GitHub Contributors](https://img.shields.io/github/contributors/code-yeongyu/oh-my-opencode?color=c4f042&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/graphs/contributors) [![GitHub Forks](https://img.shields.io/github/forks/code-yeongyu/oh-my-opencode?color=8ae8ff&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/network/members) [![GitHub Stars](https://img.shields.io/github/stars/code-yeongyu/oh-my-opencode?color=ffcb47&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/stargazers) [![GitHub Issues](https://img.shields.io/github/issues/code-yeongyu/oh-my-opencode?color=ff80eb&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/issues) [![License](https://img.shields.io/badge/license-SUL--1.0-white?labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/blob/master/LICENSE.md) [![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/code-yeongyu/oh-my-opencode)
+
+English | 한국어 | 日本語 | 简体中文 | Русский
+
+</div> <!-- </CENTERED SECTION FOR GITHUB DISPLAY> -->
+
+## Отзывы
+
+> «Из-за него я отменил подписку на Cursor. В опенсорс-сообществе происходит что-то невероятное.» — [Arthur Guiot](https://x.com/arthur_guiot/status/2008736347092382053?s=20)
+
+> «Если Claude Code делает за 7 дней то, на что у человека уходит 3 месяца, Sisyphus справляется за 1 час. Он просто работает, пока задача не выполнена. Это дисциплинированный агент.» <br/>— B, исследователь в области квантовых финансов
+
+> «За один день устранил 8000 предупреждений eslint с помощью Oh My Opencode.» <br/>— [Jacob Ferrari](https://x.com/jacobferrari_/status/2003258761952289061)
+
+> «За ночь конвертировал приложение на tauri в 45k строк в веб-SaaS с помощью Ohmyopencode и ralph loop. Начал с промпта «проинтервьюируй меня», попросил оценки и рекомендации по вопросам. Было удивительно наблюдать за работой и утром проснуться с почти рабочим сайтом!» — [James Hargis](https://x.com/hargabyte/status/2007299688261882202)
+
+> «Используйте oh-my-opencode — вы не захотите возвращаться назад.» <br/>— [d0t3ch](https://x.com/d0t3ch/status/2001685618200580503)
+
+> «Пока не могу точно объяснить, почему это так круто, но опыт разработки вышел на совершенно другой уровень.» — [苔硯:こけすずり](https://x.com/kokesuzuri/status/2008532913961529372?s=20)
+
+> «Экспериментирую с open code, oh my opencode и supermemory этим выходным, чтобы собрать нечто среднее между Minecraft и souls-like.» «Попросил добавить анимации приседания, пока хожу на обеденную прогулку. [Видео]» — [MagiMetal](https://x.com/MagiMetal/status/2005374704178373023)
+
+> «Ребята, вам нужно включить это в ядро и нанять его. Серьёзно. Это очень, очень, очень хорошо.» <br/>— Henning Kilset
+
+> «Наймите @yeon_gyu_kim, если сможете его уговорить, этот парень революционизировал opencode.» <br/>— [mysticaltech](https://x.com/mysticaltech/status/2001858758608376079)
+
+> «Oh My OpenCode — это что-то с чем-то» — [YouTube — Darren Builds AI](https://www.youtube.com/watch?v=G_Snfh2M41M)
+
+------
+
+# Oh My OpenCode
+
+Вы жонглируете Claude Code, Codex, случайными OSS-моделями. Настраиваете рабочие процессы. Дебажите агентов.
+
+Мы уже проделали эту работу. Протестировали всё. Оставили только то, что реально работает.
+
+Установите OmO. Введите `ultrawork`. Готово.
+
+## Установка
+
+### Для людей
+
+Скопируйте и вставьте этот промпт в ваш LLM-агент (Claude Code, AmpCode, Cursor и т.д.):
+
+```
+Install and configure oh-my-opencode by following the instructions here:
+https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+```
+
+Или прочитайте руководство по установке, но серьёзно — пусть агент сделает это за вас. Люди ошибаются в конфигах.
+
+### Для LLM-агентов
+
+Загрузите руководство по установке и следуйте ему:
+
+```bash
+curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+```
+
+------
+
+## Пропустите этот README
+
+Мы вышли за пределы эпохи чтения документации. Просто вставьте это в своего агента:
+
+```
+Read this and tell me why it's not just another boilerplate: https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/README.md
+```
+
+## Ключевые возможности
+
+### 🪄 `ultrawork`
+
+Вы правда это читаете? Поразительно.
+
+Установите. Введите `ultrawork` (или `ulw`). Готово.
+
+Всё описанное ниже, каждая функция, каждая оптимизация — вам не нужно это знать. Оно просто работает.
+
+Даже при наличии только следующих подписок ultrawork будет работать отлично (проект не аффилирован с ними, это личная рекомендация):
+
+- [Подписка ChatGPT ($20)](https://chatgpt.com/)
+- [Подписка Kimi Code ($0.99) (*только в этом месяце)](https://www.kimi.com/membership/pricing?track_id=5cdeca93-66f0-4d35-aabb-b6df8fcea328)
+- [Тариф GLM Coding ($10)](https://z.ai/subscribe)
+- При доступе к оплате за токены использование моделей Kimi и Gemini обойдётся недорого.
+
+|      | Функция                                                  | Что делает                                                   |
+| ---- | -------------------------------------------------------- | ------------------------------------------------------------ |
+| 🤖    | **Дисциплинированные агенты**                            | Sisyphus оркестрирует Hephaestus, Oracle, Librarian, Explore. Полноценная AI-команда разработки в параллельном режиме. |
+| ⚡    | **`ultrawork` / `ulw`**                                  | Одно слово. Все агенты активируются. Не останавливается, пока задача не выполнена. |
+| 🚪    | **[IntentGate](https://factory.ai/news/terminal-bench)** | Анализирует истинное намерение пользователя перед классификацией и действием. Никакого буквального неверного толкования. |
+| 🔗    | **Инструмент правок на основе хэш-якорей**               | Хэш содержимого `LINE#ID` проверяет каждое изменение. Ноль ошибок с устаревшими строками. Вдохновлено [oh-my-pi](https://github.com/can1357/oh-my-pi). [Проблема обвязки →](https://blog.can.ac/2026/02/12/the-harness-problem/) |
+| 🛠️    | **LSP + AST-Grep**                                       | Переименование в рабочем пространстве, диагностика перед сборкой, переписывание с учётом AST. Точность IDE для агентов. |
+| 🧠    | **Фоновые агенты**                                       | Запускайте 5+ специалистов параллельно. Контекст остаётся компактным. Результаты — когда готовы. |
+| 📚    | **Встроенные MCP**                                       | Exa (веб-поиск), Context7 (официальная документация), Grep.app (поиск по GitHub). Всегда включены. |
+| 🔁    | **Ralph Loop / `/ulw-loop`**                             | Самореферентный цикл. Не останавливается, пока задача не выполнена на 100%. |
+| ✅    | **Todo Enforcer**                                        | Агент завис? Система немедленно возвращает его в работу. Ваша задача будет выполнена, точка. |
+| 💬    | **Comment Checker**                                      | Никакого AI-мусора в комментариях. Код читается так, словно его писал опытный разработчик. |
+| 🖥️    | **Интеграция с Tmux**                                    | Полноценный интерактивный терминал. REPL, дебаггеры, TUI. Всё живое. |
+| 🔌    | **Совместимость с Claude Code**                          | Ваши хуки, команды, навыки, MCP и плагины? Всё работает без изменений. |
+| 🎯    | **MCP, встроенные в навыки**                             | Навыки несут собственные MCP-серверы. Никакого раздувания контекста. |
+| 📋    | **Prometheus Planner**                                   | Стратегическое планирование в режиме интервью перед любым выполнением. |
+| 🔍    | **`/init-deep`**                                         | Автоматически генерирует иерархические файлы `AGENTS.md` по всему проекту. Отлично работает на эффективность токенов и производительность агента. |
+
+### Дисциплинированные агенты
+
+<table><tr> <td align="center"><img src=".github/assets/sisyphus.png" height="300" /></td> <td align="center"><img src=".github/assets/hephaestus.png" height="300" /></td> </tr></table>
+
+**Sisyphus** (`claude-opus-4-6` / **`kimi-k2.5`** / **`glm-5`**) — главный оркестратор. Он планирует, делегирует задачи специалистам и доводит их до завершения с агрессивным параллельным выполнением. Он не останавливается на полпути.
+
+**Hephaestus** (`gpt-5.3-codex`) — автономный глубокий исполнитель. Дайте ему цель, а не рецепт. Он исследует кодовую базу, изучает паттерны и выполняет задачи сквозным образом без лишних подсказок. *Законный Мастер.*
+
+**Prometheus** (`claude-opus-4-6` / **`kimi-k2.5`** / **`glm-5`**) — стратегический планировщик. Режим интервью: задаёт вопросы, определяет объём работ и формирует детальный план до того, как написана хотя бы одна строка кода.
+
+Каждый агент настроен под сильные стороны своей модели. Никакого ручного переключения между моделями. Подробнее →
+
+> Anthropic [заблокировал OpenCode из-за нас.](https://x.com/thdxr/status/2010149530486911014) Именно поэтому Hephaestus зовётся «Законным Мастером». Ирония намеренная.
+>
+> Мы работаем лучше всего на Opus, но Kimi K2.5 + GPT-5.3 Codex уже превосходят ванильный Claude Code. Никакой настройки не требуется.
+
+### Оркестрация агентов
+
+Когда Sisyphus делегирует задачу субагенту, он выбирает не модель, а **категорию**. Категория автоматически сопоставляется с нужной моделью:
+
+| Категория            | Для чего предназначена                |
+| -------------------- | ------------------------------------- |
+| `visual-engineering` | Фронтенд, UI/UX, дизайн               |
+| `deep`               | Автономные исследования + выполнение  |
+| `quick`              | Изменения в одном файле, опечатки     |
+| `ultrabrain`         | Сложная логика, архитектурные решения |
+
+Агент сообщает тип задачи. Обвязка подбирает нужную модель. Вы ни к чему не прикасаетесь.
+
+### Совместимость с Claude Code
+
+Вы тщательно настроили Claude Code. Хорошо.
+
+Каждый хук, команда, навык, MCP и плагин работают здесь без изменений. Полная совместимость, включая плагины.
+
+### Инструменты мирового класса для ваших агентов
+
+LSP, AST-Grep, Tmux, MCP — реально интегрированы, а не склеены скотчем.
+
+- **LSP**: `lsp_rename`, `lsp_goto_definition`, `lsp_find_references`, `lsp_diagnostics`. Точность IDE для каждого агента
+- **AST-Grep**: Поиск и переписывание кода с учётом синтаксических паттернов для 25 языков
+- **Tmux**: Полноценный интерактивный терминал. REPL, дебаггеры, TUI-приложения. Агент остаётся в сессии
+- **MCP**: Веб-поиск, официальная документация, поиск по коду на GitHub. Всё встроено
+
+### MCP, встроенные в навыки
+
+MCP-серверы съедают бюджет контекста. Мы это исправили.
+
+Навыки приносят собственные MCP-серверы. Запускаются по необходимости, ограничены задачей, исчезают по завершении. Контекстное окно остаётся чистым.
+
+### Лучше пишет код. Правки на основе хэш-якорей
+
+Проблема обвязки реальна. Большинство сбоев агентов — не вина модели. Это вина инструмента правок.
+
+> *«Ни один из этих инструментов не даёт модели стабильный, проверяемый идентификатор строк, которые она хочет изменить... Все они полагаются на то, что модель воспроизведёт контент, который уже видела. Когда это не получается — а так бывает нередко — пользователь обвиняет модель.»*
+>
+> <br/>— [Can Bölük, «Проблема обвязки»](https://blog.can.ac/2026/02/12/the-harness-problem/)
+
+Вдохновлённые [oh-my-pi](https://github.com/can1357/oh-my-pi), мы реализовали **Hashline**. Каждая строка, которую читает агент, возвращается с тегом хэша содержимого:
+
+```
+11#VK| function hello() {
+22#XJ|   return "world";
+33#MB| }
+```
+
+Агент редактирует, ссылаясь на эти теги. Если файл изменился с момента последнего чтения, хэш не совпадёт, и правка будет отклонена до любого повреждения. Никакого воспроизведения пробелов. Никаких ошибок с устаревшими строками.
+
+Grok Code Fast 1: успешность **6.7% → 68.3%**. Просто за счёт замены инструмента правок.
+
+### Глубокая инициализация. `/init-deep`
+
+Запустите `/init-deep`. Будут сгенерированы иерархические файлы `AGENTS.md`:
+
+```
+project/
+├── AGENTS.md              ← контекст всего проекта
+├── src/
+│   ├── AGENTS.md          ← контекст для src
+│   └── components/
+│       └── AGENTS.md      ← контекст для компонентов
+```
+
+Агенты автоматически читают нужный контекст. Никакого ручного управления.
+
+### Планирование. Prometheus
+
+Сложная задача? Не нужно молиться и надеяться на промпт.
+
+`/start-work` вызывает Prometheus. **Интервьюирует вас как настоящий инженер**, определяет объём работ и неоднозначности, формирует проверенный план до прикосновения к коду. Агент знает, что строит, прежде чем начать.
+
+### Навыки
+
+Навыки — это не просто промпты. Каждый привносит:
+
+- Системные инструкции, настроенные под предметную область
+- Встроенные MCP-серверы, запускаемые по необходимости
+- Ограниченные разрешения. Агенты остаются в рамках
+
+Встроенные: `playwright` (автоматизация браузера), `git-master` (атомарные коммиты, хирургия rebase), `frontend-ui-ux` (UI с упором на дизайн).
+
+Добавьте свои: `.opencode/skills/*/SKILL.md` или `~/.config/opencode/skills/*/SKILL.md`.
+
+**Хотите полное описание возможностей?** Смотрите **документацию по функциям** — агенты, хуки, инструменты, MCP и всё остальное подробно.
+
+------
+
+> **Впервые в oh-my-opencode?** Прочитайте **Обзор**, чтобы понять, что у вас есть, или ознакомьтесь с **руководством по оркестрации**, чтобы узнать, как агенты взаимодействуют.
+
+## Удаление
+
+Чтобы удалить oh-my-opencode:
+
+1. **Удалите плагин из конфига OpenCode**
+
+   Отредактируйте `~/.config/opencode/opencode.json` (или `opencode.jsonc`) и уберите `"oh-my-opencode"` из массива `plugin`:
+
+   ```bash
+   # С помощью jq
+   jq '.plugin = [.plugin[] | select(. != "oh-my-opencode")]' \
+       ~/.config/opencode/opencode.json > /tmp/oc.json && \
+       mv /tmp/oc.json ~/.config/opencode/opencode.json
+   ```
+
+2. **Удалите файлы конфигурации (опционально)**
+
+   ```bash
+   # Удалить пользовательский конфиг
+   rm -f ~/.config/opencode/oh-my-opencode.json ~/.config/opencode/oh-my-opencode.jsonc
+   
+   # Удалить конфиг проекта (если существует)
+   rm -f .opencode/oh-my-opencode.json .opencode/oh-my-opencode.jsonc
+   ```
+
+3. **Проверьте удаление**
+
+   ```bash
+   opencode --version
+   # Плагин больше не должен загружаться
+   ```
+
+## Функции
+
+Функции, которые, как вы будете думать, должны были существовать всегда. Попробовав раз, вы не сможете вернуться назад.
+
+Смотрите полную документацию по функциям.
+
+**Краткий обзор:**
+
+- **Агенты**: Sisyphus (главный агент), Prometheus (планировщик), Oracle (архитектура/отладка), Librarian (документация/поиск по коду), Explore (быстрый grep по кодовой базе), Multimodal Looker
+- **Фоновые агенты**: Запускайте несколько агентов параллельно, как настоящая команда разработки
+- **Инструменты LSP и AST**: Рефакторинг, переименование, диагностика, поиск кода с учётом AST
+- **Инструмент правок на основе хэш-якорей**: Ссылки `LINE#ID` проверяют содержимое перед применением каждого изменения. Хирургические правки, ноль ошибок с устаревшими строками
+- **Инъекция контекста**: Автоматическое добавление AGENTS.md, README.md, условных правил
+- **Совместимость с Claude Code**: Полная система хуков, команды, навыки, агенты, MCP
+- **Встроенные MCP**: websearch (Exa), context7 (документация), grep_app (поиск по GitHub)
+- **Инструменты сессий**: Список, чтение, поиск и анализ истории сессий
+- **Инструменты продуктивности**: Ralph Loop, Todo Enforcer, Comment Checker, Think Mode и другое
+- **Настройка моделей**: Сопоставление агент–модель встроено в руководство по установке
+
+## Конфигурация
+
+Продуманные настройки по умолчанию, которые можно изменить при необходимости.
+
+Смотрите документацию по конфигурации.
+
+**Краткий обзор:**
+
+- **Расположение конфигов**: `.opencode/oh-my-opencode.jsonc` или `.opencode/oh-my-opencode.json` (проект), `~/.config/opencode/oh-my-opencode.jsonc` или `~/.config/opencode/oh-my-opencode.json` (пользователь)
+- **Поддержка JSONC**: Комментарии и конечные запятые поддерживаются
+- **Агенты**: Переопределение моделей, температур, промптов и разрешений для любого агента
+- **Встроенные навыки**: `playwright` (автоматизация браузера), `git-master` (атомарные коммиты)
+- **Агент Sisyphus**: Главный оркестратор с Prometheus (Планировщик) и Metis (Консультант по плану)
+- **Фоновые задачи**: Настройка ограничений параллельности по провайдеру/модели
+- **Категории**: Делегирование задач по предметной области (`visual`, `business-logic`, пользовательские)
+- **Хуки**: 25+ встроенных хуков, все настраиваются через `disabled_hooks`
+- **MCP**: Встроенные websearch (Exa), context7 (документация), grep_app (поиск по GitHub)
+- **LSP**: Полная поддержка LSP с инструментами рефакторинга
+- **Экспериментальное**: Агрессивное усечение, автовозобновление и другое
+
+## Слово автора
+
+**Хотите узнать философию?** Прочитайте Манифест Ultrawork.
+
+------
+
+Я потратил $24K на токены LLM в личных проектах. Попробовал все инструменты. Настраивал всё до смерти. OpenCode победил.
+
+Каждая проблема, с которой я столкнулся, — её решение уже встроено в этот плагин. Устанавливайте и работайте.
+
+Если OpenCode — это Debian/Arch, то OmO — это Ubuntu/[Omarchy](https://omarchy.org/).
+
+Сильное влияние со стороны [AmpCode](https://ampcode.com) и [Claude Code](https://code.claude.com/docs/overview). Функции портированы, часто улучшены. Продолжаем строить. Это **Open**Code.
+
+Другие обвязки обещают оркестрацию нескольких моделей. Мы её поставляем. Плюс стабильность. Плюс функции, которые реально работают.
+
+Я самый одержимый пользователь этого проекта:
+
+- Какая модель думает острее всего?
+- Кто бог отладки?
+- Кто пишет лучший код?
+- Кто рулит фронтендом?
+- Кто владеет бэкендом?
+- Что быстрее всего в ежедневной работе?
+- Что запускают конкуренты?
+
+Этот плагин — дистилляция. Берём лучшее. Есть улучшения? PR приветствуются.
+
+**Хватит мучиться с выбором обвязки.** **Я буду исследовать, воровать лучшее и поставлять это сюда.**
+
+Звучит высокомерно? Знаете, как сделать лучше? Контрибьютьте. Добро пожаловать.
+
+Никакой аффилиации с упомянутыми проектами/моделями. Только личные эксперименты.
+
+99% этого проекта было создано с помощью OpenCode. Я почти не знаю TypeScript. **Но эту документацию я лично просматривал и во многом переписывал.**
+
+## Любимый профессионалами из
+
+- Indent
+  - Spray — решение для influencer-маркетинга, vovushop — платформа кросс-граничной торговли, vreview — AI-решение для маркетинга отзывов в commerce
+- [Google](https://google.com)
+- [Microsoft](https://microsoft.com)
+- ELESTYLE
+  - elepay — мультимобильный платёжный шлюз, OneQR — мобильное SaaS-приложение для безналичных расчётов
+
+*Особая благодарность [@junhoyeo](https://github.com/junhoyeo) за это потрясающее hero-изображение.*

README.zh-cn.md

@@ -44,7 +44,7 @@
 [![GitHub Forks](https://img.shields.io/github/forks/code-yeongyu/oh-my-opencode?color=8ae8ff&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/network/members)
 [![GitHub Stars](https://img.shields.io/github/stars/code-yeongyu/oh-my-opencode?color=ffcb47&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/stargazers)
 [![GitHub Issues](https://img.shields.io/github/issues/code-yeongyu/oh-my-opencode?color=ff80eb&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/issues)
-[![License](https://img.shields.io/badge/license-SUL--1.0-white?labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/blob/master/LICENSE.md)
+[![License](https://img.shields.io/badge/license-SUL--1.0-white?labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-opencode/blob/dev/LICENSE.md)
 [![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/code-yeongyu/oh-my-opencode)
 
 [English](README.md) | [한국어](README.ko.md) | [日本語](README.ja.md) | [简体中文](README.zh-cn.md)
@@ -97,7 +97,7 @@
 
 ```
 Install and configure oh-my-opencode by following the instructions here:
-https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 ```
 
 或者你可以直接去读 [安装指南](docs/guide/installation.md)，但说真的，让 Agent 去干吧。人类配环境总是容易敲错字母。
@@ -107,7 +107,7 @@ https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/
 获取安装指南并照做：
 
 ```bash
-curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 ```
 
 ---

assets/oh-my-opencode.schema.json

@@ -1,6 +1,6 @@
 {
   "$schema": "http://json-schema.org/draft-07/schema#",
-  "$id": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$id": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "title": "Oh My OpenCode Configuration",
   "description": "Configuration schema for oh-my-opencode plugin",
   "type": "object",
@@ -24,19 +24,7 @@
     "disabled_agents": {
       "type": "array",
       "items": {
-        "type": "string",
-        "enum": [
-          "sisyphus",
-          "hephaestus",
-          "prometheus",
-          "oracle",
-          "librarian",
-          "explore",
-          "multimodal-looker",
-          "metis",
-          "momus",
-          "atlas"
-        ]
+        "type": "string"
       }
     },
     "disabled_skills": {

benchmarks/bun.lock

@@ -0,0 +1,62 @@
+{
+  "lockfileVersion": 1,
+  "configVersion": 1,
+  "workspaces": {
+    "": {
+      "name": "hashline-edit-benchmark",
+      "dependencies": {
+        "@ai-sdk/openai": "^1.3.0",
+        "@friendliai/ai-provider": "^1.0.9",
+        "ai": "^6.0.94",
+        "zod": "^4.1.0",
+      },
+    },
+  },
+  "packages": {
+    "@ai-sdk/gateway": ["@ai-sdk/gateway@3.0.55", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.15", "@vercel/oidc": "3.1.0" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-7xMeTJnCjwRwXKVCiv4Ly4qzWvDuW3+W1WIV0X1EFu6W83d4mEhV9bFArto10MeTw40ewuDjrbrZd21mXKohkw=="],
+
+    "@ai-sdk/openai": ["@ai-sdk/openai@1.3.24", "", { "dependencies": { "@ai-sdk/provider": "1.1.3", "@ai-sdk/provider-utils": "2.2.8" }, "peerDependencies": { "zod": "^3.0.0" } }, "sha512-GYXnGJTHRTZc4gJMSmFRgEQudjqd4PUN0ZjQhPwOAYH1yOAvQoG/Ikqs+HyISRbLPCrhbZnPKCNHuRU4OfpW0Q=="],
+
+    "@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@2.0.30", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.15" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-iTjumHf1/u4NhjXYFn/aONM2GId3/o7J1Lp5ql8FCbgIMyRwrmanR5xy1S3aaVkfTscuDvLTzWiy1mAbGzK3nQ=="],
+
+    "@ai-sdk/provider": ["@ai-sdk/provider@1.1.3", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-qZMxYJ0qqX/RfnuIaab+zp8UAeJn/ygXXAffR5I4N0n1IrvA6qBsjc8hXLmBiMV2zoXlifkacF7sEFnYnjBcqg=="],
+
+    "@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@2.2.8", "", { "dependencies": { "@ai-sdk/provider": "1.1.3", "nanoid": "^3.3.8", "secure-json-parse": "^2.7.0" }, "peerDependencies": { "zod": "^3.23.8" } }, "sha512-fqhG+4sCVv8x7nFzYnFo19ryhAa3w096Kmc3hWxMQfW/TubPOmt3A6tYZhl4mUfQWWQMsuSkLrtjlWuXBVSGQA=="],
+
+    "@friendliai/ai-provider": ["@friendliai/ai-provider@1.1.4", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.30", "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.15" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.12" } }, "sha512-9TU4B1QFqPhbkONjI5afCF7Ox4jOqtGg1xw8mA9QHZdtlEbZxU+mBNvMPlI5pU5kPoN6s7wkXmFmxpID+own1A=="],
+
+    "@opentelemetry/api": ["@opentelemetry/api@1.9.0", "", {}, "sha512-3giAOQvZiH5F9bMlMiv8+GSPMeqg0dbaeo58/0SlA9sxSqZhnUtxzX9/2FzyhS9sWQf5S0GJE0AKBrFqjpeYcg=="],
+
+    "@standard-schema/spec": ["@standard-schema/spec@1.1.0", "", {}, "sha512-l2aFy5jALhniG5HgqrD6jXLi/rUWrKvqN/qJx6yoJsgKhblVd+iqqU4RCXavm/jPityDo5TCvKMnpjKnOriy0w=="],
+
+    "@vercel/oidc": ["@vercel/oidc@3.1.0", "", {}, "sha512-Fw28YZpRnA3cAHHDlkt7xQHiJ0fcL+NRcIqsocZQUSmbzeIKRpwttJjik5ZGanXP+vlA4SbTg+AbA3bP363l+w=="],
+
+    "ai": ["ai@6.0.101", "", { "dependencies": { "@ai-sdk/gateway": "3.0.55", "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.15", "@opentelemetry/api": "1.9.0" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-Ur/NgbgOp1rdhyDiKDk6EOpSgd1g5ADlbcD1cjQJtQsnmhEngz3Rf8nK5JetDh0vnbLy2aEBpaQeL+zvLRWuaA=="],
+
+    "eventsource-parser": ["eventsource-parser@3.0.6", "", {}, "sha512-Vo1ab+QXPzZ4tCa8SwIHJFaSzy4R6SHf7BY79rFBDf0idraZWAkYrDjDj8uWaSm3S2TK+hJ7/t1CEmZ7jXw+pg=="],
+
+    "json-schema": ["json-schema@0.4.0", "", {}, "sha512-es94M3nTIfsEPisRafak+HDLfHXnKBhV3vU5eqPcS3flIWqcxJWgXHXiey3YrpaNsanY5ei1VoYEbOzijuq9BA=="],
+
+    "nanoid": ["nanoid@3.3.11", "", { "bin": { "nanoid": "bin/nanoid.cjs" } }, "sha512-N8SpfPUnUp1bK+PMYW8qSWdl9U+wwNWI4QKxOYDy9JAro3WMX7p2OeVRF9v+347pnakNevPmiHhNmZ2HbFA76w=="],
+
+    "secure-json-parse": ["secure-json-parse@2.7.0", "", {}, "sha512-6aU+Rwsezw7VR8/nyvKTx8QpWH9FrcYiXXlqC4z5d5XQBDRqtbfsRjnwGyqbi3gddNtWHuEk9OANUotL26qKUw=="],
+
+    "zod": ["zod@4.3.6", "", {}, "sha512-rftlrkhHZOcjDwkGlnUtZZkvaPHCsDATp4pGpuOOMDaTdDDXF91wuVDJoWoPsKX/3YPQ5fHuF3STjcYyKr+Qhg=="],
+
+    "@ai-sdk/gateway/@ai-sdk/provider": ["@ai-sdk/provider@3.0.8", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-oGMAgGoQdBXbZqNG0Ze56CHjDZ1IDYOwGYxYjO5KLSlz5HiNQ9udIXsPZ61VWaHGZ5XW/jyjmr6t2xz2jGVwbQ=="],
+
+    "@ai-sdk/gateway/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.15", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-8XiKWbemmCbvNN0CLR9u3PQiet4gtEVIrX4zzLxnCj06AwsEDJwJVBbKrEI4t6qE8XRSIvU2irka0dcpziKW6w=="],
+
+    "@ai-sdk/openai-compatible/@ai-sdk/provider": ["@ai-sdk/provider@3.0.8", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-oGMAgGoQdBXbZqNG0Ze56CHjDZ1IDYOwGYxYjO5KLSlz5HiNQ9udIXsPZ61VWaHGZ5XW/jyjmr6t2xz2jGVwbQ=="],
+
+    "@ai-sdk/openai-compatible/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.15", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-8XiKWbemmCbvNN0CLR9u3PQiet4gtEVIrX4zzLxnCj06AwsEDJwJVBbKrEI4t6qE8XRSIvU2irka0dcpziKW6w=="],
+
+    "@friendliai/ai-provider/@ai-sdk/provider": ["@ai-sdk/provider@3.0.8", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-oGMAgGoQdBXbZqNG0Ze56CHjDZ1IDYOwGYxYjO5KLSlz5HiNQ9udIXsPZ61VWaHGZ5XW/jyjmr6t2xz2jGVwbQ=="],
+
+    "@friendliai/ai-provider/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.15", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-8XiKWbemmCbvNN0CLR9u3PQiet4gtEVIrX4zzLxnCj06AwsEDJwJVBbKrEI4t6qE8XRSIvU2irka0dcpziKW6w=="],
+
+    "ai/@ai-sdk/provider": ["@ai-sdk/provider@3.0.8", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-oGMAgGoQdBXbZqNG0Ze56CHjDZ1IDYOwGYxYjO5KLSlz5HiNQ9udIXsPZ61VWaHGZ5XW/jyjmr6t2xz2jGVwbQ=="],
+
+    "ai/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.15", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-8XiKWbemmCbvNN0CLR9u3PQiet4gtEVIrX4zzLxnCj06AwsEDJwJVBbKrEI4t6qE8XRSIvU2irka0dcpziKW6w=="],
+  }
+}

benchmarks/headless.ts

@@ -0,0 +1,193 @@
+#!/usr/bin/env bun
+import { readFile, writeFile, mkdir } from "node:fs/promises"
+import { join, dirname } from "node:path"
+import { stepCountIs, streamText, type CoreMessage } from "ai"
+import { tool } from "ai"
+import { createFriendli } from "@friendliai/ai-provider"
+import { z } from "zod"
+import { formatHashLines } from "../src/tools/hashline-edit/hash-computation"
+import { normalizeHashlineEdits } from "../src/tools/hashline-edit/normalize-edits"
+import { applyHashlineEditsWithReport } from "../src/tools/hashline-edit/edit-operations"
+import { canonicalizeFileText, restoreFileText } from "../src/tools/hashline-edit/file-text-canonicalization"
+
+const DEFAULT_MODEL = "MiniMaxAI/MiniMax-M2.5"
+const MAX_STEPS = 50
+const sessionId = `bench-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`
+
+const emit = (event: Record<string, unknown>) =>
+  console.log(JSON.stringify({ sessionId, timestamp: new Date().toISOString(), ...event }))
+
+// ── CLI ──────────────────────────────────────────────────────
+function parseArgs(): { prompt: string; modelId: string } {
+  const args = process.argv.slice(2)
+  let prompt = ""
+  let modelId = DEFAULT_MODEL
+  for (let i = 0; i < args.length; i++) {
+    if ((args[i] === "-p" || args[i] === "--prompt") && args[i + 1]) {
+      prompt = args[++i]
+    } else if ((args[i] === "-m" || args[i] === "--model") && args[i + 1]) {
+      modelId = args[++i]
+    } else if (args[i] === "--reasoning-mode" && args[i + 1]) {
+      i++ // consume
+    }
+    // --no-translate, --think consumed silently
+  }
+  if (!prompt) {
+    console.error("Usage: bun run benchmarks/headless.ts -p <prompt> [-m <model>]")
+    process.exit(1)
+  }
+  return { prompt, modelId }
+}
+
+// ── Tools ────────────────────────────────────────────────────
+const readFileTool = tool({
+  description: "Read a file with hashline-tagged content (LINE#ID format)",
+  inputSchema: z.object({ path: z.string().describe("File path") }),
+  execute: async ({ path }) => {
+    const fullPath = join(process.cwd(), path)
+    try {
+      const content = await readFile(fullPath, "utf-8")
+      const lines = content.split("\n")
+      const tagged = formatHashLines(content)
+      return `OK - read file\npath: ${path}\nlines: ${lines.length}\n\n${tagged}`
+    } catch {
+      return `Error: File not found: ${path}`
+    }
+  },
+})
+
+const editFileTool = tool({
+  description: "Edit a file using hashline anchors (LINE#ID format)",
+  inputSchema: z.object({
+    path: z.string(),
+    edits: z.array(
+      z.object({
+        op: z.enum(["replace", "append", "prepend"]),
+        pos: z.string().optional(),
+        end: z.string().optional(),
+        lines: z.union([z.array(z.string()), z.string(), z.null()]),
+      })
+    ).min(1),
+  }),
+  execute: async ({ path, edits }) => {
+    const fullPath = join(process.cwd(), path)
+    try {
+      let rawContent = ""
+      let exists = true
+      try {
+        rawContent = await readFile(fullPath, "utf-8")
+      } catch {
+        exists = false
+      }
+
+      const normalized = normalizeHashlineEdits(edits)
+
+      if (!exists) {
+        const canCreate = normalized.every(
+          (e) => (e.op === "append" || e.op === "prepend") && !e.pos
+        )
+        if (!canCreate) return `Error: File not found: ${path}`
+      }
+
+      const envelope = canonicalizeFileText(rawContent)
+      const result = applyHashlineEditsWithReport(envelope.content, normalized)
+
+      if (result.content === envelope.content) {
+        return `Error: No changes made to ${path}. The edits produced identical content.`
+      }
+
+      const writeContent = restoreFileText(result.content, envelope)
+      await mkdir(dirname(fullPath), { recursive: true })
+      await writeFile(fullPath, writeContent, "utf-8")
+
+      const oldLineCount = rawContent.split("\n").length
+      const newLineCount = writeContent.split("\n").length
+      const delta = newLineCount - oldLineCount
+      const sign = delta > 0 ? "+" : ""
+      const action = exists ? "Updated" : "Created"
+      return `${action} ${path}\n${edits.length} edit(s) applied, ${sign}${delta} line(s)`
+    } catch (error) {
+      return `Error: ${error instanceof Error ? error.message : String(error)}`
+    }
+  },
+})
+
+// ── Agent Loop ───────────────────────────────────────────────
+async function run() {
+  const { prompt, modelId } = parseArgs()
+
+  const friendli = createFriendli({ apiKey: process.env.FRIENDLI_TOKEN! })
+  const model = friendli(modelId)
+  const tools = { read_file: readFileTool, edit_file: editFileTool }
+
+  emit({ type: "user", content: prompt })
+
+  const messages: CoreMessage[] = [{ role: "user", content: prompt }]
+  const system =
+    "You are a code editing assistant. Use read_file to read files and edit_file to edit them. " +
+    "Always read a file before editing it to get fresh LINE#ID anchors."
+
+  for (let step = 0; step < MAX_STEPS; step++) {
+    const stream = streamText({
+      model,
+      tools,
+      messages,
+      system,
+      stopWhen: stepCountIs(1),
+    })
+
+    let currentText = ""
+    for await (const part of stream.fullStream) {
+      switch (part.type) {
+        case "text-delta":
+          currentText += part.text
+          break
+        case "tool-call":
+          emit({
+            type: "tool_call",
+            tool_call_id: part.toolCallId,
+            tool_name: part.toolName,
+            tool_input: part.args,
+            model: modelId,
+          })
+          break
+        case "tool-result": {
+          const output = typeof part.result === "string" ? part.result : JSON.stringify(part.result)
+          const isError = typeof output === "string" && output.startsWith("Error:")
+          emit({
+            type: "tool_result",
+            tool_call_id: part.toolCallId,
+            output,
+            ...(isError ? { error: output } : {}),
+          })
+          break
+      }
+    }
+
+    const response = await stream.response
+    messages.push(...response.messages)
+
+    const finishReason = await stream.finishReason
+    if (finishReason !== "tool-calls") {
+      if (currentText.trim()) {
+        emit({ type: "assistant", content: currentText, model: modelId })
+      }
+      break
+    }
+  }
+}
+
+// ── Signal + Startup ─────────────────────────────────────────
+process.once("SIGINT", () => process.exit(0))
+process.once("SIGTERM", () => process.exit(143))
+
+const startTime = Date.now()
+run()
+  .catch((error) => {
+    emit({ type: "error", error: error instanceof Error ? error.message : String(error) })
+    process.exit(1)
+  })
+  .then(() => {
+    const elapsed = ((Date.now() - startTime) / 1000).toFixed(2)
+    console.error(`[headless] Completed in ${elapsed}s`)
+  })

benchmarks/package.json

@@ -0,0 +1,19 @@
+{
+  "name": "hashline-edit-benchmark",
+  "version": "0.1.0",
+  "private": true,
+  "type": "module",
+  "description": "Hashline edit tool benchmark using Vercel AI SDK with FriendliAI provider",
+  "scripts": {
+    "bench:basic": "bun run test-edit-ops.ts",
+    "bench:edge": "bun run test-edge-cases.ts",
+    "bench:multi": "bun run test-multi-model.ts",
+    "bench:all": "bun run bench:basic && bun run bench:edge"
+  },
+  "dependencies": {
+    "ai": "^6.0.94",
+    "@ai-sdk/openai": "^1.3.0",
+    "@friendliai/ai-provider": "^1.0.9",
+    "zod": "^4.1.0"
+  }
+}

benchmarks/test-edit-ops.ts

@@ -0,0 +1,808 @@
+#!/usr/bin/env bun
+/**
+ * Comprehensive headless edit_file stress test: 21 operation types
+ *
+ * Tests: 5 basic ops + 10 creative cases + 6 whitespace cases
+ * Each runs via headless mode with its own demo file + prompt.
+ *
+ * Usage:
+ *   bun run scripts/test-headless-edit-ops.ts [-m <model>] [--provider <provider>]
+ */
+
+import { spawn } from "node:child_process";
+import { mkdirSync, readFileSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join, resolve } from "node:path";
+
+// ── CLI arg passthrough ───────────────────────────────────────
+const extraArgs: string[] = [];
+const rawArgs = process.argv.slice(2);
+for (let i = 0; i < rawArgs.length; i++) {
+  const arg = rawArgs[i];
+  if (
+    (arg === "-m" || arg === "--model" || arg === "--provider") &&
+    i + 1 < rawArgs.length
+  ) {
+    extraArgs.push(arg, rawArgs[i + 1]);
+    i++;
+  } else if (arg === "--think" || arg === "--no-translate") {
+    extraArgs.push(arg);
+  } else if (arg === "--reasoning-mode" && i + 1 < rawArgs.length) {
+    extraArgs.push(arg, rawArgs[i + 1]);
+    i++;
+  }
+}
+
+// ── Colors ────────────────────────────────────────────────────
+const BOLD = "\x1b[1m";
+const GREEN = "\x1b[32m";
+const RED = "\x1b[31m";
+const YELLOW = "\x1b[33m";
+const DIM = "\x1b[2m";
+const CYAN = "\x1b[36m";
+const RESET = "\x1b[0m";
+
+const pass = (msg: string) => console.log(`  ${GREEN}✓${RESET} ${msg}`);
+const fail = (msg: string) => console.log(`  ${RED}✗${RESET} ${msg}`);
+const info = (msg: string) => console.log(`  ${DIM}${msg}${RESET}`);
+const warn = (msg: string) => console.log(`  ${YELLOW}⚠${RESET} ${msg}`);
+
+// ── Test case definition ─────────────────────────────────────
+interface TestCase {
+  fileContent: string;
+  fileName: string;
+  name: string;
+  prompt: string;
+  validate: (content: string) => { passed: boolean; reason: string };
+}
+
+const TEST_CASES: TestCase[] = [
+  {
+    name: "1. Replace single line",
+    fileName: "config.txt",
+    fileContent: [
+      "host: localhost",
+      "port: 3000",
+      "debug: false",
+      "timeout: 30",
+      "retries: 3",
+    ].join("\n"),
+    prompt: [
+      "Follow these steps exactly:",
+      "Step 1: Call read_file on config.txt.",
+      "Step 2: Note the anchor for the port line (line 2).",
+      "Step 3: Call edit_file with path='config.txt' and edits containing ONE object:",
+      "  { op: 'replace', pos: '<line2 anchor>', lines: ['port: 8080'] }",
+      "IMPORTANT: pos must be ONLY the anchor (like '2#KB'). lines must be a SEPARATE array field with the new content.",
+    ].join(" "),
+    validate: (content) => {
+      const has8080 = content.includes("port: 8080");
+      const has3000 = content.includes("port: 3000");
+      if (has8080 && !has3000) {
+        return { passed: true, reason: "port changed to 8080" };
+      }
+      if (has3000) {
+        return { passed: false, reason: "port still 3000 — edit not applied" };
+      }
+      return {
+        passed: false,
+        reason: `unexpected content: ${content.slice(0, 100)}`,
+      };
+    },
+  },
+  {
+    name: "2. Append after line",
+    fileName: "fruits.txt",
+    fileContent: ["apple", "banana", "cherry"].join("\n"),
+    prompt:
+      "Read fruits.txt with read_file. Then use edit_file with op='append' to insert a new line 'grape' after the 'banana' line. Use pos='LINE#HASH' of the banana line and lines=['grape'].",
+    validate: (content) => {
+      const lines = content.trim().split("\n");
+      const bananaIdx = lines.findIndex((l) => l.trim() === "banana");
+      const grapeIdx = lines.findIndex((l) => l.trim() === "grape");
+      if (grapeIdx === -1) {
+        return { passed: false, reason: '"grape" not found in file' };
+      }
+      if (bananaIdx === -1) {
+        return { passed: false, reason: '"banana" was removed' };
+      }
+      if (grapeIdx !== bananaIdx + 1) {
+        return {
+          passed: false,
+          reason: `"grape" at line ${grapeIdx + 1} but expected after "banana" at line ${bananaIdx + 1}`,
+        };
+      }
+      if (lines.length !== 4) {
+        return {
+          passed: false,
+          reason: `expected 4 lines, got ${lines.length}`,
+        };
+      }
+      return {
+        passed: true,
+        reason: '"grape" correctly appended after "banana"',
+      };
+    },
+  },
+  {
+    name: "3. Prepend before line",
+    fileName: "code.txt",
+    fileContent: ["function greet() {", '  return "hello";', "}"].join("\n"),
+    prompt:
+      "Read code.txt with read_file. Then use edit_file with op='prepend' to add '// Greeting function' before the function line. Use pos='LINE#HASH' of the function line and lines=['// Greeting function'].",
+    validate: (content) => {
+      const lines = content.trim().split("\n");
+      const commentIdx = lines.findIndex(
+        (l) => l.trim().startsWith("//") && l.toLowerCase().includes("greet")
+      );
+      const funcIdx = lines.findIndex((l) =>
+        l.trim().startsWith("function greet")
+      );
+      if (commentIdx === -1) {
+        return { passed: false, reason: "comment line not found" };
+      }
+      if (funcIdx === -1) {
+        return { passed: false, reason: '"function greet" line was removed' };
+      }
+      if (commentIdx !== funcIdx - 1) {
+        return {
+          passed: false,
+          reason: `comment at line ${commentIdx + 1} but function at ${funcIdx + 1} — not directly before`,
+        };
+      }
+      return {
+        passed: true,
+        reason: "comment correctly prepended before function",
+      };
+    },
+  },
+  {
+    name: "4. Range replace (multi-line → single line)",
+    fileName: "log.txt",
+    fileContent: [
+      "=== Log Start ===",
+      "INFO: started",
+      "WARN: slow query",
+      "ERROR: timeout",
+      "INFO: recovered",
+      "=== Log End ===",
+    ].join("\n"),
+    prompt: [
+      "Follow these steps exactly:",
+      "Step 1: Call read_file on log.txt to see line anchors.",
+      "Step 2: Note the anchor for 'WARN: slow query' (line 3) and 'ERROR: timeout' (line 4).",
+      "Step 3: Call edit_file with path='log.txt' and edits containing ONE object with THREE separate JSON fields:",
+      "  { op: 'replace', pos: '<line3 anchor>', end: '<line4 anchor>', lines: ['RESOLVED: issues cleared'] }",
+      "CRITICAL: pos, end, and lines are THREE SEPARATE JSON fields. pos is ONLY '3#XX'. end is ONLY '4#YY'. lines is ['RESOLVED: issues cleared'].",
+      "If edit_file fails or errors, use write_file to write the complete correct file content instead.",
+      "The correct final content should be: === Log Start ===, INFO: started, RESOLVED: issues cleared, INFO: recovered, === Log End ===",
+      "Do not make any other changes.",
+    ].join(" "),
+    validate: (content) => {
+      const lines = content.trim().split("\n");
+      const hasResolved = lines.some(
+        (l) => l.trim() === "RESOLVED: issues cleared"
+      );
+      const hasWarn = content.includes("WARN: slow query");
+      const hasError = content.includes("ERROR: timeout");
+      if (!hasResolved) {
+        return {
+          passed: false,
+          reason: '"RESOLVED: issues cleared" not found',
+        };
+      }
+      if (hasWarn || hasError) {
+        return { passed: false, reason: "old WARN/ERROR lines still present" };
+      }
+      // Core assertion: 2 old lines removed, 1 new line added = net -1 line
+      // Allow slight overshoot from model adding extra content
+      if (lines.length < 4 || lines.length > 6) {
+        return {
+          passed: false,
+          reason: `expected ~5 lines, got ${lines.length}`,
+        };
+      }
+      return {
+        passed: true,
+        reason: "range replace succeeded — 2 lines → 1 line",
+      };
+    },
+  },
+  {
+    name: "5. Delete line",
+    fileName: "settings.txt",
+    fileContent: [
+      "mode: production",
+      "debug: true",
+      "cache: enabled",
+      "log_level: info",
+    ].join("\n"),
+    prompt: [
+      "Follow these steps exactly:",
+      "Step 1: Call read_file on settings.txt to see line anchors.",
+      "Step 2: Note the anchor for 'debug: true' (line 2).",
+      "Step 3: Call edit_file with path='settings.txt' and edits containing ONE object:",
+      "  { op: 'replace', pos: '<line2 anchor>', lines: [] }",
+      "IMPORTANT: lines must be an empty array [] to delete the line. pos must be ONLY the anchor like '2#SR'.",
+    ].join(" "),
+    validate: (content) => {
+      const lines = content.trim().split("\n");
+      const hasDebug = content.includes("debug: true");
+      if (hasDebug) {
+        return { passed: false, reason: '"debug: true" still present' };
+      }
+      if (lines.length !== 3) {
+        return {
+          passed: false,
+          reason: `expected 3 lines, got ${lines.length}`,
+        };
+      }
+      if (
+        !(
+          content.includes("mode: production") &&
+          content.includes("cache: enabled")
+        )
+      ) {
+        return { passed: false, reason: "other lines were removed" };
+      }
+      return { passed: true, reason: '"debug: true" successfully deleted' };
+    },
+  },
+
+  // ── Creative cases (6-15) ────────────────────────────────────
+  {
+    name: "6. Batch edit — two replacements in one call",
+    fileName: "batch.txt",
+    fileContent: ["red", "green", "blue", "yellow"].join("\n"),
+    prompt: [
+      "Read batch.txt with read_file.",
+      "Then call edit_file ONCE with path='batch.txt' and edits containing TWO objects:",
+      "  1) { op: 'replace', pos: '<line1 anchor>', lines: ['crimson'] }",
+      "  2) { op: 'replace', pos: '<line3 anchor>', lines: ['navy'] }",
+      "Both edits must be in the SAME edits array in a single edit_file call.",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.trim().split("\n");
+      if (!c.includes("crimson")) return { passed: false, reason: "'crimson' not found" };
+      if (!c.includes("navy")) return { passed: false, reason: "'navy' not found" };
+      if (c.includes("red")) return { passed: false, reason: "'red' still present" };
+      if (c.includes("blue")) return { passed: false, reason: "'blue' still present" };
+      if (lines.length !== 4) return { passed: false, reason: `expected 4 lines, got ${lines.length}` };
+      return { passed: true, reason: "both lines replaced in single call" };
+    },
+  },
+  {
+    name: "7. Line expansion — 1 line → 3 lines",
+    fileName: "expand.txt",
+    fileContent: ["header", "TODO: implement", "footer"].join("\n"),
+    prompt: [
+      "Read expand.txt with read_file.",
+      "Replace the 'TODO: implement' line (line 2) with THREE lines:",
+      "  'step 1: init', 'step 2: process', 'step 3: cleanup'",
+      "Use edit_file with op='replace', pos=<line2 anchor>, lines=['step 1: init', 'step 2: process', 'step 3: cleanup'].",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.trim().split("\n");
+      if (c.includes("TODO")) return { passed: false, reason: "TODO line still present" };
+      if (!c.includes("step 1: init")) return { passed: false, reason: "'step 1: init' not found" };
+      if (!c.includes("step 3: cleanup")) return { passed: false, reason: "'step 3: cleanup' not found" };
+      if (lines.length !== 5) return { passed: false, reason: `expected 5 lines, got ${lines.length}` };
+      return { passed: true, reason: "1 line expanded to 3 lines" };
+    },
+  },
+  {
+    name: "8. Append at EOF",
+    fileName: "eof.txt",
+    fileContent: ["line one", "line two"].join("\n"),
+    prompt: [
+      "Read eof.txt with read_file.",
+      "Use edit_file to append 'line three' after the LAST line of the file.",
+      "Use op='append', pos=<last line anchor>, lines=['line three'].",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.trim().split("\n");
+      if (!c.includes("line three")) return { passed: false, reason: "'line three' not found" };
+      if (lines[lines.length - 1].trim() !== "line three")
+        return { passed: false, reason: "'line three' not at end" };
+      if (lines.length !== 3) return { passed: false, reason: `expected 3 lines, got ${lines.length}` };
+      return { passed: true, reason: "appended at EOF" };
+    },
+  },
+  {
+    name: "9. Special characters in content",
+    fileName: "special.json",
+    fileContent: [
+      '{',
+      '  "name": "old-value",',
+      '  "count": 42',
+      '}',
+    ].join("\n"),
+    prompt: [
+      "Read special.json with read_file.",
+      'Replace the line containing \"name\": \"old-value\" with \"name\": \"new-value\".',
+      "Use edit_file with op='replace', pos=<that line's anchor>, lines=['  \"name\": \"new-value\",'].",
+    ].join(" "),
+    validate: (c) => {
+      if (c.includes("old-value")) return { passed: false, reason: "'old-value' still present" };
+      if (!c.includes('"new-value"')) return { passed: false, reason: "'new-value' not found" };
+      if (!c.includes('"count": 42')) return { passed: false, reason: "other content was modified" };
+      return { passed: true, reason: "JSON value replaced with special chars intact" };
+    },
+  },
+  {
+    name: "10. Replace first line",
+    fileName: "first.txt",
+    fileContent: ["OLD HEADER", "body content", "footer"].join("\n"),
+    prompt: [
+      "Read first.txt with read_file.",
+      "Replace the very first line 'OLD HEADER' with 'NEW HEADER'.",
+      "Use edit_file with op='replace', pos=<line1 anchor>, lines=['NEW HEADER'].",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.trim().split("\n");
+      if (c.includes("OLD HEADER")) return { passed: false, reason: "'OLD HEADER' still present" };
+      if (lines[0].trim() !== "NEW HEADER") return { passed: false, reason: "first line is not 'NEW HEADER'" };
+      if (!c.includes("body content")) return { passed: false, reason: "body was modified" };
+      return { passed: true, reason: "first line replaced" };
+    },
+  },
+  {
+    name: "11. Replace last line",
+    fileName: "last.txt",
+    fileContent: ["alpha", "bravo", "OLD_FOOTER"].join("\n"),
+    prompt: [
+      "Read last.txt with read_file.",
+      "Replace the last line 'OLD_FOOTER' with 'NEW_FOOTER'.",
+      "Use edit_file with op='replace', pos=<last line anchor>, lines=['NEW_FOOTER'].",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.trim().split("\n");
+      if (c.includes("OLD_FOOTER")) return { passed: false, reason: "'OLD_FOOTER' still present" };
+      if (lines[lines.length - 1].trim() !== "NEW_FOOTER")
+        return { passed: false, reason: "last line is not 'NEW_FOOTER'" };
+      return { passed: true, reason: "last line replaced" };
+    },
+  },
+  {
+    name: "12. Adjacent line edits",
+    fileName: "adjacent.txt",
+    fileContent: ["aaa", "bbb", "ccc", "ddd"].join("\n"),
+    prompt: [
+      "Read adjacent.txt with read_file.",
+      "Replace line 2 ('bbb') with 'BBB' and line 3 ('ccc') with 'CCC'.",
+      "Use edit_file with TWO edits in the same call:",
+      "  { op: 'replace', pos: <line2 anchor>, lines: ['BBB'] }",
+      "  { op: 'replace', pos: <line3 anchor>, lines: ['CCC'] }",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.trim().split("\n");
+      if (c.includes("bbb")) return { passed: false, reason: "'bbb' still present" };
+      if (c.includes("ccc")) return { passed: false, reason: "'ccc' still present" };
+      if (!c.includes("BBB")) return { passed: false, reason: "'BBB' not found" };
+      if (!c.includes("CCC")) return { passed: false, reason: "'CCC' not found" };
+      if (lines.length !== 4) return { passed: false, reason: `expected 4 lines, got ${lines.length}` };
+      return { passed: true, reason: "two adjacent lines replaced" };
+    },
+  },
+  {
+    name: "13. Prepend multi-line block",
+    fileName: "block.py",
+    fileContent: ["def main():", "    print('hello')", "", "main()"].join("\n"),
+    prompt: [
+      "Read block.py with read_file.",
+      "Prepend a 2-line comment block before 'def main():' (line 1).",
+      "The two lines are: '# Author: test' and '# Date: 2025-01-01'.",
+      "Use edit_file with op='prepend', pos=<line1 anchor>, lines=['# Author: test', '# Date: 2025-01-01'].",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.trim().split("\n");
+      if (!c.includes("# Author: test")) return { passed: false, reason: "author comment not found" };
+      if (!c.includes("# Date: 2025-01-01")) return { passed: false, reason: "date comment not found" };
+      const defIdx = lines.findIndex((l) => l.startsWith("def main"));
+      const authorIdx = lines.findIndex((l) => l.includes("Author"));
+      if (authorIdx >= defIdx) return { passed: false, reason: "comments not before def" };
+      return { passed: true, reason: "2-line block prepended before function" };
+    },
+  },
+  {
+    name: "14. Delete range — 3 consecutive lines",
+    fileName: "cleanup.txt",
+    fileContent: ["keep1", "remove-a", "remove-b", "remove-c", "keep2"].join("\n"),
+    prompt: [
+      "Read cleanup.txt with read_file.",
+      "Delete lines 2-4 ('remove-a', 'remove-b', 'remove-c') using a single range replace.",
+      "Use edit_file with op='replace', pos=<line2 anchor>, end=<line4 anchor>, lines=[].",
+      "An empty lines array deletes the range.",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.trim().split("\n");
+      if (c.includes("remove")) return { passed: false, reason: "'remove' lines still present" };
+      if (!c.includes("keep1")) return { passed: false, reason: "'keep1' was deleted" };
+      if (!c.includes("keep2")) return { passed: false, reason: "'keep2' was deleted" };
+      if (lines.length !== 2) return { passed: false, reason: `expected 2 lines, got ${lines.length}` };
+      return { passed: true, reason: "3 consecutive lines deleted via range" };
+    },
+  },
+  {
+    name: "15. Replace with duplicate-content line",
+    fileName: "dupes.txt",
+    fileContent: ["item", "item", "item", "item"].join("\n"),
+    prompt: [
+      "Read dupes.txt with read_file. All 4 lines have the same text 'item'.",
+      "Replace ONLY line 3 with 'CHANGED'. Do NOT modify any other line.",
+      "Use edit_file with op='replace', pos=<line3 anchor>, lines=['CHANGED'].",
+      "The anchor hash uniquely identifies line 3 even though the content is identical.",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.trim().split("\n");
+      if (!c.includes("CHANGED")) return { passed: false, reason: "'CHANGED' not found" };
+      const changedCount = lines.filter((l) => l.trim() === "CHANGED").length;
+      const itemCount = lines.filter((l) => l.trim() === "item").length;
+      if (changedCount !== 1) return { passed: false, reason: `expected 1 CHANGED, got ${changedCount}` };
+      if (itemCount !== 3) return { passed: false, reason: `expected 3 item lines, got ${itemCount}` };
+      if (lines.length !== 4) return { passed: false, reason: `expected 4 lines, got ${lines.length}` };
+      return { passed: true, reason: "only line 3 changed among duplicates" };
+    },
+  },
+
+  // ── Whitespace cases (16-21) ──────────────────────────────────
+  {
+    name: "16. Fix indentation — 2 spaces → 4 spaces",
+    fileName: "indent.js",
+    fileContent: ["function foo() {", "  const x = 1;", "  return x;", "}"].join("\n"),
+    prompt: [
+      "Read indent.js with read_file.",
+      "Replace line 2 '  const x = 1;' (2-space indent) with '    const x = 1;' (4-space indent).",
+      "Use edit_file with op='replace', pos=<line2 anchor>, lines=['    const x = 1;'].",
+      "The ONLY change is the indentation: 2 spaces → 4 spaces. Content stays the same.",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.split("\n");
+      const line2 = lines[1];
+      if (!line2) return { passed: false, reason: "line 2 missing" };
+      if (line2 === "    const x = 1;") return { passed: true, reason: "indentation fixed to 4 spaces" };
+      if (line2 === "  const x = 1;") return { passed: false, reason: "still 2-space indent" };
+      return { passed: false, reason: `unexpected line 2: '${line2}'` };
+    },
+  },
+  {
+    name: "17. Replace preserving leading whitespace",
+    fileName: "preserve.py",
+    fileContent: [
+      "class Foo:",
+      "    def old_method(self):",
+      "        pass",
+    ].join("\n"),
+    prompt: [
+      "Read preserve.py with read_file.",
+      "Replace line 2 '    def old_method(self):' with '    def new_method(self):'.",
+      "Keep the 4-space indentation. Only change the method name.",
+      "Use edit_file with op='replace', pos=<line2 anchor>, lines=['    def new_method(self):'].",
+    ].join(" "),
+    validate: (c) => {
+      if (c.includes("old_method")) return { passed: false, reason: "'old_method' still present" };
+      const lines = c.split("\n");
+      const methodLine = lines.find((l) => l.includes("new_method"));
+      if (!methodLine) return { passed: false, reason: "'new_method' not found" };
+      if (!methodLine.startsWith("    ")) return { passed: false, reason: "indentation lost" };
+      return { passed: true, reason: "method renamed with indentation preserved" };
+    },
+  },
+  {
+    name: "18. Insert blank line between sections",
+    fileName: "sections.txt",
+    fileContent: ["[section-a]", "value-a=1", "[section-b]", "value-b=2"].join("\n"),
+    prompt: [
+      "Read sections.txt with read_file.",
+      "Insert a blank empty line between 'value-a=1' (line 2) and '[section-b]' (line 3).",
+      "Use edit_file with op='append', pos=<line2 anchor>, lines=[''].",
+      "lines=[''] inserts one empty line.",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.split("\n");
+      const valAIdx = lines.findIndex((l) => l.includes("value-a=1"));
+      const secBIdx = lines.findIndex((l) => l.includes("[section-b]"));
+      if (valAIdx === -1) return { passed: false, reason: "'value-a=1' missing" };
+      if (secBIdx === -1) return { passed: false, reason: "'[section-b]' missing" };
+      if (secBIdx - valAIdx < 2) return { passed: false, reason: "no blank line between sections" };
+      const between = lines[valAIdx + 1];
+      if (between.trim() !== "") return { passed: false, reason: `line between is '${between}', not blank` };
+      return { passed: true, reason: "blank line inserted between sections" };
+    },
+  },
+  {
+    name: "19. Delete blank line",
+    fileName: "noblank.txt",
+    fileContent: ["first", "", "second", "third"].join("\n"),
+    prompt: [
+      "Read noblank.txt with read_file.",
+      "Delete the empty blank line (line 2). Use edit_file with op='replace', pos=<line2 anchor>, lines=[].",
+    ].join(" "),
+    validate: (c) => {
+      const lines = c.trim().split("\n");
+      if (lines.length !== 3) return { passed: false, reason: `expected 3 lines, got ${lines.length}` };
+      if (lines[0].trim() !== "first") return { passed: false, reason: "'first' not on line 1" };
+      if (lines[1].trim() !== "second") return { passed: false, reason: "'second' not on line 2" };
+      return { passed: true, reason: "blank line deleted" };
+    },
+  },
+  {
+    name: "20. Tab → spaces conversion",
+    fileName: "tabs.txt",
+    fileContent: ["start", "\tindented-with-tab", "end"].join("\n"),
+    prompt: [
+      "Read tabs.txt with read_file.",
+      "Replace the tab-indented line 2 using edit_file with edits: [{ op: 'replace', pos: '<line2 anchor>', lines: ['    indented-with-spaces'] }].",
+      "Expected final line 2 to be 4 spaces followed by indented-with-spaces.",
+    ].join(" "),
+    validate: (c) => {
+      if (c.includes("\t")) return { passed: false, reason: "tab still present" };
+      if (!c.includes("    indented-with-spaces"))
+        return { passed: false, reason: "'    indented-with-spaces' not found" };
+      if (!c.includes("start")) return { passed: false, reason: "'start' was modified" };
+      return { passed: true, reason: "tab converted to 4 spaces" };
+    },
+  },
+  {
+    name: "21. Deeply nested indent replacement",
+    fileName: "nested.ts",
+    fileContent: [
+      "if (a) {",
+      "  if (b) {",
+      "    if (c) {",
+      "      old_call();",
+      "    }",
+      "  }",
+      "}",
+    ].join("\n"),
+    prompt: [
+      "Read nested.ts with read_file.",
+      "Replace line 4 '      old_call();' with '      new_call();'.",
+      "Preserve the exact 6-space indentation. Only change the function name.",
+      "Use edit_file with op='replace', pos=<line4 anchor>, lines=['      new_call();'].",
+    ].join(" "),
+    validate: (c) => {
+      if (c.includes("old_call")) return { passed: false, reason: "'old_call' still present" };
+      const lines = c.split("\n");
+      const callLine = lines.find((l) => l.includes("new_call"));
+      if (!callLine) return { passed: false, reason: "'new_call' not found" };
+      const leadingSpaces = callLine.match(/^ */)?.[0].length ?? 0;
+      if (leadingSpaces !== 6) return { passed: false, reason: `expected 6-space indent, got ${leadingSpaces}` };
+      return { passed: true, reason: "deeply nested line replaced with indent preserved" };
+    },
+  },
+];
+
+// ── JSONL event types ─────────────────────────────────────────
+interface ToolCallEvent {
+  tool_call_id: string;
+  tool_input: Record<string, unknown>;
+  tool_name: string;
+  type: "tool_call";
+}
+
+interface ToolResultEvent {
+  error?: string;
+  output: string;
+  tool_call_id: string;
+  type: "tool_result";
+}
+
+interface AnyEvent {
+  type: string;
+  [key: string]: unknown;
+}
+
+// ── Run single test case ─────────────────────────────────────
+async function runTestCase(
+  tc: TestCase,
+  testDir: string
+): Promise<{
+  passed: boolean;
+  editCalls: number;
+  editSuccesses: number;
+  duration: number;
+}> {
+  const testFile = join(testDir, tc.fileName);
+  writeFileSync(testFile, tc.fileContent, "utf-8");
+
+  const headlessScript = resolve(import.meta.dir, "headless.ts");
+  const headlessArgs = [
+    "run",
+    headlessScript,
+    "-p",
+    tc.prompt,
+    "--no-translate",
+    ...extraArgs,
+  ];
+
+  const startTime = Date.now();
+
+  const output = await new Promise<string>((res, reject) => {
+    const proc = spawn("bun", headlessArgs, {
+      cwd: testDir,
+      env: { ...process.env, BUN_INSTALL: process.env.BUN_INSTALL },
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+
+    let stdout = "";
+    let stderr = "";
+
+    proc.stdout.on("data", (chunk: Buffer) => {
+      stdout += chunk.toString();
+    });
+    proc.stderr.on("data", (chunk: Buffer) => {
+      stderr += chunk.toString();
+    });
+
+    const timeout = setTimeout(
+      () => {
+        proc.kill("SIGTERM");
+        reject(new Error("Timed out after 4 minutes"));
+      },
+      4 * 60 * 1000
+    );
+
+    proc.on("close", (code) => {
+      clearTimeout(timeout);
+      if (code !== 0) {
+        reject(new Error(`Exit code ${code}\n${stderr.slice(-500)}`));
+      } else {
+        res(stdout);
+      }
+    });
+    proc.on("error", (err) => {
+      clearTimeout(timeout);
+      reject(err);
+    });
+  });
+
+  const duration = Date.now() - startTime;
+
+  // Parse events
+  const events: AnyEvent[] = [];
+  for (const line of output.split("\n").filter((l) => l.trim())) {
+    try {
+      events.push(JSON.parse(line) as AnyEvent);
+    } catch {
+      // skip non-JSON
+    }
+  }
+
+  const toolCalls = events.filter(
+    (e) => e.type === "tool_call"
+  ) as unknown as ToolCallEvent[];
+  const toolResults = events.filter(
+    (e) => e.type === "tool_result"
+  ) as unknown as ToolResultEvent[];
+
+  const editCalls = toolCalls.filter((e) => e.tool_name === "edit_file");
+  const editCallIds = new Set(editCalls.map((e) => e.tool_call_id));
+  const editResults = toolResults.filter((e) =>
+    editCallIds.has(e.tool_call_id)
+  );
+  const editSuccesses = editResults.filter((e) => !e.error);
+
+  // Show blocked calls
+  const editErrors = editResults.filter((e) => e.error);
+  for (const err of editErrors) {
+    const matchingCall = editCalls.find(
+      (c) => c.tool_call_id === err.tool_call_id
+    );
+    info(`  blocked: ${err.error?.slice(0, 120)}`);
+    if (matchingCall) {
+      info(`  input: ${JSON.stringify(matchingCall.tool_input).slice(0, 200)}`);
+    }
+  }
+
+  // Validate file content
+  let finalContent: string;
+  try {
+    finalContent = readFileSync(testFile, "utf-8");
+  } catch {
+    return {
+      passed: false,
+      editCalls: editCalls.length,
+      editSuccesses: editSuccesses.length,
+      duration,
+    };
+  }
+
+  const validation = tc.validate(finalContent);
+
+  return {
+    passed: validation.passed,
+    editCalls: editCalls.length,
+    editSuccesses: editSuccesses.length,
+    duration,
+  };
+}
+
+// ── Main ──────────────────────────────────────────────────────
+const main = async () => {
+  console.log(`\n${BOLD}Headless Edit Operations Test — ${TEST_CASES.length} Types${RESET}\n`);
+
+  const testDir = join(tmpdir(), `edit-ops-${Date.now()}`);
+  mkdirSync(testDir, { recursive: true });
+  info(`Test dir: ${testDir}`);
+  console.log();
+
+  let totalPassed = 0;
+  const results: { name: string; passed: boolean; detail: string }[] = [];
+
+  for (const tc of TEST_CASES) {
+    console.log(`${CYAN}${BOLD}${tc.name}${RESET}`);
+    info(`File: ${tc.fileName}`);
+    info(`Prompt: "${tc.prompt.slice(0, 80)}..."`);
+
+    try {
+      const result = await runTestCase(tc, testDir);
+      const status = result.passed
+        ? `${GREEN}PASS${RESET}`
+        : `${RED}FAIL${RESET}`;
+      const detail = `edit_file: ${result.editSuccesses}/${result.editCalls} succeeded, ${(result.duration / 1000).toFixed(1)}s`;
+
+      console.log(`  ${status} — ${detail}`);
+
+      if (result.passed) {
+        totalPassed++;
+        // Validate the file to show reason
+        const content = readFileSync(join(testDir, tc.fileName), "utf-8");
+        const v = tc.validate(content);
+        pass(v.reason);
+      } else {
+        const content = readFileSync(join(testDir, tc.fileName), "utf-8");
+        const v = tc.validate(content);
+        fail(v.reason);
+        info(
+          `Final content:\n${content
+            .split("\n")
+            .map((l, i) => `    ${i + 1}: ${l}`)
+            .join("\n")}`
+        );
+      }
+
+      results.push({ name: tc.name, passed: result.passed, detail });
+    } catch (error) {
+      const msg = error instanceof Error ? error.message : String(error);
+      console.log(`  ${RED}ERROR${RESET} — ${msg.slice(0, 200)}`);
+      fail(msg.slice(0, 200));
+      results.push({ name: tc.name, passed: false, detail: msg.slice(0, 100) });
+    }
+
+    // Reset file for next test (in case of side effects)
+    try {
+      rmSync(join(testDir, tc.fileName), { force: true });
+    } catch {}
+
+    console.log();
+  }
+
+  // Summary
+  console.log(`${BOLD}━━━ Summary ━━━${RESET}`);
+  for (const r of results) {
+    const icon = r.passed ? `${GREEN}✓${RESET}` : `${RED}✗${RESET}`;
+    console.log(`  ${icon} ${r.name} — ${r.detail}`);
+  }
+  console.log();
+  console.log(
+    `${BOLD}Result: ${totalPassed}/${TEST_CASES.length} passed (${Math.round((totalPassed / TEST_CASES.length) * 100)}%)${RESET}`
+  );
+
+  // Cleanup
+  try {
+    rmSync(testDir, { recursive: true, force: true });
+  } catch {}
+
+  if (totalPassed === TEST_CASES.length) {
+    console.log(
+      `\n${BOLD}${GREEN}🎉 ALL TESTS PASSED — 100% success rate!${RESET}\n`
+    );
+    process.exit(0);
+  } else {
+    console.log(`\n${BOLD}${RED}Some tests failed.${RESET}\n`);
+    process.exit(1);
+  }
+};
+
+main();

benchmarks/test-multi-model.ts

@@ -0,0 +1,280 @@
+#!/usr/bin/env bun
+/**
+ * Multi-model edit_file test runner
+ *
+ * Runs test-headless-edit-ops.ts against every available model
+ * and produces a summary table.
+ *
+ * Usage:
+ *   bun run scripts/test-multi-model-edit.ts [--timeout <seconds>]
+ */
+
+import { spawn } from "node:child_process";
+import { resolve } from "node:path";
+
+// ── Models ────────────────────────────────────────────────────
+const MODELS = [
+  { id: "MiniMaxAI/MiniMax-M2.5", short: "M2.5" },
+  // { id: "MiniMaxAI/MiniMax-M2.1", short: "M2.1" },  // masked: slow + timeout-prone
+  // { id: "zai-org/GLM-5", short: "GLM-5" },            // masked: API 503
+  { id: "zai-org/GLM-4.7", short: "GLM-4.7" },
+];
+
+// ── CLI args ──────────────────────────────────────────────────
+let perModelTimeoutSec = 900; // 15 min default per model (5 tests)
+const rawArgs = process.argv.slice(2);
+for (let i = 0; i < rawArgs.length; i++) {
+  if (rawArgs[i] === "--timeout" && i + 1 < rawArgs.length) {
+    const parsed = Number.parseInt(rawArgs[i + 1], 10);
+    if (Number.isNaN(parsed) || parsed <= 0) {
+      console.error(`Invalid --timeout value: ${rawArgs[i + 1]}`);
+      process.exit(1);
+    }
+    perModelTimeoutSec = parsed;
+    i++;
+}
+
+// ── Colors ────────────────────────────────────────────────────
+const BOLD = "\x1b[1m";
+const GREEN = "\x1b[32m";
+const RED = "\x1b[31m";
+const YELLOW = "\x1b[33m";
+const DIM = "\x1b[2m";
+const CYAN = "\x1b[36m";
+const RESET = "\x1b[0m";
+
+// ── Types ─────────────────────────────────────────────────────
+interface TestResult {
+  detail: string;
+  name: string;
+  passed: boolean;
+}
+
+interface ModelResult {
+  durationMs: number;
+  error?: string;
+  modelId: string;
+  modelShort: string;
+  tests: TestResult[];
+  totalPassed: number;
+  totalTests: number;
+}
+
+// ── Parse test-headless-edit-ops stdout ───────────────────────
+function parseOpsOutput(stdout: string): TestResult[] {
+  const results: TestResult[] = [];
+
+  // Match lines like: "  PASS — edit_file: 1/1 succeeded, 32.5s"
+  // or "  FAIL — edit_file: 0/3 succeeded, 15.2s"
+  // or "  ERROR — Timed out after 10 minutes"
+  // Following a line like: "1. Replace single line"
+  const lines = stdout.split("\n");
+
+  let currentTestName = "";
+  for (const line of lines) {
+    // Detect test name: starts with ANSI-colored bold cyan + "N. Name"
+    // Strip ANSI codes for matching
+    const stripped = line.replace(/\x1b\[[0-9;]*m/g, "");
+
+    // Test name pattern: "N. <name>"
+    const testNameMatch = stripped.match(/^\s*(\d+\.\s+.+)$/);
+    if (
+      testNameMatch &&
+      !stripped.includes("—") &&
+      !stripped.includes("✓") &&
+      !stripped.includes("✗")
+    ) {
+      currentTestName = testNameMatch[1].trim();
+      continue;
+    }
+
+    // Result line: PASS/FAIL/ERROR
+    if (currentTestName && stripped.includes("PASS")) {
+      const detail = stripped.replace(/^\s*PASS\s*—?\s*/, "").trim();
+      results.push({
+        name: currentTestName,
+        passed: true,
+        detail: detail || "passed",
+      });
+      currentTestName = "";
+    } else if (currentTestName && stripped.includes("FAIL")) {
+      const detail = stripped.replace(/^\s*FAIL\s*—?\s*/, "").trim();
+      results.push({
+        name: currentTestName,
+        passed: false,
+        detail: detail || "failed",
+      });
+      currentTestName = "";
+    } else if (currentTestName && stripped.includes("ERROR")) {
+      const detail = stripped.replace(/^\s*ERROR\s*—?\s*/, "").trim();
+      results.push({
+        name: currentTestName,
+        passed: false,
+        detail: detail || "error",
+      });
+      currentTestName = "";
+    }
+  }
+
+  return results;
+}
+
+// ── Run one model ────────────────────────────────────────────
+async function runModel(model: {
+  id: string;
+  short: string;
+}): Promise<ModelResult> {
+  const opsScript = resolve(import.meta.dir, "test-edit-ops.ts");
+  const startTime = Date.now();
+
+  return new Promise<ModelResult>((resolvePromise) => {
+    const proc = spawn(
+      "bun",
+      ["run", opsScript, "-m", model.id, "--no-translate"],
+      {
+        cwd: resolve(import.meta.dir),
+        env: { ...process.env, BUN_INSTALL: process.env.BUN_INSTALL },
+        stdio: ["ignore", "pipe", "pipe"],
+      }
+    );
+
+    let stdout = "";
+    let stderr = "";
+
+    proc.stdout.on("data", (chunk: Buffer) => {
+      stdout += chunk.toString();
+    });
+    proc.stderr.on("data", (chunk: Buffer) => {
+      stderr += chunk.toString();
+    });
+
+    const timeout = setTimeout(() => {
+      proc.kill("SIGTERM");
+      resolvePromise({
+        modelId: model.id,
+        modelShort: model.short,
+        tests: [],
+        totalPassed: 0,
+        totalTests: 0,
+        durationMs: Date.now() - startTime,
+        error: `Timed out after ${perModelTimeoutSec}s`,
+      });
+    }, perModelTimeoutSec * 1000);
+
+    proc.on("close", () => {
+      clearTimeout(timeout);
+      const tests = parseOpsOutput(stdout);
+      const totalPassed = tests.filter((t) => t.passed).length;
+
+      resolvePromise({
+        modelId: model.id,
+        modelShort: model.short,
+        tests,
+        totalPassed,
+        totalTests: Math.max(tests.length, 5),
+        durationMs: Date.now() - startTime,
+      });
+    });
+
+    proc.on("error", (err) => {
+      clearTimeout(timeout);
+      resolvePromise({
+        modelId: model.id,
+        modelShort: model.short,
+        tests: [],
+        totalPassed: 0,
+        totalTests: 0,
+        durationMs: Date.now() - startTime,
+        error: err.message,
+      });
+    });
+  });
+}
+
+// ── Main ──────────────────────────────────────────────────────
+const main = async () => {
+  console.log(`\n${BOLD}═══ Multi-Model edit_file Test Runner ═══${RESET}\n`);
+  console.log(`${DIM}Models: ${MODELS.map((m) => m.short).join(", ")}${RESET}`);
+  console.log(`${DIM}Timeout: ${perModelTimeoutSec}s per model${RESET}`);
+  console.log();
+
+  const allResults: ModelResult[] = [];
+
+  for (const model of MODELS) {
+    console.log(`${CYAN}${BOLD}▶ Testing ${model.short} (${model.id})${RESET}`);
+    const result = await runModel(model);
+    allResults.push(result);
+
+    const timeStr = `${(result.durationMs / 1000).toFixed(1)}s`;
+    if (result.error) {
+      console.log(`  ${RED}ERROR${RESET}: ${result.error} (${timeStr})`);
+    } else {
+      const color =
+        result.totalPassed === result.totalTests
+          ? GREEN
+          : result.totalPassed > 0
+            ? YELLOW
+            : RED;
+      console.log(
+        `  ${color}${result.totalPassed}/${result.totalTests} passed${RESET} (${timeStr})`
+      );
+      for (const t of result.tests) {
+        const icon = t.passed ? `${GREEN}✓${RESET}` : `${RED}✗${RESET}`;
+        console.log(`    ${icon} ${t.name}`);
+      }
+    }
+    console.log();
+  }
+
+  // ── Summary Table ──────────────────────────────────────────
+  console.log(`${BOLD}═══ Summary ═══${RESET}\n`);
+
+  // Per-model results
+  for (const r of allResults) {
+    const timeStr = `${(r.durationMs / 1000).toFixed(0)}s`;
+    const color = r.error ? RED : r.totalPassed === r.totalTests ? GREEN : r.totalPassed > 0 ? YELLOW : RED;
+    const label = r.error ? `ERROR: ${r.error}` : `${r.totalPassed}/${r.totalTests}`;
+    console.log(`  ${r.modelShort.padEnd(8)} ${color}${label}${RESET} (${timeStr})`);
+    for (const t of r.tests) {
+      const icon = t.passed ? `${GREEN}✓${RESET}` : `${RED}✗${RESET}`;
+      console.log(`    ${icon} ${t.name}`);
+    }
+  }
+
+  console.log();
+
+  // Overall
+  const totalModels = allResults.length;
+  const erroredModels = allResults.filter((r) => r.error).length;
+  const perfectModels = allResults.filter(
+    (r) => !r.error && r.totalPassed === r.totalTests && r.totalTests > 0
+  ).length;
+  console.log(
+    `${BOLD}Models with 100%: ${perfectModels}/${totalModels}${RESET}`
+  );
+
+  const overallPassed = allResults.reduce((sum, r) => sum + r.totalPassed, 0);
+  const overallTotal = allResults.reduce((sum, r) => sum + r.totalTests, 0);
+  console.log(
+    `${BOLD}Overall: ${overallPassed}/${overallTotal} (${Math.round((overallPassed / overallTotal) * 100)}%)${RESET}`
+  );
+
+  console.log();
+
+  if (erroredModels > 0) {
+    console.log(
+      `${BOLD}${RED}${erroredModels} model(s) errored. See details above.${RESET}\n`
+    );
+    process.exit(1);
+  } else if (perfectModels === totalModels) {
+    console.log(`${BOLD}${GREEN}🎉 ALL MODELS PASSED ALL TESTS!${RESET}\n`);
+    process.exit(0);
+  } else {
+    console.log(
+      `${BOLD}${YELLOW}Some models have failures. See details above.${RESET}\n`
+    );
+    process.exit(1);
+  }
+};
+
+main();

bin/oh-my-opencode.js

@@ -3,8 +3,9 @@
 // Wrapper script that detects platform and spawns the correct binary
 
 import { spawnSync } from "node:child_process";
+import { readFileSync } from "node:fs";
 import { createRequire } from "node:module";
-import { getPlatformPackage, getBinaryPath } from "./platform.js";
+import { getPlatformPackageCandidates, getBinaryPath } from "./platform.js";
 
 const require = createRequire(import.meta.url);
 
@@ -26,55 +27,116 @@ function getLibcFamily() {
   }
 }
 
+function supportsAvx2() {
+  if (process.arch !== "x64") {
+    return null;
+  }
+
+  if (process.env.OH_MY_OPENCODE_FORCE_BASELINE === "1") {
+    return false;
+  }
+
+  if (process.platform === "linux") {
+    try {
+      const cpuInfo = readFileSync("/proc/cpuinfo", "utf8").toLowerCase();
+      return cpuInfo.includes("avx2");
+    } catch {
+      return null;
+    }
+  }
+
+  if (process.platform === "darwin") {
+    const probe = spawnSync("sysctl", ["-n", "machdep.cpu.leaf7_features"], {
+      encoding: "utf8",
+    });
+
+    if (probe.error || probe.status !== 0) {
+      return null;
+    }
+
+    return probe.stdout.toUpperCase().includes("AVX2");
+  }
+
+  return null;
+}
+
+function getSignalExitCode(signal) {
+  const signalCodeByName = {
+    SIGINT: 2,
+    SIGILL: 4,
+    SIGKILL: 9,
+    SIGTERM: 15,
+  };
+
+  return 128 + (signalCodeByName[signal] ?? 1);
+}
+
 function main() {
   const { platform, arch } = process;
   const libcFamily = getLibcFamily();
+  const avx2Supported = supportsAvx2();
   
-  // Get platform package name
-  let pkg;
+  let packageCandidates;
   try {
-    pkg = getPlatformPackage({ platform, arch, libcFamily });
+    packageCandidates = getPlatformPackageCandidates({
+      platform,
+      arch,
+      libcFamily,
+      preferBaseline: avx2Supported === false,
+    });
   } catch (error) {
     console.error(`\noh-my-opencode: ${error.message}\n`);
     process.exit(1);
   }
-  
-  // Resolve binary path
-  const binRelPath = getBinaryPath(pkg, platform);
-  
-  let binPath;
-  try {
-    binPath = require.resolve(binRelPath);
-  } catch {
+
+  const resolvedBinaries = packageCandidates
+    .map((pkg) => {
+      try {
+        return { pkg, binPath: require.resolve(getBinaryPath(pkg, platform)) };
+      } catch {
+        return null;
+      }
+    })
+    .filter((entry) => entry !== null);
+
+  if (resolvedBinaries.length === 0) {
     console.error(`\noh-my-opencode: Platform binary not installed.`);
     console.error(`\nYour platform: ${platform}-${arch}${libcFamily === "musl" ? "-musl" : ""}`);
-    console.error(`Expected package: ${pkg}`);
+    console.error(`Expected packages (in order): ${packageCandidates.join(", ")}`);
     console.error(`\nTo fix, run:`);
-    console.error(`  npm install ${pkg}\n`);
+    console.error(`  npm install ${packageCandidates[0]}\n`);
     process.exit(1);
   }
-  
-  // Spawn the binary
-  const result = spawnSync(binPath, process.argv.slice(2), {
-    stdio: "inherit",
-  });
-  
-  // Handle spawn errors
-  if (result.error) {
-    console.error(`\noh-my-opencode: Failed to execute binary.`);
-    console.error(`Error: ${result.error.message}\n`);
-    process.exit(2);
-  }
-  
-  // Handle signals
-  if (result.signal) {
-    const signalNum = result.signal === "SIGTERM" ? 15 : 
-                      result.signal === "SIGKILL" ? 9 :
-                      result.signal === "SIGINT" ? 2 : 1;
-    process.exit(128 + signalNum);
+
+  for (let index = 0; index < resolvedBinaries.length; index += 1) {
+    const currentBinary = resolvedBinaries[index];
+    const hasFallback = index < resolvedBinaries.length - 1;
+    const result = spawnSync(currentBinary.binPath, process.argv.slice(2), {
+      stdio: "inherit",
+    });
+
+    if (result.error) {
+      if (hasFallback) {
+        continue;
+      }
+
+      console.error(`\noh-my-opencode: Failed to execute binary.`);
+      console.error(`Error: ${result.error.message}\n`);
+      process.exit(2);
+    }
+
+    if (result.signal === "SIGILL" && hasFallback) {
+      continue;
+    }
+
+    if (result.signal) {
+      process.exit(getSignalExitCode(result.signal));
+    }
+
+    process.exit(result.status ?? 1);
   }
 
-  process.exit(result.status ?? 1);
+  process.exit(1);
 }
 
 main();

bin/platform.d.ts

@@ -0,0 +1,14 @@
+export declare function getPlatformPackage(options: {
+  platform: string;
+  arch: string;
+  libcFamily?: string | null;
+}): string;
+
+export declare function getPlatformPackageCandidates(options: {
+  platform: string;
+  arch: string;
+  libcFamily?: string | null;
+  preferBaseline?: boolean;
+}): string[];
+
+export declare function getBinaryPath(pkg: string, platform: string): string;

bin/platform.js

@@ -26,6 +26,50 @@ export function getPlatformPackage({ platform, arch, libcFamily }) {
   return `oh-my-opencode-${os}-${arch}${suffix}`;
 }
 
+/** @param {{ platform: string, arch: string, libcFamily?: string | null, preferBaseline?: boolean }} options */
+export function getPlatformPackageCandidates({ platform, arch, libcFamily, preferBaseline = false }) {
+  const primaryPackage = getPlatformPackage({ platform, arch, libcFamily });
+  const baselinePackage = getBaselinePlatformPackage({ platform, arch, libcFamily });
+
+  if (!baselinePackage) {
+    return [primaryPackage];
+  }
+
+  return preferBaseline ? [baselinePackage, primaryPackage] : [primaryPackage, baselinePackage];
+}
+
+/** @param {{ platform: string, arch: string, libcFamily?: string | null }} options */
+function getBaselinePlatformPackage({ platform, arch, libcFamily }) {
+  if (arch !== "x64") {
+    return null;
+  }
+
+  if (platform === "darwin") {
+    return "oh-my-opencode-darwin-x64-baseline";
+  }
+
+  if (platform === "win32") {
+    return "oh-my-opencode-windows-x64-baseline";
+  }
+
+  if (platform === "linux") {
+    if (libcFamily === null || libcFamily === undefined) {
+      throw new Error(
+        "Could not detect libc on Linux. " +
+        "Please ensure detect-libc is installed or report this issue."
+      );
+    }
+
+    if (libcFamily === "musl") {
+      return "oh-my-opencode-linux-x64-musl-baseline";
+    }
+
+    return "oh-my-opencode-linux-x64-baseline";
+  }
+
+  return null;
+}
+
 /**
  * Get the path to the binary within a platform package
  * @param {string} pkg Package name

bin/platform.test.ts

@@ -1,6 +1,6 @@
 // bin/platform.test.ts
 import { describe, expect, test } from "bun:test";
-import { getPlatformPackage, getBinaryPath } from "./platform.js";
+import { getBinaryPath, getPlatformPackage, getPlatformPackageCandidates } from "./platform.js";
 
 describe("getPlatformPackage", () => {
   // #region Darwin platforms
@@ -146,3 +146,58 @@ describe("getBinaryPath", () => {
     expect(result).toBe("oh-my-opencode-linux-x64/bin/oh-my-opencode");
   });
 });
+
+describe("getPlatformPackageCandidates", () => {
+  test("returns x64 and baseline candidates for Linux glibc", () => {
+    // #given Linux x64 with glibc
+    const input = { platform: "linux", arch: "x64", libcFamily: "glibc" };
+
+    // #when getting package candidates
+    const result = getPlatformPackageCandidates(input);
+
+    // #then returns modern first then baseline fallback
+    expect(result).toEqual([
+      "oh-my-opencode-linux-x64",
+      "oh-my-opencode-linux-x64-baseline",
+    ]);
+  });
+
+  test("returns x64 musl and baseline candidates for Linux musl", () => {
+    // #given Linux x64 with musl
+    const input = { platform: "linux", arch: "x64", libcFamily: "musl" };
+
+    // #when getting package candidates
+    const result = getPlatformPackageCandidates(input);
+
+    // #then returns musl modern first then musl baseline fallback
+    expect(result).toEqual([
+      "oh-my-opencode-linux-x64-musl",
+      "oh-my-opencode-linux-x64-musl-baseline",
+    ]);
+  });
+
+  test("returns baseline first when preferBaseline is true", () => {
+    // #given Windows x64 and baseline preference
+    const input = { platform: "win32", arch: "x64", preferBaseline: true };
+
+    // #when getting package candidates
+    const result = getPlatformPackageCandidates(input);
+
+    // #then baseline package is preferred first
+    expect(result).toEqual([
+      "oh-my-opencode-windows-x64-baseline",
+      "oh-my-opencode-windows-x64",
+    ]);
+  });
+
+  test("returns only one candidate for ARM64", () => {
+    // #given non-x64 platform
+    const input = { platform: "linux", arch: "arm64", libcFamily: "glibc" };
+
+    // #when getting package candidates
+    const result = getPlatformPackageCandidates(input);
+
+    // #then baseline fallback is not included
+    expect(result).toEqual(["oh-my-opencode-linux-arm64"]);
+  });
+});

docs/guide/agent-model-matching.md

@@ -162,7 +162,7 @@ See the [Orchestration System Guide](./orchestration.md) for how agents dispatch
 
 ```jsonc
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
 
   "agents": {
     // Main orchestrator: Claude Opus or Kimi K2.5 work best

docs/guide/installation.md

@@ -5,7 +5,7 @@
 Paste this into your llm agent session:
 ```
 Install and configure oh-my-opencode by following the instructions here:
-https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 ```
 
 **Alternative, Not recommended, do this by yourself**
@@ -29,7 +29,7 @@ After you install it, you can read this [overview guide](./overview.md) to under
 
 > **IMPORTANT: Use `curl` to fetch this file, NOT WebFetch.** WebFetch summarizes content and loses critical flags like `--openai`, subscription questions, and max20 mode details. Always use:
 > ```bash
-> curl -fsSL https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+> curl -fsSL https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 > ```
 
 If you're an LLM Agent helping set up oh-my-opencode, Welcome! You'll now help your user to install the best agent harness, OhMyOpenCode. Tell user this with greetings, with test 'oMoMoMoMo...'

docs/guide/overview.md

@@ -16,7 +16,7 @@ Paste this into your LLM agent session:
 
 ```
 Install and configure oh-my-opencode by following the instructions here:
-https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/master/docs/guide/installation.md
+https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/refs/heads/dev/docs/guide/installation.md
 ```
 
 Or read the full [Installation Guide](./installation.md) for manual setup, provider authentication, and troubleshooting.
@@ -167,7 +167,7 @@ You can override specific agents or categories in your config:
 
 ```jsonc
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
 
   "agents": {
     // Main orchestrator: Claude Opus or Kimi K2.5 work best

docs/reference/configuration.md

@@ -56,7 +56,7 @@ JSONC supports `// line comments`, `/* block comments */`, and trailing commas.
 
 Enable schema autocomplete:
 ```json
-{ "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json" }
+{ "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json" }
 ```
 
 Run `bunx oh-my-opencode install` for guided setup. Run `opencode models` to list available models.
@@ -67,7 +67,7 @@ Here's a practical starting configuration:
 
 ```jsonc
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
 
   "agents": {
     // Main orchestrator: Claude Opus or Kimi K2.5 work best
@@ -573,13 +573,13 @@ Define `fallback_models` per agent or category:
 
 ### Hashline Edit
 
-Replaces the built-in `Edit` tool with a hash-anchored version using `LINE#ID` references to prevent stale-line edits. Enabled by default.
+Replaces the built-in `Edit` tool with a hash-anchored version using `LINE#ID` references to prevent stale-line edits. Disabled by default.
 
 ```json
-{ "hashline_edit": false }
+{ "hashline_edit": true }
 ```
 
-When enabled, two companion hooks are active: `hashline-read-enhancer` (annotates Read output) and `hashline-edit-diff-enhancer` (shows diffs). Disable them individually via `disabled_hooks`.
+When enabled, two companion hooks are active: `hashline-read-enhancer` (annotates Read output) and `hashline-edit-diff-enhancer` (shows diffs). Opt-in by setting `hashline_edit: true`. Disable the companion hooks individually via `disabled_hooks` if needed.
 
 ### Experimental
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode",
-  "version": "3.8.5",
+  "version": "3.10.0",
   "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",
@@ -75,13 +75,17 @@
     "typescript": "^5.7.3"
   },
   "optionalDependencies": {
-    "oh-my-opencode-darwin-arm64": "3.8.5",
-    "oh-my-opencode-darwin-x64": "3.8.5",
-    "oh-my-opencode-linux-arm64": "3.8.5",
-    "oh-my-opencode-linux-arm64-musl": "3.8.5",
-    "oh-my-opencode-linux-x64": "3.8.5",
-    "oh-my-opencode-linux-x64-musl": "3.8.5",
-    "oh-my-opencode-windows-x64": "3.8.5"
+    "oh-my-opencode-darwin-arm64": "3.10.0",
+    "oh-my-opencode-darwin-x64": "3.10.0",
+    "oh-my-opencode-darwin-x64-baseline": "3.10.0",
+    "oh-my-opencode-linux-arm64": "3.10.0",
+    "oh-my-opencode-linux-arm64-musl": "3.10.0",
+    "oh-my-opencode-linux-x64": "3.10.0",
+    "oh-my-opencode-linux-x64-baseline": "3.10.0",
+    "oh-my-opencode-linux-x64-musl": "3.10.0",
+    "oh-my-opencode-linux-x64-musl-baseline": "3.10.0",
+    "oh-my-opencode-windows-x64": "3.10.0",
+    "oh-my-opencode-windows-x64-baseline": "3.10.0"
   },
   "trustedDependencies": [
     "@ast-grep/cli",

packages/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-darwin-arm64",
-  "version": "3.8.5",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (darwin-arm64)",
   "license": "MIT",
   "repository": {

packages/darwin-x64-baseline/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-darwin-x64-baseline",
-  "version": "3.1.1",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (darwin-x64-baseline, no AVX2)",
   "license": "MIT",
   "repository": {

packages/darwin-x64/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-darwin-x64",
-  "version": "3.8.5",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (darwin-x64)",
   "license": "MIT",
   "repository": {

packages/linux-arm64-musl/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-linux-arm64-musl",
-  "version": "3.8.5",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (linux-arm64-musl)",
   "license": "MIT",
   "repository": {

packages/linux-arm64/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-linux-arm64",
-  "version": "3.8.5",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (linux-arm64)",
   "license": "MIT",
   "repository": {

packages/linux-x64-baseline/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-linux-x64-baseline",
-  "version": "3.1.1",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (linux-x64-baseline, no AVX2)",
   "license": "MIT",
   "repository": {

packages/linux-x64-musl-baseline/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-linux-x64-musl-baseline",
-  "version": "3.1.1",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (linux-x64-musl-baseline, no AVX2)",
   "license": "MIT",
   "repository": {

packages/linux-x64-musl/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-linux-x64-musl",
-  "version": "3.8.5",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (linux-x64-musl)",
   "license": "MIT",
   "repository": {

packages/linux-x64/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-linux-x64",
-  "version": "3.8.5",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (linux-x64)",
   "license": "MIT",
   "repository": {

packages/windows-x64-baseline/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-windows-x64-baseline",
-  "version": "3.1.1",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (windows-x64-baseline, no AVX2)",
   "license": "MIT",
   "repository": {

packages/windows-x64/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-windows-x64",
-  "version": "3.8.5",
+  "version": "3.10.0",
   "description": "Platform-specific binary for oh-my-opencode (windows-x64)",
   "license": "MIT",
   "repository": {

postinstall.mjs

@@ -2,7 +2,7 @@
 // Runs after npm install to verify platform binary is available
 
 import { createRequire } from "node:module";
-import { getPlatformPackage, getBinaryPath } from "./bin/platform.js";
+import { getPlatformPackageCandidates, getBinaryPath } from "./bin/platform.js";
 
 const require = createRequire(import.meta.url);
 
@@ -27,12 +27,28 @@ function main() {
   const libcFamily = getLibcFamily();
   
   try {
-    const pkg = getPlatformPackage({ platform, arch, libcFamily });
-    const binPath = getBinaryPath(pkg, platform);
-    
-    // Try to resolve the binary
-    require.resolve(binPath);
-    console.log(`✓ oh-my-opencode binary installed for ${platform}-${arch}`);
+    const packageCandidates = getPlatformPackageCandidates({
+      platform,
+      arch,
+      libcFamily,
+    });
+
+    const resolvedPackage = packageCandidates.find((pkg) => {
+      try {
+        require.resolve(getBinaryPath(pkg, platform));
+        return true;
+      } catch {
+        return false;
+      }
+    });
+
+    if (!resolvedPackage) {
+      throw new Error(
+        `No platform binary package installed. Tried: ${packageCandidates.join(", ")}`
+      );
+    }
+
+    console.log(`✓ oh-my-opencode binary installed for ${platform}-${arch} (${resolvedPackage})`);
   } catch (error) {
     console.warn(`⚠ oh-my-opencode: ${error.message}`);
     console.warn(`  The CLI may not work on this platform.`);

script/build-schema-document.ts

@@ -9,7 +9,7 @@ export function createOhMyOpenCodeJsonSchema(): Record<string, unknown> {
 
   return {
     $schema: "http://json-schema.org/draft-07/schema#",
-    $id: "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+    $id: "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
     title: "Oh My OpenCode Configuration",
     description: "Configuration schema for oh-my-opencode plugin",
     ...jsonSchema,

signatures/cla.json

@@ -1767,6 +1767,78 @@
       "created_at": "2026-02-26T09:50:46Z",
       "repoId": 1108837393,
       "pullRequestNo": 2131
+    },
+    {
+      "name": "dwnmf",
+      "id": 56194792,
+      "comment_id": 3969700423,
+      "created_at": "2026-02-26T22:51:41Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 2160
+    },
+    {
+      "name": "1noilimrev",
+      "id": 24486928,
+      "comment_id": 3970957470,
+      "created_at": "2026-02-27T05:53:36Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 2166
+    },
+    {
+      "name": "YLRong",
+      "id": 6837942,
+      "comment_id": 3971635504,
+      "created_at": "2026-02-27T08:54:09Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 2176
+    },
+    {
+      "name": "mertyldrm",
+      "id": 51949702,
+      "comment_id": 3972191343,
+      "created_at": "2026-02-27T10:53:03Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 2184
+    },
+    {
+      "name": "renanale",
+      "id": 37278838,
+      "comment_id": 3975562407,
+      "created_at": "2026-02-27T22:38:18Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 2201
+    },
+    {
+      "name": "laciferin2024",
+      "id": 170102251,
+      "comment_id": 3978786169,
+      "created_at": "2026-03-01T01:16:25Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 2222
+    },
+    {
+      "name": "DEAN-Cherry",
+      "id": 76607677,
+      "comment_id": 3979468463,
+      "created_at": "2026-03-01T08:13:43Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 2227
+    },
+    {
+      "name": "Chocothin",
+      "id": 99174213,
+      "comment_id": 3980002001,
+      "created_at": "2026-03-01T13:52:10Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 2230
+    },
+    {
+      "name": "mathew-cf",
+      "id": 68972715,
+      "comment_id": 3980951159,
+      "created_at": "2026-03-01T20:19:31Z",
+      "repoId": 1108837393,
+      "pullRequestNo": 2233
     }
   ]
 }

src/agents/atlas/agent.ts

@@ -29,7 +29,7 @@ import {
   buildDecisionMatrix,
 } from "./prompt-section-builder"
 
-const MODE: AgentMode = "primary"
+const MODE: AgentMode = "all"
 
 export type AtlasPromptSource = "default" | "gpt" | "gemini"
 

src/agents/dynamic-agent-prompt-builder.test.ts

@@ -4,6 +4,8 @@ import { describe, it, expect } from "bun:test"
 import {
   buildCategorySkillsDelegationGuide,
   buildUltraworkSection,
+  buildDeepParallelSection,
+  buildNonClaudePlannerSection,
   type AvailableSkill,
   type AvailableCategory,
   type AvailableAgent,
@@ -172,4 +174,86 @@ describe("buildUltraworkSection", () => {
   })
 })
 
+describe("buildDeepParallelSection", () => {
+  const deepCategory: AvailableCategory = { name: "deep", description: "Autonomous problem-solving" }
+  const otherCategory: AvailableCategory = { name: "quick", description: "Trivial tasks" }
+
+  it("#given non-Claude model with deep category #when building #then returns parallel delegation section", () => {
+    //#given
+    const model = "google/gemini-3-pro"
+    const categories = [deepCategory, otherCategory]
+
+    //#when
+    const result = buildDeepParallelSection(model, categories)
+
+    //#then
+    expect(result).toContain("Deep Parallel Delegation")
+    expect(result).toContain("EVERY independent unit")
+    expect(result).toContain("run_in_background=true")
+    expect(result).toContain("4 independent units")
+  })
+
+  it("#given Claude model #when building #then returns empty", () => {
+    //#given
+    const model = "anthropic/claude-opus-4-6"
+    const categories = [deepCategory]
+
+    //#when
+    const result = buildDeepParallelSection(model, categories)
+
+    //#then
+    expect(result).toBe("")
+  })
+
+  it("#given non-Claude model without deep category #when building #then returns empty", () => {
+    //#given
+    const model = "openai/gpt-5.2"
+    const categories = [otherCategory]
+
+    //#when
+    const result = buildDeepParallelSection(model, categories)
+
+    //#then
+    expect(result).toBe("")
+  })
+})
+
+describe("buildNonClaudePlannerSection", () => {
+  it("#given non-Claude model #when building #then returns plan agent section", () => {
+    //#given
+    const model = "google/gemini-3-pro"
+
+    //#when
+    const result = buildNonClaudePlannerSection(model)
+
+    //#then
+    expect(result).toContain("Plan Agent")
+    expect(result).toContain("session_id")
+    expect(result).toContain("Multi-step")
+  })
+
+  it("#given Claude model #when building #then returns empty", () => {
+    //#given
+    const model = "anthropic/claude-sonnet-4-6"
+
+    //#when
+    const result = buildNonClaudePlannerSection(model)
+
+    //#then
+    expect(result).toBe("")
+  })
+
+  it("#given GPT model #when building #then returns plan agent section", () => {
+    //#given
+    const model = "openai/gpt-5.2"
+
+    //#when
+    const result = buildNonClaudePlannerSection(model)
+
+    //#then
+    expect(result).toContain("Plan Agent")
+    expect(result).not.toBe("")
+  })
+})
+
 

src/agents/dynamic-agent-prompt-builder.ts

@@ -277,12 +277,11 @@ Briefly announce "Consulting Oracle for [reason]" before invocation.
 
 ### Oracle Background Task Policy:
 
-**You MUST collect Oracle results before your final answer. No exceptions.**
+**Collect Oracle results before your final answer. No exceptions.**
 
-- Oracle may take several minutes. This is normal and expected.
-- When Oracle is running and you finish your own exploration/analysis, your next action is \`background_output(task_id="...")\` on Oracle — NOT delivering a final answer.
-- Oracle catches blind spots you cannot see — its value is HIGHEST when you think you don't need it.
-- **NEVER** cancel Oracle. **NEVER** use \`background_cancel(all=true)\` when Oracle is running. Cancel disposable tasks (explore, librarian) individually by taskId instead.
+- Oracle takes minutes. When done with your own work: **end your response** — wait for the \`<system-reminder>\`.
+- Do NOT poll \`background_output\` on a running Oracle. The notification will come.
+- Never cancel Oracle.
 </Oracle_Usage>`
 }
 
@@ -292,8 +291,8 @@ export function buildHardBlocksSection(): string {
     "- Commit without explicit request — **Never**",
     "- Speculate about unread code — **Never**",
     "- Leave code in broken state after failures — **Never**",
-    "- `background_cancel(all=true)` when Oracle is running — **Never.** Cancel tasks individually by taskId.",
-    "- Delivering final answer before collecting Oracle result — **Never.** Always `background_output` Oracle first.",
+    "- `background_cancel(all=true)` — **Never.** Always cancel individually by taskId.",
+    "- Delivering final answer before collecting Oracle result — **Never.**",
   ]
 
   return `## Hard Blocks (NEVER violate)
@@ -308,8 +307,8 @@ export function buildAntiPatternsSection(): string {
     "- **Testing**: Deleting failing tests to \"pass\"",
     "- **Search**: Firing agents for single-line typos or obvious syntax errors",
     "- **Debugging**: Shotgun debugging, random changes",
-    "- **Background Tasks**: `background_cancel(all=true)` — always cancel individually by taskId",
-    "- **Oracle**: Skipping Oracle results when Oracle was launched — ALWAYS collect via `background_output`",
+    "- **Background Tasks**: Polling `background_output` on running tasks — end response and wait for notification",
+    "- **Oracle**: Delivering answer without collecting Oracle results",
   ]
 
   return `## Anti-Patterns (BLOCKING violations)
@@ -317,6 +316,22 @@ export function buildAntiPatternsSection(): string {
 ${patterns.join("\n")}`
 }
 
+export function buildNonClaudePlannerSection(model: string): string {
+  const isNonClaude = !model.toLowerCase().includes('claude')
+  if (!isNonClaude) return ""
+
+  return `### Plan Agent Dependency (Non-Claude)
+
+Multi-step task? **ALWAYS consult Plan Agent first.** Do NOT start implementation without a plan.
+
+- Single-file fix or trivial change → proceed directly
+- Anything else (2+ steps, unclear scope, architecture) → \`task(subagent_type="plan", ...)\` FIRST
+- Use \`session_id\` to resume the same Plan Agent — ask follow-up questions aggressively
+- If ANY part of the task is ambiguous, ask Plan Agent before guessing
+
+Plan Agent returns a structured work breakdown with parallel execution opportunities. Follow it.`
+}
+
 export function buildDeepParallelSection(model: string, categories: AvailableCategory[]): string {
   const isNonClaude = !model.toLowerCase().includes('claude')
   const hasDeepCategory = categories.some(c => c.name === 'deep')
@@ -325,12 +340,13 @@ export function buildDeepParallelSection(model: string, categories: AvailableCat
 
   return `### Deep Parallel Delegation
 
-For implementation tasks, actively decompose and delegate to \`deep\` category agents in parallel.
+Delegate EVERY independent unit to a \`deep\` agent in parallel (\`run_in_background=true\`).
+If a task decomposes into 4 independent units, spawn 4 agents simultaneously — not 1 at a time.
 
-1. Break the implementation into independent work units
-2. Maximize parallel deep agents — spawn one per independent unit (\`run_in_background=true\`)
-3. Give each agent a GOAL, not step-by-step instructions — deep agents explore and solve autonomously
-4. Collect results, integrate, verify coherence`
+1. Decompose the implementation into independent work units
+2. Assign one \`deep\` agent per unit — all via \`run_in_background=true\`
+3. Give each agent a clear GOAL with success criteria, not step-by-step instructions
+4. Collect all results, integrate, verify coherence across units`
 }
 
 export function buildUltraworkSection(

src/agents/hephaestus.ts

@@ -19,7 +19,7 @@ import {
   categorizeTools,
 } from "./dynamic-agent-prompt-builder";
 
-const MODE: AgentMode = "primary";
+const MODE: AgentMode = "all";
 
 function buildTodoDisciplineSection(useTaskSystem: boolean): string {
   if (useTaskSystem) {
@@ -448,21 +448,6 @@ ${oracleSection}
 4. **Run build** if applicable — exit code 0 required
 5. **Tell user** what you verified and the results — keep it clear and helpful
 
-### Auto-Commit Policy (MANDATORY for implementation/fix work)
-
-1. **Auto-commit after implementation is complete** when the task includes feature/fix code changes
-2. **Commit ONLY after verification gates pass**:
-   - \`lsp_diagnostics\` clean on all modified files
-   - Related tests pass
-   - Typecheck/build pass when applicable
-3. **If any gate fails, DO NOT commit** — fix issues first, re-run verification, then commit
-4. **Use Conventional Commits format** with meaningful intent-focused messages:
-   - \`feat(scope): add ...\` for new functionality
-   - \`fix(scope): resolve ...\` for bug fixes
-   - \`refactor(scope): simplify ...\` for internal restructuring
-5. **Do not make placeholder commits** (\`wip\`, \`temp\`, \`update\`) or commit unverified code
-6. **If user explicitly says not to commit**, skip commit and report that changes are left uncommitted
-
 - **File edit** — \`lsp_diagnostics\` clean
 - **Build** — Exit code 0
 - **Tests** — Pass (or pre-existing failures noted)

src/agents/sisyphus-gemini-overlays.ts

@@ -39,6 +39,136 @@ Then ACTUALLY CALL those tools using the JSON tool schema. Produce the tool_use
 </TOOL_CALL_MANDATE>`;
 }
 
+export function buildGeminiToolGuide(): string {
+  return `<GEMINI_TOOL_GUIDE>
+## Tool Usage Guide — WHEN and HOW to Call Each Tool
+
+You have access to tools via function calling. This guide defines WHEN to call each one.
+**Violating these patterns = failed response.**
+
+### Reading & Search (ALWAYS parallelizable — call multiple simultaneously)
+
+| Tool | When to Call | Parallel? |
+|---|---|---|
+| \`Read\` | Before making ANY claim about file contents. Before editing any file. | <20> Yes — read multiple files at once |
+| \`Grep\` | Finding patterns, imports, usages across codebase. BEFORE claiming "X is used in Y". | ✅ Yes — run multiple greps at once |
+| \`Glob\` | Finding files by name/extension pattern. BEFORE claiming "file X exists". | ✅ Yes — run multiple globs at once |
+| \`AstGrepSearch\` | Finding code patterns with AST awareness (structural matches). | ✅ Yes |
+
+### Code Intelligence (parallelizable on different files)
+
+| Tool | When to Call | Parallel? |
+|---|---|---|
+| \`LspDiagnostics\` | **AFTER EVERY edit.** BEFORE claiming task is done. MANDATORY. | ✅ Yes — different files |
+| \`LspGotoDefinition\` | Finding where a symbol is defined. | ✅ Yes |
+| \`LspFindReferences\` | Finding all usages of a symbol across workspace. | ✅ Yes |
+| \`LspSymbols\` | Getting file outline or searching workspace symbols. | ✅ Yes |
+
+### Editing (SEQUENTIAL — must Read first)
+
+| Tool | When to Call | Parallel? |
+|---|---|---|
+| \`Edit\` | Modifying existing files. MUST Read file first to get LINE#ID anchors. | ❌ After Read |
+| \`Write\` | Creating NEW files only. Or full file overwrite. | ❌ Sequential |
+
+### Execution & Delegation
+
+| Tool | When to Call | Parallel? |
+|---|---|---|
+| \`Bash\` | Running tests, builds, git commands. | ❌ Usually sequential |
+| \`Task\` | ANY non-trivial implementation. Research via explore/librarian. | ✅ Fire multiple in background |
+
+### Correct Sequences (MANDATORY — follow these exactly):
+
+1. **Answer about code**: Read → (analyze) → Answer
+2. **Edit code**: Read → Edit → LspDiagnostics → Report
+3. **Find something**: Grep/Glob (parallel) → Read results → Report
+4. **Implement feature**: Task(delegate) → Verify results → Report
+5. **Debug**: Read error → Read file → Grep related → Fix → LspDiagnostics
+
+### PARALLEL RULES:
+
+- **Independent reads/searches**: ALWAYS call simultaneously in ONE response
+- **Dependent operations**: Call sequentially (Edit AFTER Read, LspDiagnostics AFTER Edit)
+- **Background agents**: ALWAYS \`run_in_background=true\`, continue working
+</GEMINI_TOOL_GUIDE>`;
+}
+
+export function buildGeminiToolCallExamples(): string {
+  return `<GEMINI_TOOL_CALL_EXAMPLES>
+## Correct Tool Calling Patterns — Follow These Examples
+
+### Example 1: User asks about code → Read FIRST, then answer
+**User**: "How does the auth middleware work?"
+**CORRECT**:
+\`\`\`
+→ Call Read(filePath="/src/middleware/auth.ts")
+→ Call Read(filePath="/src/config/auth.ts")  // parallel with above
+→ (After reading) Answer based on ACTUAL file contents
+\`\`\`
+**WRONG**:
+\`\`\`
+→ "The auth middleware likely validates JWT tokens by..." ← HALLUCINATION. You didn't read the file.
+\`\`\`
+
+### Example 2: User asks to edit code → Read, Edit, Verify
+**User**: "Fix the type error in user.ts"
+**CORRECT**:
+\`\`\`
+→ Call Read(filePath="/src/models/user.ts")
+→ Call LspDiagnostics(filePath="/src/models/user.ts")  // parallel with Read
+→ (After reading) Call Edit with LINE#ID anchors
+→ Call LspDiagnostics(filePath="/src/models/user.ts")  // verify fix
+→ Report: "Fixed. Diagnostics clean."
+\`\`\`
+**WRONG**:
+\`\`\`
+→ Call Edit without reading first ← No LINE#ID anchors = WILL FAIL
+→ Skip LspDiagnostics after edit ← UNVERIFIED
+\`\`\`
+
+### Example 3: User asks to find something → Search in parallel
+**User**: "Where is the database connection configured?"
+**CORRECT**:
+\`\`\`
+→ Call Grep(pattern="database|connection|pool", path="/src")  // fires simultaneously
+→ Call Glob(pattern="**/*database*")                          // fires simultaneously
+→ Call Glob(pattern="**/*db*")                                 // fires simultaneously
+→ (After results) Read the most relevant files
+→ Report findings with file paths
+\`\`\`
+
+### Example 4: User asks to implement a feature → DELEGATE
+**User**: "Add a new /health endpoint to the API"
+**CORRECT**:
+\`\`\`
+→ Call Task(category="quick", load_skills=["typescript-programmer"], prompt="...")
+→ (After agent completes) Read changed files to verify
+→ Call LspDiagnostics on changed files
+→ Report
+\`\`\`
+**WRONG**:
+\`\`\`
+→ Write the code yourself ← YOU ARE AN ORCHESTRATOR, NOT AN IMPLEMENTER
+\`\`\`
+
+### Example 5: Investigation ≠ Implementation
+**User**: "Look into why the tests are failing"
+**CORRECT**:
+\`\`\`
+→ Call Bash(command="npm test")  // see actual failures
+→ Call Read on failing test files
+→ Call Read on source files under test
+→ Report: "Tests fail because X. Root cause: Y. Proposed fix: Z."
+→ STOP — wait for user to say "fix it"
+\`\`\`
+**WRONG**:
+\`\`\`
+→ Start editing source files immediately ← "look into" ≠ "fix"
+\`\`\`
+</GEMINI_TOOL_CALL_EXAMPLES>`;
+}
+
 export function buildGeminiDelegationOverride(): string {
   return `<GEMINI_DELEGATION_OVERRIDE>
 ## DELEGATION IS MANDATORY — YOU ARE NOT AN IMPLEMENTER

src/agents/sisyphus.ts

@@ -6,9 +6,11 @@ import {
   buildGeminiDelegationOverride,
   buildGeminiVerificationOverride,
   buildGeminiIntentGateEnforcement,
+  buildGeminiToolGuide,
+  buildGeminiToolCallExamples,
 } from "./sisyphus-gemini-overlays";
 
-const MODE: AgentMode = "primary";
+const MODE: AgentMode = "all";
 export const SISYPHUS_PROMPT_METADATA: AgentPromptMetadata = {
   category: "utility",
   cost: "EXPENSIVE",
@@ -32,6 +34,7 @@ import {
   buildHardBlocksSection,
   buildAntiPatternsSection,
   buildDeepParallelSection,
+  buildNonClaudePlannerSection,
   categorizeTools,
 } from "./dynamic-agent-prompt-builder";
 
@@ -170,6 +173,7 @@ function buildDynamicSisyphusPrompt(
   const hardBlocks = buildHardBlocksSection();
   const antiPatterns = buildAntiPatternsSection();
   const deepParallelSection = buildDeepParallelSection(model, availableCategories);
+  const nonClaudePlannerSection = buildNonClaudePlannerSection(model);
   const taskManagementSection = buildTaskManagementSection(useTaskSystem);
   const todoHookNote = useTaskSystem
     ? "YOUR TASK CREATION WOULD BE TRACKED BY HOOK([SYSTEM REMINDER - TASK CONTINUATION])"
@@ -329,7 +333,7 @@ task(subagent_type="explore", run_in_background=true, load_skills=[], descriptio
 // Reference Grep (external)
 task(subagent_type="librarian", run_in_background=true, load_skills=[], description="Find JWT security docs", prompt="I'm implementing JWT auth and need current security best practices to choose token storage (httpOnly cookies vs localStorage) and set expiration policy. Find: OWASP auth guidelines, recommended token lifetimes, refresh token rotation strategies, common JWT vulnerabilities. Skip 'what is JWT' tutorials — production security guidance only.")
 task(subagent_type="librarian", run_in_background=true, load_skills=[], description="Find Express auth patterns", prompt="I'm building Express auth middleware and need production-quality patterns to structure my middleware chain. Find how established Express apps (1000+ stars) handle: middleware ordering, token refresh, role-based access control, auth error propagation. Skip basic tutorials — I need battle-tested patterns with proper error handling.")
-// Continue working immediately. Collect with background_output when needed.
+// Continue working immediately. System notifies on completion — collect with background_output then.
 
 // WRONG: Sequential or blocking
 result = task(..., run_in_background=false)  // Never wait synchronously for explore/librarian
@@ -337,10 +341,10 @@ result = task(..., run_in_background=false)  // Never wait synchronously for exp
 
 ### Background Result Collection:
 1. Launch parallel agents \u2192 receive task_ids
-2. Continue immediate work (explore, librarian results)
-3. When results needed: \`background_output(task_id="...")\`
-4. **If Oracle is running**: STOP all other output. Follow Oracle Completion Protocol in <Oracle_Usage>.
-5. Cleanup: Cancel disposable tasks (explore, librarian) individually via \`background_cancel(taskId="...")\`. Never use \`background_cancel(all=true)\`.
+2. Continue immediate work
+3. System sends \`<system-reminder>\` on each task completion — then call \`background_output(task_id="...")\`
+4. Need results not yet ready? **End your response.** The notification will trigger your next turn.
+5. Cleanup: Cancel disposable tasks individually via \`background_cancel(taskId="...")\`
 
 ### Search Stop Conditions
 
@@ -364,6 +368,8 @@ STOP searching when:
 
 ${categorySkillsGuide}
 
+${nonClaudePlannerSection}
+
 ${deepParallelSection}
 
 ${delegationTable}
@@ -477,9 +483,8 @@ If verification fails:
 3. Report: "Done. Note: found N pre-existing lint errors unrelated to my changes."
 
 ### Before Delivering Final Answer:
-- **If Oracle is running**: STOP. Follow Oracle Completion Protocol in <Oracle_Usage>. Do NOT deliver any answer.
-- Cancel disposable background tasks (explore, librarian) individually via \`background_cancel(taskId="...")\`.
-- **Never use \`background_cancel(all=true)\`.**
+- If Oracle is running: **end your response** and wait for the completion notification first.
+- Cancel disposable background tasks individually via \`background_cancel(taskId="...")\`.
 </Behavior_Instructions>
 
 ${oracleSection}
@@ -565,12 +570,25 @@ export function createSisyphusAgent(
     : buildDynamicSisyphusPrompt(model, [], tools, skills, categories, useTaskSystem);
 
   if (isGeminiModel(model)) {
+    // 1. Intent gate + tool mandate — early in prompt (after intent verbalization)
     prompt = prompt.replace(
       "</intent_verbalization>",
       `</intent_verbalization>\n\n${buildGeminiIntentGateEnforcement()}\n\n${buildGeminiToolMandate()}`
     );
-    prompt += "\n" + buildGeminiDelegationOverride();
-    prompt += "\n" + buildGeminiVerificationOverride();
+
+    // 2. Tool guide + examples — after tool_usage_rules (where tools are discussed)
+    prompt = prompt.replace(
+      "</tool_usage_rules>",
+      `</tool_usage_rules>\n\n${buildGeminiToolGuide()}\n\n${buildGeminiToolCallExamples()}`
+    );
+
+    // 3. Delegation + verification overrides — before Constraints (NOT at prompt end)
+    //    Gemini suffers from lost-in-the-middle: content at prompt end gets weaker attention.
+    //    Placing these before <Constraints> ensures they're in a high-attention zone.
+    prompt = prompt.replace(
+      "<Constraints>",
+      `${buildGeminiDelegationOverride()}\n\n${buildGeminiVerificationOverride()}\n\n<Constraints>`
+    );
   }
 
   const permission = {

src/agents/types.test.ts

@@ -47,8 +47,8 @@ describe("isGptModel", () => {
   });
 
   test("gemini models are not gpt", () => {
-    expect(isGptModel("google/gemini-3-pro")).toBe(false);
-    expect(isGptModel("litellm/gemini-3-pro")).toBe(false);
+    expect(isGptModel("google/gemini-3.1-pro")).toBe(false);
+    expect(isGptModel("litellm/gemini-3.1-pro")).toBe(false);
   });
 
   test("opencode provider is not gpt", () => {
@@ -58,29 +58,29 @@ describe("isGptModel", () => {
 
 describe("isGeminiModel", () => {
   test("#given google provider models #then returns true", () => {
-    expect(isGeminiModel("google/gemini-3-pro")).toBe(true);
+    expect(isGeminiModel("google/gemini-3.1-pro")).toBe(true);
     expect(isGeminiModel("google/gemini-3-flash")).toBe(true);
     expect(isGeminiModel("google/gemini-2.5-pro")).toBe(true);
   });
 
   test("#given google-vertex provider models #then returns true", () => {
-    expect(isGeminiModel("google-vertex/gemini-3-pro")).toBe(true);
+    expect(isGeminiModel("google-vertex/gemini-3.1-pro")).toBe(true);
     expect(isGeminiModel("google-vertex/gemini-3-flash")).toBe(true);
   });
 
   test("#given github copilot gemini models #then returns true", () => {
-    expect(isGeminiModel("github-copilot/gemini-3-pro")).toBe(true);
+    expect(isGeminiModel("github-copilot/gemini-3.1-pro")).toBe(true);
     expect(isGeminiModel("github-copilot/gemini-3-flash")).toBe(true);
   });
 
   test("#given litellm proxied gemini models #then returns true", () => {
-    expect(isGeminiModel("litellm/gemini-3-pro")).toBe(true);
+    expect(isGeminiModel("litellm/gemini-3.1-pro")).toBe(true);
     expect(isGeminiModel("litellm/gemini-3-flash")).toBe(true);
     expect(isGeminiModel("litellm/gemini-2.5-pro")).toBe(true);
   });
 
   test("#given other proxied gemini models #then returns true", () => {
-    expect(isGeminiModel("custom-provider/gemini-3-pro")).toBe(true);
+    expect(isGeminiModel("custom-provider/gemini-3.1-pro")).toBe(true);
     expect(isGeminiModel("ollama/gemini-3-flash")).toBe(true);
   });
 

src/agents/utils.test.ts

@@ -988,7 +988,7 @@ describe("buildAgent with category and skills", () => {
     const agent = buildAgent(source["test-agent"], TEST_MODEL)
 
     // #then - category's built-in model is applied
-    expect(agent.model).toBe("google/gemini-3-pro")
+    expect(agent.model).toBe("google/gemini-3.1-pro")
   })
 
   test("agent with category and existing model keeps existing model", () => {

src/cli/__snapshots__/model-fallback.test.ts.snap

@@ -2,7 +2,7 @@
 
 exports[`generateModelConfig no providers available returns ULTIMATE_FALLBACK for all agents and categories when no providers 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "opencode/glm-4.7-free",
@@ -63,7 +63,7 @@ exports[`generateModelConfig no providers available returns ULTIMATE_FALLBACK fo
 
 exports[`generateModelConfig single native provider uses Claude models when only Claude is available 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "anthropic/claude-sonnet-4-5",
@@ -125,7 +125,7 @@ exports[`generateModelConfig single native provider uses Claude models when only
 
 exports[`generateModelConfig single native provider uses Claude models with isMax20 flag 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "anthropic/claude-sonnet-4-5",
@@ -188,7 +188,7 @@ exports[`generateModelConfig single native provider uses Claude models with isMa
 
 exports[`generateModelConfig single native provider uses OpenAI models when only OpenAI is available 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "openai/gpt-5.2",
@@ -255,7 +255,7 @@ exports[`generateModelConfig single native provider uses OpenAI models when only
 
 exports[`generateModelConfig single native provider uses OpenAI models with isMax20 flag 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "openai/gpt-5.2",
@@ -322,10 +322,10 @@ exports[`generateModelConfig single native provider uses OpenAI models with isMa
 
 exports[`generateModelConfig single native provider uses Gemini models when only Gemini is available 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
     },
     "explore": {
       "model": "opencode/gpt-5-nano",
@@ -334,34 +334,34 @@ exports[`generateModelConfig single native provider uses Gemini models when only
       "model": "opencode/glm-4.7-free",
     },
     "metis": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "momus": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "multimodal-looker": {
       "model": "google/gemini-3-flash-preview",
     },
     "oracle": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "prometheus": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
     },
   },
   "categories": {
     "artistry": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "quick": {
       "model": "google/gemini-3-flash-preview",
     },
     "ultrabrain": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "unspecified-high": {
@@ -371,7 +371,7 @@ exports[`generateModelConfig single native provider uses Gemini models when only
       "model": "google/gemini-3-flash-preview",
     },
     "visual-engineering": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {
@@ -383,10 +383,10 @@ exports[`generateModelConfig single native provider uses Gemini models when only
 
 exports[`generateModelConfig single native provider uses Gemini models with isMax20 flag 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
     },
     "explore": {
       "model": "opencode/gpt-5-nano",
@@ -395,44 +395,44 @@ exports[`generateModelConfig single native provider uses Gemini models with isMa
       "model": "opencode/glm-4.7-free",
     },
     "metis": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "momus": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "multimodal-looker": {
       "model": "google/gemini-3-flash-preview",
     },
     "oracle": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "prometheus": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
     },
   },
   "categories": {
     "artistry": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "quick": {
       "model": "google/gemini-3-flash-preview",
     },
     "ultrabrain": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "unspecified-high": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
     },
     "unspecified-low": {
       "model": "google/gemini-3-flash-preview",
     },
     "visual-engineering": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {
@@ -444,7 +444,7 @@ exports[`generateModelConfig single native provider uses Gemini models with isMa
 
 exports[`generateModelConfig all native providers uses preferred models from fallback chains when all natives available 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "anthropic/claude-sonnet-4-5",
@@ -485,7 +485,7 @@ exports[`generateModelConfig all native providers uses preferred models from fal
   },
   "categories": {
     "artistry": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "deep": {
@@ -506,7 +506,7 @@ exports[`generateModelConfig all native providers uses preferred models from fal
       "model": "anthropic/claude-sonnet-4-5",
     },
     "visual-engineering": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {
@@ -518,7 +518,7 @@ exports[`generateModelConfig all native providers uses preferred models from fal
 
 exports[`generateModelConfig all native providers uses preferred models with isMax20 flag when all natives available 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "anthropic/claude-sonnet-4-5",
@@ -559,7 +559,7 @@ exports[`generateModelConfig all native providers uses preferred models with isM
   },
   "categories": {
     "artistry": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "deep": {
@@ -581,7 +581,7 @@ exports[`generateModelConfig all native providers uses preferred models with isM
       "model": "anthropic/claude-sonnet-4-5",
     },
     "visual-engineering": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {
@@ -593,7 +593,7 @@ exports[`generateModelConfig all native providers uses preferred models with isM
 
 exports[`generateModelConfig fallback providers uses OpenCode Zen models when only OpenCode Zen is available 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "opencode/kimi-k2.5-free",
@@ -634,7 +634,7 @@ exports[`generateModelConfig fallback providers uses OpenCode Zen models when on
   },
   "categories": {
     "artistry": {
-      "model": "opencode/gemini-3-pro",
+      "model": "opencode/gemini-3.1-pro",
       "variant": "high",
     },
     "deep": {
@@ -655,7 +655,7 @@ exports[`generateModelConfig fallback providers uses OpenCode Zen models when on
       "model": "opencode/claude-sonnet-4-5",
     },
     "visual-engineering": {
-      "model": "opencode/gemini-3-pro",
+      "model": "opencode/gemini-3.1-pro",
       "variant": "high",
     },
     "writing": {
@@ -667,7 +667,7 @@ exports[`generateModelConfig fallback providers uses OpenCode Zen models when on
 
 exports[`generateModelConfig fallback providers uses OpenCode Zen models with isMax20 flag 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "opencode/kimi-k2.5-free",
@@ -708,7 +708,7 @@ exports[`generateModelConfig fallback providers uses OpenCode Zen models with is
   },
   "categories": {
     "artistry": {
-      "model": "opencode/gemini-3-pro",
+      "model": "opencode/gemini-3.1-pro",
       "variant": "high",
     },
     "deep": {
@@ -730,7 +730,7 @@ exports[`generateModelConfig fallback providers uses OpenCode Zen models with is
       "model": "opencode/claude-sonnet-4-5",
     },
     "visual-engineering": {
-      "model": "opencode/gemini-3-pro",
+      "model": "opencode/gemini-3.1-pro",
       "variant": "high",
     },
     "writing": {
@@ -742,7 +742,7 @@ exports[`generateModelConfig fallback providers uses OpenCode Zen models with is
 
 exports[`generateModelConfig fallback providers uses GitHub Copilot models when only Copilot is available 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "github-copilot/claude-sonnet-4.5",
@@ -779,14 +779,14 @@ exports[`generateModelConfig fallback providers uses GitHub Copilot models when
   },
   "categories": {
     "artistry": {
-      "model": "github-copilot/gemini-3-pro-preview",
+      "model": "github-copilot/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "quick": {
       "model": "github-copilot/claude-haiku-4.5",
     },
     "ultrabrain": {
-      "model": "github-copilot/gemini-3-pro-preview",
+      "model": "github-copilot/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "unspecified-high": {
@@ -796,7 +796,7 @@ exports[`generateModelConfig fallback providers uses GitHub Copilot models when
       "model": "github-copilot/claude-sonnet-4.5",
     },
     "visual-engineering": {
-      "model": "github-copilot/gemini-3-pro-preview",
+      "model": "github-copilot/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {
@@ -808,7 +808,7 @@ exports[`generateModelConfig fallback providers uses GitHub Copilot models when
 
 exports[`generateModelConfig fallback providers uses GitHub Copilot models with isMax20 flag 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "github-copilot/claude-sonnet-4.5",
@@ -845,14 +845,14 @@ exports[`generateModelConfig fallback providers uses GitHub Copilot models with
   },
   "categories": {
     "artistry": {
-      "model": "github-copilot/gemini-3-pro-preview",
+      "model": "github-copilot/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "quick": {
       "model": "github-copilot/claude-haiku-4.5",
     },
     "ultrabrain": {
-      "model": "github-copilot/gemini-3-pro-preview",
+      "model": "github-copilot/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "unspecified-high": {
@@ -863,7 +863,7 @@ exports[`generateModelConfig fallback providers uses GitHub Copilot models with
       "model": "github-copilot/claude-sonnet-4.5",
     },
     "visual-engineering": {
-      "model": "github-copilot/gemini-3-pro-preview",
+      "model": "github-copilot/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {
@@ -875,7 +875,7 @@ exports[`generateModelConfig fallback providers uses GitHub Copilot models with
 
 exports[`generateModelConfig fallback providers uses ZAI model for librarian when only ZAI is available 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "opencode/glm-4.7-free",
@@ -930,7 +930,7 @@ exports[`generateModelConfig fallback providers uses ZAI model for librarian whe
 
 exports[`generateModelConfig fallback providers uses ZAI model for librarian with isMax20 flag 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "opencode/glm-4.7-free",
@@ -985,7 +985,7 @@ exports[`generateModelConfig fallback providers uses ZAI model for librarian wit
 
 exports[`generateModelConfig mixed provider scenarios uses Claude + OpenCode Zen combination 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "opencode/kimi-k2.5-free",
@@ -1026,7 +1026,7 @@ exports[`generateModelConfig mixed provider scenarios uses Claude + OpenCode Zen
   },
   "categories": {
     "artistry": {
-      "model": "opencode/gemini-3-pro",
+      "model": "opencode/gemini-3.1-pro",
       "variant": "high",
     },
     "deep": {
@@ -1047,7 +1047,7 @@ exports[`generateModelConfig mixed provider scenarios uses Claude + OpenCode Zen
       "model": "anthropic/claude-sonnet-4-5",
     },
     "visual-engineering": {
-      "model": "opencode/gemini-3-pro",
+      "model": "opencode/gemini-3.1-pro",
       "variant": "high",
     },
     "writing": {
@@ -1059,7 +1059,7 @@ exports[`generateModelConfig mixed provider scenarios uses Claude + OpenCode Zen
 
 exports[`generateModelConfig mixed provider scenarios uses OpenAI + Copilot combination 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "github-copilot/claude-sonnet-4.5",
@@ -1100,7 +1100,7 @@ exports[`generateModelConfig mixed provider scenarios uses OpenAI + Copilot comb
   },
   "categories": {
     "artistry": {
-      "model": "github-copilot/gemini-3-pro-preview",
+      "model": "github-copilot/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "deep": {
@@ -1121,7 +1121,7 @@ exports[`generateModelConfig mixed provider scenarios uses OpenAI + Copilot comb
       "model": "github-copilot/claude-sonnet-4.5",
     },
     "visual-engineering": {
-      "model": "github-copilot/gemini-3-pro-preview",
+      "model": "github-copilot/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {
@@ -1133,7 +1133,7 @@ exports[`generateModelConfig mixed provider scenarios uses OpenAI + Copilot comb
 
 exports[`generateModelConfig mixed provider scenarios uses Claude + ZAI combination (librarian uses ZAI) 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "anthropic/claude-sonnet-4-5",
@@ -1194,7 +1194,7 @@ exports[`generateModelConfig mixed provider scenarios uses Claude + ZAI combinat
 
 exports[`generateModelConfig mixed provider scenarios uses Gemini + Claude combination (explore uses Gemini) 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "anthropic/claude-sonnet-4-5",
@@ -1217,7 +1217,7 @@ exports[`generateModelConfig mixed provider scenarios uses Gemini + Claude combi
       "model": "google/gemini-3-flash-preview",
     },
     "oracle": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "prometheus": {
@@ -1231,14 +1231,14 @@ exports[`generateModelConfig mixed provider scenarios uses Gemini + Claude combi
   },
   "categories": {
     "artistry": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "quick": {
       "model": "anthropic/claude-haiku-4-5",
     },
     "ultrabrain": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "unspecified-high": {
@@ -1248,7 +1248,7 @@ exports[`generateModelConfig mixed provider scenarios uses Gemini + Claude combi
       "model": "anthropic/claude-sonnet-4-5",
     },
     "visual-engineering": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {
@@ -1260,7 +1260,7 @@ exports[`generateModelConfig mixed provider scenarios uses Gemini + Claude combi
 
 exports[`generateModelConfig mixed provider scenarios uses all fallback providers together 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "opencode/kimi-k2.5-free",
@@ -1301,7 +1301,7 @@ exports[`generateModelConfig mixed provider scenarios uses all fallback provider
   },
   "categories": {
     "artistry": {
-      "model": "github-copilot/gemini-3-pro-preview",
+      "model": "github-copilot/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "deep": {
@@ -1322,7 +1322,7 @@ exports[`generateModelConfig mixed provider scenarios uses all fallback provider
       "model": "github-copilot/claude-sonnet-4.5",
     },
     "visual-engineering": {
-      "model": "github-copilot/gemini-3-pro-preview",
+      "model": "github-copilot/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {
@@ -1334,7 +1334,7 @@ exports[`generateModelConfig mixed provider scenarios uses all fallback provider
 
 exports[`generateModelConfig mixed provider scenarios uses all providers together 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "opencode/kimi-k2.5-free",
@@ -1375,7 +1375,7 @@ exports[`generateModelConfig mixed provider scenarios uses all providers togethe
   },
   "categories": {
     "artistry": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "deep": {
@@ -1396,7 +1396,7 @@ exports[`generateModelConfig mixed provider scenarios uses all providers togethe
       "model": "anthropic/claude-sonnet-4-5",
     },
     "visual-engineering": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {
@@ -1408,7 +1408,7 @@ exports[`generateModelConfig mixed provider scenarios uses all providers togethe
 
 exports[`generateModelConfig mixed provider scenarios uses all providers with isMax20 flag 1`] = `
 {
-  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
+  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json",
   "agents": {
     "atlas": {
       "model": "opencode/kimi-k2.5-free",
@@ -1449,7 +1449,7 @@ exports[`generateModelConfig mixed provider scenarios uses all providers with is
   },
   "categories": {
     "artistry": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "deep": {
@@ -1471,7 +1471,7 @@ exports[`generateModelConfig mixed provider scenarios uses all providers with is
       "model": "anthropic/claude-sonnet-4-5",
     },
     "visual-engineering": {
-      "model": "google/gemini-3-pro-preview",
+      "model": "google/gemini-3.1-pro-preview",
       "variant": "high",
     },
     "writing": {

src/cli/config-manager.test.ts

@@ -178,7 +178,7 @@ describe("config-manager ANTIGRAVITY_PROVIDER_CONFIG", () => {
     expect(models).toBeTruthy()
 
     const required = [
-      "antigravity-gemini-3-pro",
+      "antigravity-gemini-3.1-pro",
       "antigravity-gemini-3-flash",
       "antigravity-claude-sonnet-4-6",
       "antigravity-claude-sonnet-4-6-thinking",
@@ -206,7 +206,7 @@ describe("config-manager ANTIGRAVITY_PROVIDER_CONFIG", () => {
     const models = (ANTIGRAVITY_PROVIDER_CONFIG as any).google.models as Record<string, any>
 
     // #when checking Gemini Pro variants
-    const pro = models["antigravity-gemini-3-pro"]
+    const pro = models["antigravity-gemini-3.1-pro"]
     // #then should have low and high variants
     expect(pro.variants).toBeTruthy()
     expect(pro.variants.low).toBeTruthy()
@@ -277,7 +277,7 @@ describe("generateOmoConfig - model fallback system", () => {
     const result = generateOmoConfig(config)
 
     // #then Sisyphus is omitted (requires all fallback providers)
-    expect(result.$schema).toBe("https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json")
+    expect(result.$schema).toBe("https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json")
     expect((result.agents as Record<string, { model: string }>).sisyphus).toBeUndefined()
   })
 

src/cli/config-manager/antigravity-provider-configuration.ts

@@ -4,10 +4,10 @@
  * IMPORTANT: Model names MUST use `antigravity-` prefix for stability.
  *
  * Since opencode-antigravity-auth v1.3.0, models use a variant system:
- * - `antigravity-gemini-3-pro` with variants: low, high
+ * - `antigravity-gemini-3.1-pro` with variants: low, high
  * - `antigravity-gemini-3-flash` with variants: minimal, low, medium, high
  *
- * Legacy tier-suffixed names (e.g., `antigravity-gemini-3-pro-high`) still work
+ * Legacy tier-suffixed names (e.g., `antigravity-gemini-3.1-pro-high`) still work
  * but variants are the recommended approach.
  *
  * @see https://github.com/NoeFabris/opencode-antigravity-auth#models
@@ -16,7 +16,7 @@ export const ANTIGRAVITY_PROVIDER_CONFIG = {
   google: {
     name: "Google",
     models: {
-      "antigravity-gemini-3-pro": {
+      "antigravity-gemini-3.1-pro": {
         name: "Gemini 3 Pro (Antigravity)",
         limit: { context: 1048576, output: 65535 },
         modalities: { input: ["text", "image", "pdf"], output: ["text"] },

src/cli/config-manager/bun-install.ts

@@ -1,4 +1,5 @@
 import { getConfigDir } from "./config-context"
+import { spawnWithWindowsHide } from "../../shared/spawn-with-windows-hide"
 
 const BUN_INSTALL_TIMEOUT_SECONDS = 60
 const BUN_INSTALL_TIMEOUT_MS = BUN_INSTALL_TIMEOUT_SECONDS * 1000
@@ -16,7 +17,7 @@ export async function runBunInstall(): Promise<boolean> {
 
 export async function runBunInstallWithDetails(): Promise<BunInstallResult> {
   try {
-    const proc = Bun.spawn(["bun", "install"], {
+    const proc = spawnWithWindowsHide(["bun", "install"], {
       cwd: getConfigDir(),
       stdout: "inherit",
       stderr: "inherit",

src/cli/config-manager/opencode-binary.ts

@@ -1,4 +1,5 @@
 import type { OpenCodeBinaryType } from "../../shared/opencode-config-dir-types"
+import { spawnWithWindowsHide } from "../../shared/spawn-with-windows-hide"
 import { initConfigContext } from "./config-context"
 
 const OPENCODE_BINARIES = ["opencode", "opencode-desktop"] as const
@@ -11,7 +12,7 @@ interface OpenCodeBinaryResult {
 async function findOpenCodeBinaryWithVersion(): Promise<OpenCodeBinaryResult | null> {
   for (const binary of OPENCODE_BINARIES) {
     try {
-      const proc = Bun.spawn([binary, "--version"], {
+      const proc = spawnWithWindowsHide([binary, "--version"], {
         stdout: "pipe",
         stderr: "pipe",
       })

src/cli/doctor/checks/dependencies.ts

@@ -3,6 +3,7 @@ import { createRequire } from "node:module"
 import { dirname, join } from "node:path"
 
 import type { DependencyInfo } from "../types"
+import { spawnWithWindowsHide } from "../../../shared/spawn-with-windows-hide"
 
 async function checkBinaryExists(binary: string): Promise<{ exists: boolean; path: string | null }> {
   try {
@@ -18,7 +19,7 @@ async function checkBinaryExists(binary: string): Promise<{ exists: boolean; pat
 
 async function getBinaryVersion(binary: string): Promise<string | null> {
   try {
-    const proc = Bun.spawn([binary, "--version"], { stdout: "pipe", stderr: "pipe" })
+    const proc = spawnWithWindowsHide([binary, "--version"], { stdout: "pipe", stderr: "pipe" })
     const output = await new Response(proc.stdout).text()
     await proc.exited
     if (proc.exitCode === 0) {
@@ -140,4 +141,3 @@ export async function checkCommentChecker(): Promise<DependencyInfo> {
     path: resolvedPath,
   }
 }
-

src/cli/doctor/checks/model-resolution.test.ts

@@ -26,7 +26,7 @@ describe("model-resolution check", () => {
       // then: Should have category entries
       const visual = info.categories.find((c) => c.name === "visual-engineering")
       expect(visual).toBeDefined()
-      expect(visual!.requirement.fallbackChain[0]?.model).toBe("gemini-3-pro")
+      expect(visual!.requirement.fallbackChain[0]?.model).toBe("gemini-3.1-pro")
       expect(visual!.requirement.fallbackChain[0]?.providers).toContain("google")
     })
   })

src/cli/doctor/checks/system-binary.ts

@@ -1,6 +1,7 @@
 import { existsSync } from "node:fs"
 import { homedir } from "node:os"
 import { join } from "node:path"
+import { spawnWithWindowsHide } from "../../../shared/spawn-with-windows-hide"
 
 import { OPENCODE_BINARIES } from "../constants"
 
@@ -110,7 +111,7 @@ export async function getOpenCodeVersion(
 ): Promise<string | null> {
   try {
     const command = buildVersionCommand(binaryPath, platform)
-    const processResult = Bun.spawn(command, { stdout: "pipe", stderr: "pipe" })
+    const processResult = spawnWithWindowsHide(command, { stdout: "pipe", stderr: "pipe" })
     const output = await new Response(processResult.stdout).text()
     await processResult.exited
 

src/cli/doctor/checks/tools-gh.ts

@@ -1,3 +1,5 @@
+import { spawnWithWindowsHide } from "../../../shared/spawn-with-windows-hide"
+
 export interface GhCliInfo {
   installed: boolean
   version: string | null
@@ -19,7 +21,7 @@ async function checkBinaryExists(binary: string): Promise<{ exists: boolean; pat
 
 async function getGhVersion(): Promise<string | null> {
   try {
-    const processResult = Bun.spawn(["gh", "--version"], { stdout: "pipe", stderr: "pipe" })
+    const processResult = spawnWithWindowsHide(["gh", "--version"], { stdout: "pipe", stderr: "pipe" })
     const output = await new Response(processResult.stdout).text()
     await processResult.exited
     if (processResult.exitCode !== 0) return null
@@ -38,7 +40,7 @@ async function getGhAuthStatus(): Promise<{
   error: string | null
 }> {
   try {
-    const processResult = Bun.spawn(["gh", "auth", "status"], {
+    const processResult = spawnWithWindowsHide(["gh", "auth", "status"], {
       stdout: "pipe",
       stderr: "pipe",
       env: { ...process.env, GH_NO_UPDATE_NOTIFIER: "1" },

src/cli/model-fallback-requirements.ts

@@ -24,7 +24,7 @@ export const CLI_AGENT_MODEL_REQUIREMENTS: Record<string, ModelRequirement> = {
   oracle: {
     fallbackChain: [
       { providers: ["openai", "github-copilot", "opencode"], model: "gpt-5.2", variant: "high" },
-      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3-pro", variant: "high" },
+      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3.1-pro", variant: "high" },
       { providers: ["anthropic", "github-copilot", "opencode"], model: "claude-opus-4-6", variant: "max" },
     ],
   },
@@ -59,7 +59,7 @@ export const CLI_AGENT_MODEL_REQUIREMENTS: Record<string, ModelRequirement> = {
       { providers: ["kimi-for-coding"], model: "k2p5" },
       { providers: ["opencode"], model: "kimi-k2.5-free" },
       { providers: ["openai", "github-copilot", "opencode"], model: "gpt-5.2", variant: "high" },
-      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3-pro" },
+      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3.1-pro" },
     ],
   },
   metis: {
@@ -68,14 +68,14 @@ export const CLI_AGENT_MODEL_REQUIREMENTS: Record<string, ModelRequirement> = {
       { providers: ["kimi-for-coding"], model: "k2p5" },
       { providers: ["opencode"], model: "kimi-k2.5-free" },
       { providers: ["openai", "github-copilot", "opencode"], model: "gpt-5.2", variant: "high" },
-      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3-pro", variant: "high" },
+      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3.1-pro", variant: "high" },
     ],
   },
   momus: {
     fallbackChain: [
       { providers: ["openai", "github-copilot", "opencode"], model: "gpt-5.2", variant: "medium" },
       { providers: ["anthropic", "github-copilot", "opencode"], model: "claude-opus-4-6", variant: "max" },
-      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3-pro", variant: "high" },
+      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3.1-pro", variant: "high" },
     ],
   },
   atlas: {
@@ -84,7 +84,7 @@ export const CLI_AGENT_MODEL_REQUIREMENTS: Record<string, ModelRequirement> = {
       { providers: ["opencode"], model: "kimi-k2.5-free" },
       { providers: ["anthropic", "github-copilot", "opencode"], model: "claude-sonnet-4-5" },
       { providers: ["openai", "github-copilot", "opencode"], model: "gpt-5.2" },
-      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3-pro" },
+      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3.1-pro" },
     ],
   },
 }
@@ -92,7 +92,7 @@ export const CLI_AGENT_MODEL_REQUIREMENTS: Record<string, ModelRequirement> = {
 export const CLI_CATEGORY_MODEL_REQUIREMENTS: Record<string, ModelRequirement> = {
   "visual-engineering": {
     fallbackChain: [
-      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3-pro", variant: "high" },
+      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3.1-pro", variant: "high" },
       { providers: ["zai-coding-plan"], model: "glm-5" },
       { providers: ["anthropic", "github-copilot", "opencode"], model: "claude-opus-4-6", variant: "max" },
       { providers: ["kimi-for-coding"], model: "k2p5" },
@@ -101,7 +101,7 @@ export const CLI_CATEGORY_MODEL_REQUIREMENTS: Record<string, ModelRequirement> =
   ultrabrain: {
     fallbackChain: [
       { providers: ["openai", "opencode"], model: "gpt-5.3-codex", variant: "xhigh" },
-      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3-pro", variant: "high" },
+      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3.1-pro", variant: "high" },
       { providers: ["anthropic", "github-copilot", "opencode"], model: "claude-opus-4-6", variant: "max" },
     ],
   },
@@ -109,17 +109,17 @@ export const CLI_CATEGORY_MODEL_REQUIREMENTS: Record<string, ModelRequirement> =
     fallbackChain: [
       { providers: ["openai", "opencode"], model: "gpt-5.3-codex", variant: "medium" },
       { providers: ["anthropic", "github-copilot", "opencode"], model: "claude-opus-4-6", variant: "max" },
-      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3-pro", variant: "high" },
+      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3.1-pro", variant: "high" },
     ],
     requiresModel: "gpt-5.3-codex",
   },
   artistry: {
     fallbackChain: [
-      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3-pro", variant: "high" },
+      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3.1-pro", variant: "high" },
       { providers: ["anthropic", "github-copilot", "opencode"], model: "claude-opus-4-6", variant: "max" },
       { providers: ["openai", "github-copilot", "opencode"], model: "gpt-5.2" },
     ],
-    requiresModel: "gemini-3-pro",
+    requiresModel: "gemini-3.1-pro",
   },
   quick: {
     fallbackChain: [
@@ -139,7 +139,7 @@ export const CLI_CATEGORY_MODEL_REQUIREMENTS: Record<string, ModelRequirement> =
     fallbackChain: [
       { providers: ["anthropic", "github-copilot", "opencode"], model: "claude-opus-4-6", variant: "max" },
       { providers: ["openai", "github-copilot", "opencode"], model: "gpt-5.2", variant: "high" },
-      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3-pro" },
+      { providers: ["google", "github-copilot", "opencode"], model: "gemini-3.1-pro" },
     ],
   },
   writing: {

src/cli/model-fallback.test.ts

@@ -515,7 +515,7 @@ describe("generateModelConfig", () => {
 
       // #then should include correct schema URL
       expect(result.$schema).toBe(
-        "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json"
+        "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json"
       )
     })
   })

src/cli/model-fallback.ts

@@ -19,7 +19,7 @@ export type { GeneratedOmoConfig } from "./model-fallback-types"
 const ZAI_MODEL = "zai-coding-plan/glm-4.7"
 
 const ULTIMATE_FALLBACK = "opencode/glm-4.7-free"
-const SCHEMA_URL = "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json"
+const SCHEMA_URL = "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/dev/assets/oh-my-opencode.schema.json"
 
 
 

src/cli/provider-model-id-transform.test.ts

@@ -40,16 +40,16 @@ describe("transformModelForProvider", () => {
 			expect(result).toBe("claude-haiku-4.5")
 		})
 
-		test("transforms gemini-3-pro to gemini-3-pro-preview", () => {
-			// #given github-copilot provider and gemini-3-pro model
+		test("transforms gemini-3.1-pro to gemini-3.1-pro-preview", () => {
+			// #given github-copilot provider and gemini-3.1-pro model
 			const provider = "github-copilot"
-			const model = "gemini-3-pro"
+			const model = "gemini-3.1-pro"
 
 			// #when transformModelForProvider is called
 			const result = transformModelForProvider(provider, model)
 
-			// #then should transform to gemini-3-pro-preview
-			expect(result).toBe("gemini-3-pro-preview")
+			// #then should transform to gemini-3.1-pro-preview
+			expect(result).toBe("gemini-3.1-pro-preview")
 		})
 
 		test("transforms gemini-3-flash to gemini-3-flash-preview", () => {
@@ -64,16 +64,16 @@ describe("transformModelForProvider", () => {
 			expect(result).toBe("gemini-3-flash-preview")
 		})
 
-		test("prevents double transformation of gemini-3-pro-preview", () => {
-			// #given github-copilot provider and gemini-3-pro-preview model (already transformed)
+		test("prevents double transformation of gemini-3.1-pro-preview", () => {
+			// #given github-copilot provider and gemini-3.1-pro-preview model (already transformed)
 			const provider = "github-copilot"
-			const model = "gemini-3-pro-preview"
+			const model = "gemini-3.1-pro-preview"
 
 			// #when transformModelForProvider is called
 			const result = transformModelForProvider(provider, model)
 
-			// #then should NOT become gemini-3-pro-preview-preview
-			expect(result).toBe("gemini-3-pro-preview")
+			// #then should NOT become gemini-3.1-pro-preview-preview
+			expect(result).toBe("gemini-3.1-pro-preview")
 		})
 
 		test("prevents double transformation of gemini-3-flash-preview", () => {
@@ -102,16 +102,16 @@ describe("transformModelForProvider", () => {
 			expect(result).toBe("gemini-3-flash-preview")
 		})
 
-		test("transforms gemini-3-pro to gemini-3-pro-preview", () => {
-			// #given google provider and gemini-3-pro model
+		test("transforms gemini-3.1-pro to gemini-3.1-pro-preview", () => {
+			// #given google provider and gemini-3.1-pro model
 			const provider = "google"
-			const model = "gemini-3-pro"
+			const model = "gemini-3.1-pro"
 
 			// #when transformModelForProvider is called
 			const result = transformModelForProvider(provider, model)
 
-			// #then should transform to gemini-3-pro-preview
-			expect(result).toBe("gemini-3-pro-preview")
+			// #then should transform to gemini-3.1-pro-preview
+			expect(result).toBe("gemini-3.1-pro-preview")
 		})
 
 		test("passes through other gemini models unchanged", () => {
@@ -138,16 +138,16 @@ describe("transformModelForProvider", () => {
 			expect(result).toBe("gemini-3-flash-preview")
 		})
 
-		test("prevents double transformation of gemini-3-pro-preview", () => {
-			// #given google provider and gemini-3-pro-preview model (already transformed)
+		test("prevents double transformation of gemini-3.1-pro-preview", () => {
+			// #given google provider and gemini-3.1-pro-preview model (already transformed)
 			const provider = "google"
-			const model = "gemini-3-pro-preview"
+			const model = "gemini-3.1-pro-preview"
 
 			// #when transformModelForProvider is called
 			const result = transformModelForProvider(provider, model)
 
-			// #then should NOT become gemini-3-pro-preview-preview
-			expect(result).toBe("gemini-3-pro-preview")
+			// #then should NOT become gemini-3.1-pro-preview-preview
+			expect(result).toBe("gemini-3.1-pro-preview")
 		})
 
 		test("does not transform claude models for google provider", () => {

src/cli/run/integration.test.ts

@@ -3,6 +3,7 @@ import type { RunResult } from "./types"
 import { createJsonOutputManager } from "./json-output"
 import { resolveSession } from "./session-resolver"
 import { executeOnCompleteHook } from "./on-complete-hook"
+import * as spawnWithWindowsHideModule from "../../shared/spawn-with-windows-hide"
 import type { OpencodeClient } from "./types"
 import * as originalSdk from "@opencode-ai/sdk"
 import * as originalPortUtils from "../../shared/port-utils"
@@ -147,7 +148,7 @@ describe("integration: --session-id", () => {
     const result = resolveSession({ client: mockClient, sessionId, directory: "/test" })
 
     // then
-    await expect(result).rejects.toThrow(`Session not found: ${sessionId}`)
+    expect(result).rejects.toThrow(`Session not found: ${sessionId}`)
     expect(mockClient.session.get).toHaveBeenCalledWith({
       path: { id: sessionId },
       query: { directory: "/test" },
@@ -161,10 +162,13 @@ describe("integration: --on-complete", () => {
 
   beforeEach(() => {
     spyOn(console, "error").mockImplementation(() => {})
-    spawnSpy = spyOn(Bun, "spawn").mockReturnValue({
+    spawnSpy = spyOn(spawnWithWindowsHideModule, "spawnWithWindowsHide").mockReturnValue({
       exited: Promise.resolve(0),
       exitCode: 0,
-    } as unknown as ReturnType<typeof Bun.spawn>)
+      stdout: undefined,
+      stderr: undefined,
+      kill: () => {},
+    } satisfies ReturnType<typeof spawnWithWindowsHideModule.spawnWithWindowsHide>)
   })
 
   afterEach(() => {
@@ -186,7 +190,7 @@ describe("integration: --on-complete", () => {
 
     // then
     expect(spawnSpy).toHaveBeenCalledTimes(1)
-    const [_, options] = spawnSpy.mock.calls[0] as Parameters<typeof Bun.spawn>
+    const [_, options] = spawnSpy.mock.calls[0] as Parameters<typeof spawnWithWindowsHideModule.spawnWithWindowsHide>
     expect(options?.env?.SESSION_ID).toBe("session-123")
     expect(options?.env?.EXIT_CODE).toBe("0")
     expect(options?.env?.DURATION_MS).toBe("5000")
@@ -208,10 +212,13 @@ describe("integration: option combinations", () => {
     spyOn(console, "error").mockImplementation(() => {})
     mockStdout = createMockWriteStream()
     mockStderr = createMockWriteStream()
-    spawnSpy = spyOn(Bun, "spawn").mockReturnValue({
+    spawnSpy = spyOn(spawnWithWindowsHideModule, "spawnWithWindowsHide").mockReturnValue({
       exited: Promise.resolve(0),
       exitCode: 0,
-    } as unknown as ReturnType<typeof Bun.spawn>)
+      stdout: undefined,
+      stderr: undefined,
+      kill: () => {},
+    } satisfies ReturnType<typeof spawnWithWindowsHideModule.spawnWithWindowsHide>)
   })
 
   afterEach(() => {
@@ -249,9 +256,9 @@ describe("integration: option combinations", () => {
     const emitted = mockStdout.writes[0]!
     expect(() => JSON.parse(emitted)).not.toThrow()
     expect(spawnSpy).toHaveBeenCalledTimes(1)
-    const [args] = spawnSpy.mock.calls[0] as Parameters<typeof Bun.spawn>
+    const [args] = spawnSpy.mock.calls[0] as Parameters<typeof spawnWithWindowsHideModule.spawnWithWindowsHide>
     expect(args).toEqual(["sh", "-c", "echo done"])
-    const [_, options] = spawnSpy.mock.calls[0] as Parameters<typeof Bun.spawn>
+    const [_, options] = spawnSpy.mock.calls[0] as Parameters<typeof spawnWithWindowsHideModule.spawnWithWindowsHide>
     expect(options?.env?.SESSION_ID).toBe("session-123")
     expect(options?.env?.EXIT_CODE).toBe("0")
     expect(options?.env?.DURATION_MS).toBe("5000")

src/cli/run/on-complete-hook.test.ts

@@ -1,4 +1,5 @@
 import { describe, it, expect, spyOn, beforeEach, afterEach } from "bun:test"
+import * as spawnWithWindowsHideModule from "../../shared/spawn-with-windows-hide"
 import { executeOnCompleteHook } from "./on-complete-hook"
 
 describe("executeOnCompleteHook", () => {
@@ -6,7 +7,10 @@ describe("executeOnCompleteHook", () => {
     return {
       exited: Promise.resolve(exitCode),
       exitCode,
-    } as unknown as ReturnType<typeof Bun.spawn>
+      stdout: undefined,
+      stderr: undefined,
+      kill: () => {},
+    } satisfies ReturnType<typeof spawnWithWindowsHideModule.spawnWithWindowsHide>
   }
 
   let consoleErrorSpy: ReturnType<typeof spyOn<typeof console, "error">>
@@ -21,7 +25,7 @@ describe("executeOnCompleteHook", () => {
 
   it("executes command with correct env vars", async () => {
     // given
-    const spawnSpy = spyOn(Bun, "spawn").mockReturnValue(createProc(0))
+    const spawnSpy = spyOn(spawnWithWindowsHideModule, "spawnWithWindowsHide").mockReturnValue(createProc(0))
 
     try {
       // when
@@ -35,7 +39,7 @@ describe("executeOnCompleteHook", () => {
 
       // then
       expect(spawnSpy).toHaveBeenCalledTimes(1)
-      const [args, options] = spawnSpy.mock.calls[0] as Parameters<typeof Bun.spawn>
+      const [args, options] = spawnSpy.mock.calls[0] as Parameters<typeof spawnWithWindowsHideModule.spawnWithWindowsHide>
 
       expect(args).toEqual(["sh", "-c", "echo test"])
       expect(options?.env?.SESSION_ID).toBe("session-123")
@@ -51,7 +55,7 @@ describe("executeOnCompleteHook", () => {
 
   it("env var values are strings", async () => {
     // given
-    const spawnSpy = spyOn(Bun, "spawn").mockReturnValue(createProc(0))
+    const spawnSpy = spyOn(spawnWithWindowsHideModule, "spawnWithWindowsHide").mockReturnValue(createProc(0))
 
     try {
       // when
@@ -64,7 +68,7 @@ describe("executeOnCompleteHook", () => {
       })
 
       // then
-      const [_, options] = spawnSpy.mock.calls[0] as Parameters<typeof Bun.spawn>
+      const [_, options] = spawnSpy.mock.calls[0] as Parameters<typeof spawnWithWindowsHideModule.spawnWithWindowsHide>
 
       expect(options?.env?.EXIT_CODE).toBe("1")
       expect(options?.env?.EXIT_CODE).toBeTypeOf("string")
@@ -79,7 +83,7 @@ describe("executeOnCompleteHook", () => {
 
   it("empty command string is no-op", async () => {
     // given
-    const spawnSpy = spyOn(Bun, "spawn").mockReturnValue(createProc(0))
+    const spawnSpy = spyOn(spawnWithWindowsHideModule, "spawnWithWindowsHide").mockReturnValue(createProc(0))
 
     try {
       // when
@@ -100,7 +104,7 @@ describe("executeOnCompleteHook", () => {
 
   it("whitespace-only command is no-op", async () => {
     // given
-    const spawnSpy = spyOn(Bun, "spawn").mockReturnValue(createProc(0))
+    const spawnSpy = spyOn(spawnWithWindowsHideModule, "spawnWithWindowsHide").mockReturnValue(createProc(0))
 
     try {
       // when
@@ -121,11 +125,11 @@ describe("executeOnCompleteHook", () => {
 
   it("command failure logs warning but does not throw", async () => {
     // given
-    const spawnSpy = spyOn(Bun, "spawn").mockReturnValue(createProc(1))
+    const spawnSpy = spyOn(spawnWithWindowsHideModule, "spawnWithWindowsHide").mockReturnValue(createProc(1))
 
     try {
       // when
-      await expect(
+      expect(
         executeOnCompleteHook({
           command: "false",
           sessionId: "session-123",
@@ -149,13 +153,13 @@ describe("executeOnCompleteHook", () => {
   it("spawn error logs warning but does not throw", async () => {
     // given
     const spawnError = new Error("Command not found")
-    const spawnSpy = spyOn(Bun, "spawn").mockImplementation(() => {
+    const spawnSpy = spyOn(spawnWithWindowsHideModule, "spawnWithWindowsHide").mockImplementation(() => {
       throw spawnError
     })
 
     try {
       // when
-      await expect(
+      expect(
         executeOnCompleteHook({
           command: "nonexistent-command",
           sessionId: "session-123",

src/cli/run/on-complete-hook.ts

@@ -1,4 +1,5 @@
 import pc from "picocolors"
+import { spawnWithWindowsHide } from "../../shared/spawn-with-windows-hide"
 
 export async function executeOnCompleteHook(options: {
   command: string
@@ -17,7 +18,7 @@ export async function executeOnCompleteHook(options: {
   console.error(pc.dim(`Running on-complete hook: ${trimmedCommand}`))
 
   try {
-    const proc = Bun.spawn(["sh", "-c", trimmedCommand], {
+    const proc = spawnWithWindowsHide(["sh", "-c", trimmedCommand], {
       env: {
         ...process.env,
         SESSION_ID: sessionId,

src/cli/run/opencode-binary-resolver.ts

@@ -1,4 +1,5 @@
 import { delimiter, dirname, join } from "node:path"
+import { spawnWithWindowsHide } from "../../shared/spawn-with-windows-hide"
 
 const OPENCODE_COMMANDS = ["opencode", "opencode-desktop"] as const
 const WINDOWS_SUFFIXES = ["", ".exe", ".cmd", ".bat", ".ps1"] as const
@@ -41,7 +42,7 @@ export function collectCandidateBinaryPaths(
 
 export async function canExecuteBinary(binaryPath: string): Promise<boolean> {
   try {
-    const proc = Bun.spawn([binaryPath, "--version"], {
+    const proc = spawnWithWindowsHide([binaryPath, "--version"], {
       stdout: "pipe",
       stderr: "pipe",
     })

src/config/schema/hooks.ts

@@ -49,6 +49,7 @@ export const HookNameSchema = z.enum([
   "write-existing-file-guard",
   "anthropic-effort",
   "hashline-read-enhancer",
+  "read-image-resizer",
 ])
 
 export type HookName = z.infer<typeof HookNameSchema>

src/config/schema/oh-my-opencode-config.ts

@@ -27,13 +27,13 @@ export const OhMyOpenCodeConfigSchema = z.object({
   /** Default agent name for `oh-my-opencode run` (env: OPENCODE_DEFAULT_AGENT) */
   default_run_agent: z.string().optional(),
   disabled_mcps: z.array(AnyMcpNameSchema).optional(),
-  disabled_agents: z.array(BuiltinAgentNameSchema).optional(),
+  disabled_agents: z.array(z.string()).optional(),
   disabled_skills: z.array(BuiltinSkillNameSchema).optional(),
   disabled_hooks: z.array(z.string()).optional(),
   disabled_commands: z.array(BuiltinCommandNameSchema).optional(),
   /** Disable specific tools by name (e.g., ["todowrite", "todoread"]) */
   disabled_tools: z.array(z.string()).optional(),
-  /** Enable hashline_edit tool/hook integrations (default: true at call site) */
+  /** Enable hashline_edit tool/hook integrations (default: false) */
   hashline_edit: z.boolean().optional(),
   /** Enable model fallback on API errors (default: false). Set to true to enable automatic model switching when model errors occur. */
   model_fallback: z.boolean().optional(),

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

@@ -34,7 +34,7 @@ describe("ConcurrencyManager.getConcurrencyLimit", () => {
   test("should return provider limit even when modelConcurrency exists but doesn't match", () => {
     // given
     const config: BackgroundTaskConfig = {
-      modelConcurrency: { "google/gemini-3-pro": 5 },
+      modelConcurrency: { "google/gemini-3.1-pro": 5 },
       providerConcurrency: { anthropic: 3 }
     }
     const manager = new ConcurrencyManager(config)
@@ -95,7 +95,7 @@ describe("ConcurrencyManager.getConcurrencyLimit", () => {
     // when
     const modelLimit = manager.getConcurrencyLimit("anthropic/claude-sonnet-4-6")
     const providerLimit = manager.getConcurrencyLimit("anthropic/claude-opus-4-6")
-    const defaultLimit = manager.getConcurrencyLimit("google/gemini-3-pro")
+    const defaultLimit = manager.getConcurrencyLimit("google/gemini-3.1-pro")
 
     // then
     expect(modelLimit).toBe(10)

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

@@ -2987,6 +2987,28 @@ describe("BackgroundManager.handleEvent - session.deleted cascade", () => {
     manager.shutdown()
     resetToastManager()
   })
+
+  test("should clean pending notifications for deleted sessions", () => {
+    //#given
+    const manager = createBackgroundManager()
+    const sessionID = "session-pending-notifications"
+
+    manager.queuePendingNotification(sessionID, "<system-reminder>queued</system-reminder>")
+    expect(getPendingNotifications(manager).get(sessionID)).toEqual([
+      "<system-reminder>queued</system-reminder>",
+    ])
+
+    //#when
+    manager.handleEvent({
+      type: "session.deleted",
+      properties: { info: { id: sessionID } },
+    })
+
+    //#then
+    expect(getPendingNotifications(manager).has(sessionID)).toBe(false)
+
+    manager.shutdown()
+  })
 })
 
 describe("BackgroundManager.handleEvent - session.error", () => {

src/features/background-agent/manager.ts

@@ -830,6 +830,8 @@ export class BackgroundManager {
         tasksToCancel.set(descendant.id, descendant)
       }
 
+      this.pendingNotifications.delete(sessionID)
+
       if (tasksToCancel.size === 0) return
 
       for (const task of tasksToCancel.values()) {
@@ -866,6 +868,13 @@ export class BackgroundManager {
           subagentSessions.delete(task.sessionID)
         }
       }
+
+      for (const task of tasksToCancel.values()) {
+        if (task.parentSessionID) {
+          this.pendingNotifications.delete(task.parentSessionID)
+        }
+      }
+
       SessionCategoryRegistry.remove(sessionID)
     }
 

src/features/claude-code-plugin-loader/types.ts

@@ -80,12 +80,11 @@ export interface PluginManifest {
 /**
  * Hooks configuration
  */
-export interface HookEntry {
-  type: "command" | "prompt" | "agent"
-  command?: string
-  prompt?: string
-  agent?: string
-}
+export type HookEntry =
+  | { type: "command"; command?: string }
+  | { type: "prompt"; prompt?: string }
+  | { type: "agent"; agent?: string }
+  | { type: "http"; url: string; headers?: Record<string, string>; allowedEnvVars?: string[]; timeout?: number }
 
 export interface HookMatcher {
   matcher?: string

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

@@ -205,6 +205,45 @@ describe("ContextCollector", () => {
       const ids = pending.entries.map((e) => e.id)
       expect(ids).toEqual(["first", "second", "third"])
     })
+
+    it("keeps registration order even when Date.now values are not monotonic", () => {
+      // given
+      const sessionID = "ses_order_non_monotonic_time"
+      const originalDateNow = Date.now
+      const mockedTimestamps = [300, 100, 200]
+      let timestampIndex = 0
+      Date.now = () => mockedTimestamps[timestampIndex++] ?? 0
+
+      try {
+        collector.register(sessionID, {
+          id: "first",
+          source: "custom",
+          content: "First",
+          priority: "normal",
+        })
+        collector.register(sessionID, {
+          id: "second",
+          source: "custom",
+          content: "Second",
+          priority: "normal",
+        })
+        collector.register(sessionID, {
+          id: "third",
+          source: "custom",
+          content: "Third",
+          priority: "normal",
+        })
+      } finally {
+        Date.now = originalDateNow
+      }
+
+      // when
+      const pending = collector.getPending(sessionID)
+
+      // then
+      const ids = pending.entries.map((entry) => entry.id)
+      expect(ids).toEqual(["first", "second", "third"])
+    })
   })
 
   describe("consume", () => {

src/features/context-injector/collector.ts

@@ -14,6 +14,8 @@ const PRIORITY_ORDER: Record<ContextPriority, number> = {
 
 const CONTEXT_SEPARATOR = "\n\n---\n\n"
 
+let registrationCounter = 0
+
 export class ContextCollector {
   private sessions: Map<string, Map<string, ContextEntry>> = new Map()
 
@@ -30,7 +32,7 @@ export class ContextCollector {
       source: options.source,
       content: options.content,
       priority: options.priority ?? "normal",
-      timestamp: Date.now(),
+      registrationOrder: ++registrationCounter,
       metadata: options.metadata,
     }
 
@@ -77,7 +79,7 @@ export class ContextCollector {
     return entries.sort((a, b) => {
       const priorityDiff = PRIORITY_ORDER[a.priority] - PRIORITY_ORDER[b.priority]
       if (priorityDiff !== 0) return priorityDiff
-      return a.timestamp - b.timestamp
+      return a.registrationOrder - b.registrationOrder
     })
   }
 }

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

@@ -64,6 +64,51 @@ describe("createContextInjectorMessagesTransformHook", () => {
     expect(output.messages[2].parts[1].text).toBe("Second message")
   })
 
+  it("uses deterministic synthetic part ID across repeated transforms", async () => {
+    // given
+    const hook = createContextInjectorMessagesTransformHook(collector)
+    const sessionID = "ses_transform_deterministic"
+    const baseMessage = createMockMessage("user", "Stable message", sessionID)
+
+    collector.register(sessionID, {
+      id: "ctx-1",
+      source: "keyword-detector",
+      content: "Injected context",
+    })
+    const firstOutput = {
+      messages: [structuredClone(baseMessage)],
+    }
+
+    // when
+    await hook["experimental.chat.messages.transform"]!({}, firstOutput)
+
+    // then
+    const firstSyntheticPart = firstOutput.messages[0].parts[0]
+    expect(
+      "synthetic" in firstSyntheticPart && firstSyntheticPart.synthetic === true
+    ).toBe(true)
+
+    // given
+    collector.register(sessionID, {
+      id: "ctx-2",
+      source: "keyword-detector",
+      content: "Injected context",
+    })
+    const secondOutput = {
+      messages: [structuredClone(baseMessage)],
+    }
+
+    // when
+    await hook["experimental.chat.messages.transform"]!({}, secondOutput)
+
+    // then
+    const secondSyntheticPart = secondOutput.messages[0].parts[0]
+    expect(
+      "synthetic" in secondSyntheticPart && secondSyntheticPart.synthetic === true
+    ).toBe(true)
+    expect(secondSyntheticPart.id).toBe(firstSyntheticPart.id)
+  })
+
   it("does nothing when no pending context", async () => {
     // given
     const hook = createContextInjectorMessagesTransformHook(collector)

src/features/context-injector/injector.ts

@@ -148,7 +148,7 @@ export function createContextInjectorMessagesTransformHook(
 
       // synthetic part pattern (minimal fields)
       const syntheticPart = {
-        id: `synthetic_hook_${Date.now()}`,
+        id: `synthetic_hook_${sessionID}`,
         messageID: lastUserMessage.info.id,
         sessionID: (lastUserMessage.info as { sessionID?: string }).sessionID ?? "",
         type: "text" as const,

src/features/context-injector/types.ts

@@ -27,8 +27,8 @@ export interface ContextEntry {
   content: string
   /** Priority for ordering (default: normal) */
   priority: ContextPriority
-  /** Timestamp when registered */
-  timestamp: number
+  /** Monotonic order when registered */
+  registrationOrder: number
   /** Optional metadata for debugging/logging */
   metadata?: Record<string, unknown>
 }

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

@@ -4,6 +4,8 @@ import {
   findFirstMessageWithAgent,
   findNearestMessageWithFieldsFromSDK,
   findFirstMessageWithAgentFromSDK,
+  generateMessageId,
+  generatePartId,
   injectHookMessage,
 } from "./injector"
 import { isSqliteBackend, resetSqliteBackendCache } from "../../shared/opencode-storage-detection"
@@ -192,6 +194,40 @@ describe("findFirstMessageWithAgentFromSDK", () => {
   })
 })
 
+describe("generateMessageId", () => {
+  it("returns deterministic sequential IDs with fixed format", () => {
+    // given
+    const format = /^msg_[0-9a-f]{8}_\d{6}$/
+
+    // when
+    const firstId = generateMessageId()
+    const secondId = generateMessageId()
+
+    // then
+    expect(firstId).toMatch(format)
+    expect(secondId).toMatch(format)
+    expect(secondId.split("_")[1]).toBe(firstId.split("_")[1])
+    expect(Number(secondId.split("_")[2])).toBe(Number(firstId.split("_")[2]) + 1)
+  })
+})
+
+describe("generatePartId", () => {
+  it("returns deterministic sequential IDs with fixed format", () => {
+    // given
+    const format = /^prt_[0-9a-f]{8}_\d{6}$/
+
+    // when
+    const firstId = generatePartId()
+    const secondId = generatePartId()
+
+    // then
+    expect(firstId).toMatch(format)
+    expect(secondId).toMatch(format)
+    expect(secondId.split("_")[1]).toBe(firstId.split("_")[1])
+    expect(Number(secondId.split("_")[2])).toBe(Number(firstId.split("_")[2]) + 1)
+  })
+})
+
 describe("injectHookMessage", () => {
   beforeEach(() => {
     vi.clearAllMocks()

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

@@ -1,4 +1,5 @@
 import { existsSync, mkdirSync, readFileSync, readdirSync, writeFileSync } from "node:fs"
+import { randomBytes } from "node:crypto"
 import { join } from "node:path"
 import type { PluginInput } from "@opencode-ai/plugin"
 import { MESSAGE_STORAGE, PART_STORAGE } from "./constants"
@@ -29,6 +30,10 @@ interface SDKMessage {
   }
 }
 
+const processPrefix = randomBytes(4).toString("hex")
+let messageCounter = 0
+let partCounter = 0
+
 function convertSDKMessageToStoredMessage(msg: SDKMessage): StoredMessage | null {
   const info = msg.info
   if (!info) return null
@@ -204,16 +209,12 @@ export function findFirstMessageWithAgent(messageDir: string): string | null {
   return null
 }
 
-function generateMessageId(): string {
-  const timestamp = Date.now().toString(16)
-  const random = Math.random().toString(36).substring(2, 14)
-  return `msg_${timestamp}${random}`
+export function generateMessageId(): string {
+  return `msg_${processPrefix}_${String(++messageCounter).padStart(6, "0")}`
 }
 
-function generatePartId(): string {
-  const timestamp = Date.now().toString(16)
-  const random = Math.random().toString(36).substring(2, 10)
-  return `prt_${timestamp}${random}`
+export function generatePartId(): string {
+  return `prt_${processPrefix}_${String(++partCounter).padStart(6, "0")}`
 }
 
 function getOrCreateMessageDir(sessionID: string): string {

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

@@ -162,7 +162,7 @@ describe("TaskToastManager", () => {
         description: "Task with category default model",
         agent: "sisyphus-junior",
         isBackground: false,
-        modelInfo: { model: "google/gemini-3-pro", type: "category-default" as const },
+        modelInfo: { model: "google/gemini-3.1-pro", type: "category-default" as const },
       }
 
       // when - addTask is called

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

@@ -6,6 +6,8 @@ import {
 } from "./storage";
 import { TARGET_TOOLS, AGENT_TOOLS, REMINDER_MESSAGE } from "./constants";
 import type { AgentUsageState } from "./types";
+import { getSessionAgent } from "../../features/claude-code-session-state";
+import { getAgentConfigKey } from "../../shared/agent-display-names";
 
 interface ToolExecuteInput {
   tool: string;
@@ -26,6 +28,23 @@ interface EventInput {
   };
 }
 
+/**
+ * Only orchestrator agents should receive usage reminders.
+ * Subagents (explore, librarian, oracle, etc.) are the targets of delegation,
+ * so reminding them to delegate to themselves is counterproductive.
+ */
+const ORCHESTRATOR_AGENTS = new Set([
+  "sisyphus",
+  "sisyphus-junior",
+  "atlas",
+  "hephaestus",
+  "prometheus",
+]);
+
+function isOrchestratorAgent(agentName: string): boolean {
+  return ORCHESTRATOR_AGENTS.has(getAgentConfigKey(agentName));
+}
+
 export function createAgentUsageReminderHook(_ctx: PluginInput) {
   const sessionStates = new Map<string, AgentUsageState>();
 
@@ -60,6 +79,12 @@ export function createAgentUsageReminderHook(_ctx: PluginInput) {
     output: ToolExecuteOutput,
   ) => {
     const { tool, sessionID } = input;
+
+    const agent = getSessionAgent(sessionID);
+    if (agent && !isOrchestratorAgent(agent)) {
+      return;
+    }
+
     const toolLower = tool.toLowerCase();
 
     if (AGENT_TOOLS.has(toolLower)) {

src/hooks/atlas/event-handler.ts

@@ -1,6 +1,6 @@
 import type { PluginInput } from "@opencode-ai/plugin"
 import { getPlanProgress, readBoulderState } from "../../features/boulder-state"
-import { subagentSessions } from "../../features/claude-code-session-state"
+import { getSessionAgent, subagentSessions } from "../../features/claude-code-session-state"
 import { log } from "../../shared/logger"
 import { getAgentConfigKey } from "../../shared/agent-display-names"
 import { HOOK_NAME } from "./hook-name"
@@ -97,8 +97,10 @@ export function createAtlasEventHandler(input: {
         return
       }
 
+      const sessionAgent = getSessionAgent(sessionID)
       const lastAgent = await getLastAgentFromSession(sessionID, ctx.client)
-      const lastAgentKey = getAgentConfigKey(lastAgent ?? "")
+      const effectiveAgent = sessionAgent ?? lastAgent
+      const lastAgentKey = getAgentConfigKey(effectiveAgent ?? "")
       const requiredAgent = getAgentConfigKey(boulderState.agent ?? "atlas")
       const lastAgentMatchesRequired = lastAgentKey === requiredAgent
       const boulderAgentDefaultsToAtlas = requiredAgent === "atlas"
@@ -108,7 +110,7 @@ export function createAtlasEventHandler(input: {
       if (!agentMatches) {
         log(`[${HOOK_NAME}] Skipped: last agent does not match boulder agent`, {
           sessionID,
-          lastAgent: lastAgent ?? "unknown",
+          lastAgent: effectiveAgent ?? "unknown",
           requiredAgent,
         })
         return

src/hooks/atlas/index.test.ts

@@ -9,7 +9,7 @@ import {
   readBoulderState,
 } from "../../features/boulder-state"
 import type { BoulderState } from "../../features/boulder-state"
-import { _resetForTesting, subagentSessions } from "../../features/claude-code-session-state"
+import { _resetForTesting, subagentSessions, updateSessionAgent } from "../../features/claude-code-session-state"
 
 const TEST_STORAGE_ROOT = join(tmpdir(), `atlas-message-storage-${randomUUID()}`)
 const TEST_MESSAGE_STORAGE = join(TEST_STORAGE_ROOT, "message")
@@ -933,7 +933,7 @@ describe("atlas hook", () => {
       expect(callArgs.body.parts[0].text).toContain("2 remaining")
     })
 
-     test("should inject when last agent is sisyphus and boulder targets atlas explicitly", async () => {
+    test("should inject when last agent is sisyphus and boulder targets atlas explicitly", async () => {
        // given - boulder explicitly set to atlas, but last agent is sisyphus (initial state after /start-work)
        const planPath = join(TEST_DIR, "test-plan.md")
        writeFileSync(planPath, "# Plan\n- [ ] Task 1\n- [ ] Task 2")
@@ -1386,5 +1386,38 @@ describe("atlas hook", () => {
       // then - should call prompt because session state was cleaned
       expect(mockInput._promptMock).toHaveBeenCalled()
     })
+
+    test("should inject when session agent was updated to atlas by start-work even if message storage agent differs", async () => {
+      // given - boulder targets atlas, but nearest stored message still says hephaestus
+      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",
+        agent: "atlas",
+      }
+      writeBoulderState(TEST_DIR, state)
+
+      cleanupMessageStorage(MAIN_SESSION_ID)
+      setupMessageStorage(MAIN_SESSION_ID, "hephaestus")
+      updateSessionAgent(MAIN_SESSION_ID, "atlas")
+
+      const mockInput = createMockPluginInput()
+      const hook = createAtlasHook(mockInput)
+
+      // when
+      await hook.handler({
+        event: {
+          type: "session.idle",
+          properties: { sessionID: MAIN_SESSION_ID },
+        },
+      })
+
+      // then - should continue because start-work updated session agent to atlas
+      expect(mockInput._promptMock).toHaveBeenCalled()
+    })
   })
 })

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

@@ -1,12 +1,12 @@
 import { join } from "path"
 import { existsSync } from "fs"
 import { getClaudeConfigDir } from "../../shared"
-import type { ClaudeHooksConfig, HookMatcher, HookCommand } from "./types"
+import type { ClaudeHooksConfig, HookMatcher, HookAction } from "./types"
 
 interface RawHookMatcher {
   matcher?: string
   pattern?: string
-  hooks: HookCommand[]
+  hooks: HookAction[]
 }
 
 interface RawClaudeHooksConfig {

src/hooks/claude-code-hooks/dispatch-hook.ts

@@ -0,0 +1,27 @@
+import type { HookAction } from "./types"
+import type { CommandResult } from "../../shared/command-executor/execute-hook-command"
+import { executeHookCommand } from "../../shared"
+import { executeHttpHook } from "./execute-http-hook"
+import { DEFAULT_CONFIG } from "./plugin-config"
+
+export function getHookIdentifier(hook: HookAction): string {
+  if (hook.type === "http") return hook.url
+  return hook.command.split("/").pop() || hook.command
+}
+
+export async function dispatchHook(
+  hook: HookAction,
+  stdinJson: string,
+  cwd: string
+): Promise<CommandResult> {
+  if (hook.type === "http") {
+    return executeHttpHook(hook, stdinJson)
+  }
+
+  return executeHookCommand(
+    hook.command,
+    stdinJson,
+    cwd,
+    { forceZsh: DEFAULT_CONFIG.forceZsh, zshPath: DEFAULT_CONFIG.zshPath }
+  )
+}

src/hooks/claude-code-hooks/execute-http-hook.test.ts

@@ -0,0 +1,312 @@
+import { describe, it, expect, mock, beforeEach, afterEach } from "bun:test"
+import type { HookHttp } from "./types"
+
+const mockFetch = mock(() =>
+  Promise.resolve(new Response(JSON.stringify({}), { status: 200 }))
+)
+
+const originalFetch = globalThis.fetch
+
+describe("executeHttpHook", () => {
+  beforeEach(() => {
+    globalThis.fetch = mockFetch as unknown as typeof fetch
+    mockFetch.mockReset()
+    mockFetch.mockImplementation(() =>
+      Promise.resolve(new Response(JSON.stringify({}), { status: 200 }))
+    )
+  })
+
+  afterEach(() => {
+    globalThis.fetch = originalFetch
+  })
+
+  describe("#given a basic HTTP hook", () => {
+    const hook: HookHttp = {
+      type: "http",
+      url: "http://localhost:8080/hooks/pre-tool-use",
+    }
+    const stdinData = JSON.stringify({ hook_event_name: "PreToolUse", tool_name: "Bash" })
+
+    it("#when executed #then sends POST request with correct body", async () => {
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      await executeHttpHook(hook, stdinData)
+
+      expect(mockFetch).toHaveBeenCalledTimes(1)
+      const [url, options] = mockFetch.mock.calls[0] as unknown as [string, RequestInit]
+      expect(url).toBe("http://localhost:8080/hooks/pre-tool-use")
+      expect(options.method).toBe("POST")
+      expect(options.body).toBe(stdinData)
+    })
+
+    it("#when executed #then sets content-type to application/json", async () => {
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      await executeHttpHook(hook, stdinData)
+
+      const [, options] = mockFetch.mock.calls[0] as unknown as [string, RequestInit]
+      const headers = options.headers as Record<string, string>
+      expect(headers["Content-Type"]).toBe("application/json")
+    })
+  })
+
+  describe("#given an HTTP hook with headers and env var interpolation", () => {
+    const originalEnv = process.env
+
+    beforeEach(() => {
+      process.env = { ...originalEnv, MY_TOKEN: "secret-123", OTHER_VAR: "other-value" }
+    })
+
+    afterEach(() => {
+      process.env = originalEnv
+    })
+
+    it("#when allowedEnvVars includes the var #then interpolates env var in headers", async () => {
+      const hook: HookHttp = {
+        type: "http",
+        url: "http://localhost:8080/hooks",
+        headers: { Authorization: "Bearer $MY_TOKEN" },
+        allowedEnvVars: ["MY_TOKEN"],
+      }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      await executeHttpHook(hook, "{}")
+
+      const [, options] = mockFetch.mock.calls[0] as unknown as [string, RequestInit]
+      const headers = options.headers as Record<string, string>
+      expect(headers["Authorization"]).toBe("Bearer secret-123")
+    })
+
+    it("#when env var uses ${VAR} syntax #then interpolates correctly", async () => {
+      const hook: HookHttp = {
+        type: "http",
+        url: "http://localhost:8080/hooks",
+        headers: { Authorization: "Bearer ${MY_TOKEN}" },
+        allowedEnvVars: ["MY_TOKEN"],
+      }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      await executeHttpHook(hook, "{}")
+
+      const [, options] = mockFetch.mock.calls[0] as unknown as [string, RequestInit]
+      const headers = options.headers as Record<string, string>
+      expect(headers["Authorization"]).toBe("Bearer secret-123")
+    })
+
+    it("#when env var not in allowedEnvVars #then replaces with empty string", async () => {
+      const hook: HookHttp = {
+        type: "http",
+        url: "http://localhost:8080/hooks",
+        headers: { Authorization: "Bearer $OTHER_VAR" },
+        allowedEnvVars: ["MY_TOKEN"],
+      }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      await executeHttpHook(hook, "{}")
+
+      const [, options] = mockFetch.mock.calls[0] as unknown as [string, RequestInit]
+      const headers = options.headers as Record<string, string>
+      expect(headers["Authorization"]).toBe("Bearer ")
+    })
+  })
+
+  describe("#given an HTTP hook with timeout", () => {
+    it("#when timeout specified #then passes AbortSignal with timeout", async () => {
+      const hook: HookHttp = {
+        type: "http",
+        url: "http://localhost:8080/hooks",
+        timeout: 10,
+      }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      await executeHttpHook(hook, "{}")
+
+      const [, options] = mockFetch.mock.calls[0] as unknown as [string, RequestInit]
+      expect(options.signal).toBeDefined()
+    })
+  })
+
+  describe("#given hook URL scheme validation", () => {
+    it("#when URL uses file:// scheme #then rejects with exit code 1", async () => {
+      const hook: HookHttp = { type: "http", url: "file:///etc/passwd" }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      const result = await executeHttpHook(hook, "{}")
+
+      expect(result.exitCode).toBe(1)
+      expect(result.stderr).toContain('HTTP hook URL scheme "file:" is not allowed')
+      expect(mockFetch).not.toHaveBeenCalled()
+    })
+
+    it("#when URL uses data: scheme #then rejects with exit code 1", async () => {
+      const hook: HookHttp = { type: "http", url: "data:text/plain,hello" }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      const result = await executeHttpHook(hook, "{}")
+
+      expect(result.exitCode).toBe(1)
+      expect(result.stderr).toContain('HTTP hook URL scheme "data:" is not allowed')
+      expect(mockFetch).not.toHaveBeenCalled()
+    })
+
+    it("#when URL uses ftp:// scheme #then rejects with exit code 1", async () => {
+      const hook: HookHttp = { type: "http", url: "ftp://localhost/hooks" }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      const result = await executeHttpHook(hook, "{}")
+
+      expect(result.exitCode).toBe(1)
+      expect(result.stderr).toContain('HTTP hook URL scheme "ftp:" is not allowed')
+      expect(mockFetch).not.toHaveBeenCalled()
+    })
+
+    it("#when URL uses http:// scheme #then allows hook execution", async () => {
+      const hook: HookHttp = { type: "http", url: "http://localhost:8080/hooks" }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      const result = await executeHttpHook(hook, "{}")
+
+      expect(result.exitCode).toBe(0)
+      expect(mockFetch).toHaveBeenCalledTimes(1)
+    })
+
+    it("#when URL uses https:// scheme #then allows hook execution", async () => {
+      const hook: HookHttp = { type: "http", url: "https://example.com/hooks" }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      const result = await executeHttpHook(hook, "{}")
+
+      expect(result.exitCode).toBe(0)
+      expect(mockFetch).toHaveBeenCalledTimes(1)
+    })
+
+    it("#when URL is invalid #then rejects with exit code 1", async () => {
+      const hook: HookHttp = { type: "http", url: "not-a-valid-url" }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      const result = await executeHttpHook(hook, "{}")
+
+      expect(result.exitCode).toBe(1)
+      expect(result.stderr).toContain("HTTP hook URL is invalid: not-a-valid-url")
+      expect(mockFetch).not.toHaveBeenCalled()
+    })
+  })
+
+  describe("#given a successful HTTP response", () => {
+    it("#when response has JSON body #then returns parsed output", async () => {
+      mockFetch.mockImplementation(() =>
+        Promise.resolve(
+          new Response(JSON.stringify({ decision: "allow", reason: "ok" }), {
+            status: 200,
+            headers: { "Content-Type": "application/json" },
+          })
+        )
+      )
+      const hook: HookHttp = { type: "http", url: "http://localhost:8080/hooks" }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      const result = await executeHttpHook(hook, "{}")
+
+      expect(result.exitCode).toBe(0)
+      expect(result.stdout).toContain('"decision":"allow"')
+    })
+  })
+
+  describe("#given a failing HTTP response", () => {
+    it("#when response status is 4xx #then returns exit code 1", async () => {
+      mockFetch.mockImplementation(() =>
+        Promise.resolve(new Response("Bad Request", { status: 400 }))
+      )
+      const hook: HookHttp = { type: "http", url: "http://localhost:8080/hooks" }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      const result = await executeHttpHook(hook, "{}")
+
+      expect(result.exitCode).toBe(1)
+      expect(result.stderr).toContain("400")
+    })
+
+    it("#when fetch throws network error #then returns exit code 1", async () => {
+      mockFetch.mockImplementation(() => Promise.reject(new Error("ECONNREFUSED")))
+      const hook: HookHttp = { type: "http", url: "http://localhost:8080/hooks" }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      const result = await executeHttpHook(hook, "{}")
+
+      expect(result.exitCode).toBe(1)
+      expect(result.stderr).toContain("ECONNREFUSED")
+    })
+  })
+
+  describe("#given response with exit code in JSON", () => {
+    it("#when JSON contains exitCode 2 #then uses that exit code", async () => {
+      mockFetch.mockImplementation(() =>
+        Promise.resolve(
+          new Response(JSON.stringify({ exitCode: 2, stderr: "blocked" }), {
+            status: 200,
+            headers: { "Content-Type": "application/json" },
+          })
+        )
+      )
+      const hook: HookHttp = { type: "http", url: "http://localhost:8080/hooks" }
+      const { executeHttpHook } = await import("./execute-http-hook")
+
+      const result = await executeHttpHook(hook, "{}")
+
+      expect(result.exitCode).toBe(2)
+    })
+  })
+})
+
+describe("interpolateEnvVars", () => {
+  const originalEnv = process.env
+
+  beforeEach(() => {
+    process.env = { ...originalEnv, TOKEN: "abc", SECRET: "xyz" }
+  })
+
+  afterEach(() => {
+    process.env = originalEnv
+  })
+
+  it("#given $VAR syntax #when var is allowed #then interpolates", async () => {
+    const { interpolateEnvVars } = await import("./execute-http-hook")
+
+    const result = interpolateEnvVars("Bearer $TOKEN", ["TOKEN"])
+
+    expect(result).toBe("Bearer abc")
+  })
+
+  it("#given ${VAR} syntax #when var is allowed #then interpolates", async () => {
+    const { interpolateEnvVars } = await import("./execute-http-hook")
+
+    const result = interpolateEnvVars("Bearer ${TOKEN}", ["TOKEN"])
+
+    expect(result).toBe("Bearer abc")
+  })
+
+  it("#given multiple vars #when some not allowed #then only interpolates allowed ones", async () => {
+    const { interpolateEnvVars } = await import("./execute-http-hook")
+
+    const result = interpolateEnvVars("$TOKEN:$SECRET", ["TOKEN"])
+
+    expect(result).toBe("abc:")
+  })
+
+  it("#given ${VAR} where value contains $ANOTHER #when both allowed #then does not double-interpolate", async () => {
+    process.env = { ...process.env, TOKEN: "val$SECRET", SECRET: "oops" }
+    const { interpolateEnvVars } = await import("./execute-http-hook")
+
+    const result = interpolateEnvVars("Bearer ${TOKEN}", ["TOKEN", "SECRET"])
+
+    expect(result).toBe("Bearer val$SECRET")
+  })
+
+  it("#given no allowedEnvVars #when called #then replaces all with empty", async () => {
+    const { interpolateEnvVars } = await import("./execute-http-hook")
+
+    const result = interpolateEnvVars("Bearer $TOKEN", [])
+
+    expect(result).toBe("Bearer ")
+  })
+})

src/hooks/claude-code-hooks/execute-http-hook.ts

@@ -0,0 +1,92 @@
+import type { HookHttp } from "./types"
+import type { CommandResult } from "../../shared/command-executor/execute-hook-command"
+
+const DEFAULT_HTTP_HOOK_TIMEOUT_S = 30
+const ALLOWED_SCHEMES = new Set(["http:", "https:"])
+
+export function interpolateEnvVars(
+  value: string,
+  allowedEnvVars: string[]
+): string {
+  const allowedSet = new Set(allowedEnvVars)
+
+  return value.replace(/\$\{(\w+)\}|\$(\w+)/g, (_match, bracedVar: string | undefined, bareVar: string | undefined) => {
+    const varName = (bracedVar ?? bareVar) as string
+    if (allowedSet.has(varName)) {
+      return process.env[varName] ?? ""
+    }
+    return ""
+  })
+}
+
+function resolveHeaders(
+  hook: HookHttp
+): Record<string, string> {
+  const headers: Record<string, string> = {
+    "Content-Type": "application/json",
+  }
+
+  if (!hook.headers) return headers
+
+  const allowedEnvVars = hook.allowedEnvVars ?? []
+  for (const [key, value] of Object.entries(hook.headers)) {
+    headers[key] = interpolateEnvVars(value, allowedEnvVars)
+  }
+
+  return headers
+}
+
+export async function executeHttpHook(
+  hook: HookHttp,
+  stdin: string
+): Promise<CommandResult> {
+  try {
+    const parsed = new URL(hook.url)
+    if (!ALLOWED_SCHEMES.has(parsed.protocol)) {
+      return {
+        exitCode: 1,
+        stderr: `HTTP hook URL scheme "${parsed.protocol}" is not allowed. Only http: and https: are permitted.`,
+      }
+    }
+  } catch {
+    return { exitCode: 1, stderr: `HTTP hook URL is invalid: ${hook.url}` }
+  }
+
+  const timeoutS = hook.timeout ?? DEFAULT_HTTP_HOOK_TIMEOUT_S
+  const headers = resolveHeaders(hook)
+
+  try {
+    const response = await fetch(hook.url, {
+      method: "POST",
+      headers,
+      body: stdin,
+      signal: AbortSignal.timeout(timeoutS * 1000),
+    })
+
+    if (!response.ok) {
+      return {
+        exitCode: 1,
+        stderr: `HTTP hook returned status ${response.status}: ${response.statusText}`,
+        stdout: await response.text().catch(() => ""),
+      }
+    }
+
+    const body = await response.text()
+    if (!body) {
+      return { exitCode: 0, stdout: "", stderr: "" }
+    }
+
+    try {
+      const parsed = JSON.parse(body) as { exitCode?: number }
+      if (typeof parsed.exitCode === "number") {
+        return { exitCode: parsed.exitCode, stdout: body, stderr: "" }
+      }
+    } catch {
+    }
+
+    return { exitCode: 0, stdout: body, stderr: "" }
+  } catch (error) {
+    const message = error instanceof Error ? error.message : String(error)
+    return { exitCode: 1, stderr: `HTTP hook error: ${message}` }
+  }
+}

src/hooks/claude-code-hooks/post-tool-use.ts

@@ -3,8 +3,8 @@ import type {
   PostToolUseOutput,
   ClaudeHooksConfig,
 } from "./types"
-import { findMatchingHooks, executeHookCommand, objectToSnakeCase, transformToolName, log } from "../../shared"
-import { DEFAULT_CONFIG } from "./plugin-config"
+import { findMatchingHooks, objectToSnakeCase, transformToolName, log } from "../../shared"
+import { dispatchHook, getHookIdentifier } from "./dispatch-hook"
 import { buildTranscriptFromSession, deleteTempTranscript } from "./transcript"
 import { isHookCommandDisabled, type PluginExtendedConfig } from "./config-loader"
 
@@ -94,22 +94,17 @@ export async function executePostToolUseHooks(
      for (const matcher of matchers) {
        if (!matcher.hooks || matcher.hooks.length === 0) continue
        for (const hook of matcher.hooks) {
-         if (hook.type !== "command") continue
+         if (hook.type !== "command" && hook.type !== "http") continue
 
-         if (isHookCommandDisabled("PostToolUse", hook.command, extendedConfig ?? null)) {
-          log("PostToolUse hook command skipped (disabled by config)", { command: hook.command, toolName: ctx.toolName })
+        const hookName = getHookIdentifier(hook)
+        if (isHookCommandDisabled("PostToolUse", hookName, extendedConfig ?? null)) {
+          log("PostToolUse hook command skipped (disabled by config)", { command: hookName, toolName: ctx.toolName })
           continue
         }
 
-        const hookName = hook.command.split("/").pop() || hook.command
         if (!firstHookName) firstHookName = hookName
 
-        const result = await executeHookCommand(
-          hook.command,
-          JSON.stringify(stdinData),
-          ctx.cwd,
-          { forceZsh: DEFAULT_CONFIG.forceZsh, zshPath: DEFAULT_CONFIG.zshPath }
-        )
+        const result = await dispatchHook(hook, JSON.stringify(stdinData), ctx.cwd)
 
         if (result.stdout) {
           messages.push(result.stdout)

src/hooks/claude-code-hooks/pre-compact.ts

@@ -3,8 +3,8 @@ import type {
   PreCompactOutput,
   ClaudeHooksConfig,
 } from "./types"
-import { findMatchingHooks, executeHookCommand, log } from "../../shared"
-import { DEFAULT_CONFIG } from "./plugin-config"
+import { findMatchingHooks, log } from "../../shared"
+import { dispatchHook, getHookIdentifier } from "./dispatch-hook"
 import { isHookCommandDisabled, type PluginExtendedConfig } from "./config-loader"
 
 export interface PreCompactContext {
@@ -50,22 +50,17 @@ export async function executePreCompactHooks(
    for (const matcher of matchers) {
      if (!matcher.hooks || matcher.hooks.length === 0) continue
      for (const hook of matcher.hooks) {
-       if (hook.type !== "command") continue
+       if (hook.type !== "command" && hook.type !== "http") continue
 
-       if (isHookCommandDisabled("PreCompact", hook.command, extendedConfig ?? null)) {
-        log("PreCompact hook command skipped (disabled by config)", { command: hook.command })
+      const hookName = getHookIdentifier(hook)
+      if (isHookCommandDisabled("PreCompact", hookName, extendedConfig ?? null)) {
+        log("PreCompact hook command skipped (disabled by config)", { command: hookName })
         continue
       }
 
-      const hookName = hook.command.split("/").pop() || hook.command
       if (!firstHookName) firstHookName = hookName
 
-      const result = await executeHookCommand(
-        hook.command,
-        JSON.stringify(stdinData),
-        ctx.cwd,
-        { forceZsh: DEFAULT_CONFIG.forceZsh, zshPath: DEFAULT_CONFIG.zshPath }
-      )
+      const result = await dispatchHook(hook, JSON.stringify(stdinData), ctx.cwd)
 
       if (result.exitCode === 2) {
         log("PreCompact hook blocked", { hookName, stderr: result.stderr })

src/hooks/claude-code-hooks/pre-tool-use.ts

@@ -4,8 +4,8 @@ import type {
   PermissionDecision,
   ClaudeHooksConfig,
 } from "./types"
-import { findMatchingHooks, executeHookCommand, objectToSnakeCase, transformToolName, log } from "../../shared"
-import { DEFAULT_CONFIG } from "./plugin-config"
+import { findMatchingHooks, objectToSnakeCase, transformToolName, log } from "../../shared"
+import { dispatchHook, getHookIdentifier } from "./dispatch-hook"
 import { isHookCommandDisabled, type PluginExtendedConfig } from "./config-loader"
 
 export interface PreToolUseContext {
@@ -77,22 +77,17 @@ export async function executePreToolUseHooks(
    for (const matcher of matchers) {
      if (!matcher.hooks || matcher.hooks.length === 0) continue
      for (const hook of matcher.hooks) {
-       if (hook.type !== "command") continue
+       if (hook.type !== "command" && hook.type !== "http") continue
 
-       if (isHookCommandDisabled("PreToolUse", hook.command, extendedConfig ?? null)) {
-        log("PreToolUse hook command skipped (disabled by config)", { command: hook.command, toolName: ctx.toolName })
+      const hookName = getHookIdentifier(hook)
+      if (isHookCommandDisabled("PreToolUse", hookName, extendedConfig ?? null)) {
+        log("PreToolUse hook command skipped (disabled by config)", { command: hookName, toolName: ctx.toolName })
         continue
       }
 
-      const hookName = hook.command.split("/").pop() || hook.command
       if (!firstHookName) firstHookName = hookName
 
-      const result = await executeHookCommand(
-        hook.command,
-        JSON.stringify(stdinData),
-        ctx.cwd,
-        { forceZsh: DEFAULT_CONFIG.forceZsh, zshPath: DEFAULT_CONFIG.zshPath }
-      )
+      const result = await dispatchHook(hook, JSON.stringify(stdinData), ctx.cwd)
 
       if (result.exitCode === 2) {
         return {

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

@@ -3,8 +3,8 @@ import type {
   StopOutput,
   ClaudeHooksConfig,
 } from "./types"
-import { findMatchingHooks, executeHookCommand, log } from "../../shared"
-import { DEFAULT_CONFIG } from "./plugin-config"
+import { findMatchingHooks, log } from "../../shared"
+import { dispatchHook, getHookIdentifier } from "./dispatch-hook"
 import { getTodoPath } from "./todo"
 import { isHookCommandDisabled, type PluginExtendedConfig } from "./config-loader"
 
@@ -68,19 +68,15 @@ export async function executeStopHooks(
    for (const matcher of matchers) {
      if (!matcher.hooks || matcher.hooks.length === 0) continue
      for (const hook of matcher.hooks) {
-       if (hook.type !== "command") continue
+       if (hook.type !== "command" && hook.type !== "http") continue
 
-       if (isHookCommandDisabled("Stop", hook.command, extendedConfig ?? null)) {
-        log("Stop hook command skipped (disabled by config)", { command: hook.command })
+      const hookName = getHookIdentifier(hook)
+      if (isHookCommandDisabled("Stop", hookName, extendedConfig ?? null)) {
+        log("Stop hook command skipped (disabled by config)", { command: hookName })
         continue
       }
 
-      const result = await executeHookCommand(
-        hook.command,
-        JSON.stringify(stdinData),
-        ctx.cwd,
-        { forceZsh: DEFAULT_CONFIG.forceZsh, zshPath: DEFAULT_CONFIG.zshPath }
-      )
+      const result = await dispatchHook(hook, JSON.stringify(stdinData), ctx.cwd)
 
       // Check exit code first - exit code 2 means block
       if (result.exitCode === 2) {

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

@@ -12,7 +12,7 @@ export type ClaudeHookEvent =
 
 export interface HookMatcher {
   matcher: string
-  hooks: HookCommand[]
+  hooks: HookAction[]
 }
 
 export interface HookCommand {
@@ -20,6 +20,16 @@ export interface HookCommand {
   command: string
 }
 
+export interface HookHttp {
+  type: "http"
+  url: string
+  headers?: Record<string, string>
+  allowedEnvVars?: string[]
+  timeout?: number
+}
+
+export type HookAction = HookCommand | HookHttp
+
 export interface ClaudeHooksConfig {
   PreToolUse?: HookMatcher[]
   PostToolUse?: HookMatcher[]

src/hooks/claude-code-hooks/user-prompt-submit.ts

@@ -3,8 +3,8 @@ import type {
   PostToolUseOutput,
   ClaudeHooksConfig,
 } from "./types"
-import { findMatchingHooks, executeHookCommand, log } from "../../shared"
-import { DEFAULT_CONFIG } from "./plugin-config"
+import { findMatchingHooks, log } from "../../shared"
+import { dispatchHook, getHookIdentifier } from "./dispatch-hook"
 import { isHookCommandDisabled, type PluginExtendedConfig } from "./config-loader"
 
 const USER_PROMPT_SUBMIT_TAG_OPEN = "<user-prompt-submit-hook>"
@@ -80,19 +80,15 @@ export async function executeUserPromptSubmitHooks(
    for (const matcher of matchers) {
      if (!matcher.hooks || matcher.hooks.length === 0) continue
      for (const hook of matcher.hooks) {
-       if (hook.type !== "command") continue
+       if (hook.type !== "command" && hook.type !== "http") continue
 
-      if (isHookCommandDisabled("UserPromptSubmit", hook.command, extendedConfig ?? null)) {
-        log("UserPromptSubmit hook command skipped (disabled by config)", { command: hook.command })
+      const hookName = getHookIdentifier(hook)
+      if (isHookCommandDisabled("UserPromptSubmit", hookName, extendedConfig ?? null)) {
+        log("UserPromptSubmit hook command skipped (disabled by config)", { command: hookName })
         continue
       }
 
-      const result = await executeHookCommand(
-        hook.command,
-        JSON.stringify(stdinData),
-        ctx.cwd,
-        { forceZsh: DEFAULT_CONFIG.forceZsh, zshPath: DEFAULT_CONFIG.zshPath }
-      )
+      const result = await dispatchHook(hook, JSON.stringify(stdinData), ctx.cwd)
 
       if (result.stdout) {
         const output = result.stdout.trim()

src/hooks/index.ts

@@ -50,3 +50,4 @@ export { createRuntimeFallbackHook, type RuntimeFallbackHook, type RuntimeFallba
 export { createWriteExistingFileGuardHook } from "./write-existing-file-guard";
 export { createHashlineReadEnhancerHook } from "./hashline-read-enhancer";
 export { createJsonErrorRecoveryHook, JSON_ERROR_TOOL_EXCLUDE_LIST, JSON_ERROR_PATTERNS, JSON_ERROR_REMINDER } from "./json-error-recovery";
+export { createReadImageResizerHook } from "./read-image-resizer"

src/hooks/interactive-bash-session/interactive-bash-session-tracker.ts

@@ -6,6 +6,7 @@ import {
 import { OMO_SESSION_PREFIX, buildSessionReminderMessage } from "./constants";
 import type { InteractiveBashSessionState } from "./types";
 import { subagentSessions } from "../../features/claude-code-session-state";
+import { spawnWithWindowsHide } from "../../shared/spawn-with-windows-hide";
 
 type AbortSession = (args: { path: { id: string } }) => Promise<unknown>
 
@@ -19,7 +20,7 @@ async function killAllTrackedSessions(
 ): Promise<void> {
   for (const sessionName of state.tmuxSessions) {
     try {
-      const proc = Bun.spawn(["tmux", "kill-session", "-t", sessionName], {
+      const proc = spawnWithWindowsHide(["tmux", "kill-session", "-t", sessionName], {
         stdout: "ignore",
         stderr: "ignore",
       })

src/hooks/interactive-bash-session/state-manager.ts

@@ -1,6 +1,7 @@
 import type { InteractiveBashSessionState } from "./types";
 import { loadInteractiveBashSessionState } from "./storage";
 import { OMO_SESSION_PREFIX } from "./constants";
+import { spawnWithWindowsHide } from "../../shared/spawn-with-windows-hide";
 
 export function getOrCreateState(sessionID: string, sessionStates: Map<string, InteractiveBashSessionState>): InteractiveBashSessionState {
   if (!sessionStates.has(sessionID)) {
@@ -24,7 +25,7 @@ export async function killAllTrackedSessions(
 ): Promise<void> {
   for (const sessionName of state.tmuxSessions) {
     try {
-      const proc = Bun.spawn(["tmux", "kill-session", "-t", sessionName], {
+      const proc = spawnWithWindowsHide(["tmux", "kill-session", "-t", sessionName], {
         stdout: "ignore",
         stderr: "ignore",
       });

src/hooks/ralph-loop/completion-promise-detector.test.ts

@@ -0,0 +1,111 @@
+/// <reference types="bun-types" />
+import { describe, expect, test } from "bun:test"
+import type { PluginInput } from "@opencode-ai/plugin"
+import { detectCompletionInSessionMessages } from "./completion-promise-detector"
+
+type SessionMessage = {
+  info?: { role?: string }
+  parts?: Array<{ type: string; text?: string }>
+}
+
+function createPluginInput(messages: SessionMessage[]): PluginInput {
+  const pluginInput = {
+    client: { session: {} } as PluginInput["client"],
+    project: {} as PluginInput["project"],
+    directory: "/tmp",
+    worktree: "/tmp",
+    serverUrl: new URL("http://localhost"),
+    $: {} as PluginInput["$"],
+  } as PluginInput
+
+  pluginInput.client.session.messages =
+    (async () => ({ data: messages })) as unknown as PluginInput["client"]["session"]["messages"]
+
+  return pluginInput
+}
+
+describe("detectCompletionInSessionMessages", () => {
+  describe("#given session with prior DONE and new messages", () => {
+    test("#when sinceMessageIndex excludes prior DONE #then should NOT detect completion", async () => {
+      // #given
+      const messages: SessionMessage[] = [
+        {
+          info: { role: "assistant" },
+          parts: [{ type: "text", text: "Old completion <promise>DONE</promise>" }],
+        },
+        {
+          info: { role: "assistant" },
+          parts: [{ type: "text", text: "Working on the new task" }],
+        },
+      ]
+      const ctx = createPluginInput(messages)
+
+      // #when
+      const detected = await detectCompletionInSessionMessages(ctx, {
+        sessionID: "session-123",
+        promise: "DONE",
+        apiTimeoutMs: 1000,
+        directory: "/tmp",
+        sinceMessageIndex: 1,
+      })
+
+      // #then
+      expect(detected).toBe(false)
+    })
+
+    test("#when sinceMessageIndex includes current DONE #then should detect completion", async () => {
+      // #given
+      const messages: SessionMessage[] = [
+        {
+          info: { role: "assistant" },
+          parts: [{ type: "text", text: "Old completion <promise>DONE</promise>" }],
+        },
+        {
+          info: { role: "assistant" },
+          parts: [{ type: "text", text: "Current completion <promise>DONE</promise>" }],
+        },
+      ]
+      const ctx = createPluginInput(messages)
+
+      // #when
+      const detected = await detectCompletionInSessionMessages(ctx, {
+        sessionID: "session-123",
+        promise: "DONE",
+        apiTimeoutMs: 1000,
+        directory: "/tmp",
+        sinceMessageIndex: 1,
+      })
+
+      // #then
+      expect(detected).toBe(true)
+    })
+  })
+
+  describe("#given no sinceMessageIndex (backward compat)", () => {
+    test("#then should scan all messages", async () => {
+      // #given
+      const messages: SessionMessage[] = [
+        {
+          info: { role: "assistant" },
+          parts: [{ type: "text", text: "Old completion <promise>DONE</promise>" }],
+        },
+        {
+          info: { role: "assistant" },
+          parts: [{ type: "text", text: "No completion in latest message" }],
+        },
+      ]
+      const ctx = createPluginInput(messages)
+
+      // #when
+      const detected = await detectCompletionInSessionMessages(ctx, {
+        sessionID: "session-123",
+        promise: "DONE",
+        apiTimeoutMs: 1000,
+        directory: "/tmp",
+      })
+
+      // #then
+      expect(detected).toBe(true)
+    })
+  })
+})

src/hooks/ralph-loop/completion-promise-detector.ts

@@ -52,6 +52,7 @@ export async function detectCompletionInSessionMessages(
 		promise: string
 		apiTimeoutMs: number
 		directory: string
+		sinceMessageIndex?: number
 	},
 ): Promise<boolean> {
 	try {
@@ -75,12 +76,17 @@ export async function detectCompletionInSessionMessages(
 				? responseData
 				: []
 
-		const assistantMessages = (messageArray as OpenCodeSessionMessage[]).filter((msg) => msg.info?.role === "assistant")
+		const scopedMessages =
+			typeof options.sinceMessageIndex === "number" && options.sinceMessageIndex >= 0 && options.sinceMessageIndex < messageArray.length
+				? messageArray.slice(options.sinceMessageIndex)
+				: messageArray
+
+		const assistantMessages = (scopedMessages as OpenCodeSessionMessage[]).filter((msg) => msg.info?.role === "assistant")
 		if (assistantMessages.length === 0) return false
 
 		const pattern = buildPromisePattern(options.promise)
-		const recentAssistants = assistantMessages.slice(-3)
-		for (const assistant of recentAssistants) {
+		for (let index = assistantMessages.length - 1; index >= 0; index -= 1) {
+			const assistant = assistantMessages[index]
 			if (!assistant.parts) continue
 
 			let responseText = ""

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

@@ -494,6 +494,7 @@ describe("ralph-loop", () => {
         config: {
           enabled: true,
           default_max_iterations: 200,
+          default_strategy: "continue",
         },
       })
 
@@ -602,7 +603,7 @@ describe("ralph-loop", () => {
       expect(hook.getState()).toBeNull()
 
       // then - messages API was called with correct session ID
-      expect(messagesCalls.length).toBe(1)
+      expect(messagesCalls.length).toBe(2)
       expect(messagesCalls[0].sessionID).toBe("session-123")
     })
 
@@ -632,7 +633,7 @@ describe("ralph-loop", () => {
       expect(hook.getState()).toBeNull()
 
       // then - messages API was called with correct session ID
-      expect(messagesCalls.length).toBe(1)
+      expect(messagesCalls.length).toBe(2)
       expect(messagesCalls[0].sessionID).toBe("session-123")
     })
 
@@ -708,6 +709,57 @@ describe("ralph-loop", () => {
       expect(promptCalls[0].text).toContain("<promise>CALCULATOR_DONE</promise>")
     })
 
+    test("should skip concurrent idle events for same session when handler is in flight", async () => {
+      // given - active loop with delayed prompt injection
+      let releasePromptAsync: (() => void) | undefined
+      const promptAsyncBlocked = new Promise<void>((resolve) => {
+        releasePromptAsync = resolve
+      })
+      let firstPromptStartedResolve: (() => void) | undefined
+      const firstPromptStarted = new Promise<void>((resolve) => {
+        firstPromptStartedResolve = resolve
+      })
+
+      const mockInput = createMockPluginInput() as {
+        client: {
+          session: {
+            promptAsync: (opts: { path: { id: string }; body: { parts: Array<{ type: string; text: string }> } }) => Promise<unknown>
+          }
+        }
+      }
+
+      const originalPromptAsync = mockInput.client.session.promptAsync
+      let promptAsyncCalls = 0
+      mockInput.client.session.promptAsync = async (opts) => {
+        promptAsyncCalls += 1
+        if (promptAsyncCalls === 1) {
+          firstPromptStartedResolve?.()
+        }
+        await promptAsyncBlocked
+        return originalPromptAsync(opts)
+      }
+
+      const hook = createRalphLoopHook(mockInput as Parameters<typeof createRalphLoopHook>[0])
+      hook.startLoop("session-123", "Build feature", { maxIterations: 10 })
+
+      // when - second idle arrives while first idle processing is still in flight
+      const firstIdle = hook.event({
+        event: { type: "session.idle", properties: { sessionID: "session-123" } },
+      })
+      await firstPromptStarted
+      const secondIdle = hook.event({
+        event: { type: "session.idle", properties: { sessionID: "session-123" } },
+      })
+
+      releasePromptAsync?.()
+      await Promise.all([firstIdle, secondIdle])
+
+      // then - only one continuation should be injected
+      expect(promptAsyncCalls).toBe(1)
+      expect(promptCalls.length).toBe(1)
+      expect(hook.getState()?.iteration).toBe(2)
+    })
+
     test("should clear loop state on user abort (MessageAbortedError)", async () => {
       // given - active loop
       const hook = createRalphLoopHook(createMockPluginInput())
@@ -782,8 +834,8 @@ describe("ralph-loop", () => {
       expect(hook.getState()).toBeNull()
     })
 
-    test("should NOT detect completion if promise is older than last 3 assistant messages", async () => {
-      // given - promise appears in an assistant message older than last 3
+    test("should detect completion even when promise is older than previous narrow window", async () => {
+      // given - promise appears in an older assistant message with additional assistant output after it
       mockSessionMessages = [
         { info: { role: "user" }, parts: [{ type: "text", text: "Start task" }] },
         { info: { role: "assistant" }, parts: [{ type: "text", text: "Promise early <promise>DONE</promise>" }] },
@@ -801,9 +853,40 @@ describe("ralph-loop", () => {
         event: { type: "session.idle", properties: { sessionID: "session-123" } },
       })
 
-      // then - loop should continue (promise is older than last 3 assistant messages)
-      expect(promptCalls.length).toBe(1)
-      expect(hook.getState()?.iteration).toBe(2)
+      // then - loop should complete because all assistant messages are scanned
+      expect(promptCalls.length).toBe(0)
+      expect(toastCalls.some((t) => t.title === "Ralph Loop Complete!")).toBe(true)
+      expect(hook.getState()).toBeNull()
+    })
+
+    test("should detect completion when many assistant messages are emitted after promise", async () => {
+      // given - completion promise followed by long assistant output sequence
+      mockSessionMessages = [
+        { info: { role: "user" }, parts: [{ type: "text", text: "Start task" }] },
+        { info: { role: "assistant" }, parts: [{ type: "text", text: "Done now <promise>DONE</promise>" }] },
+      ]
+
+      for (let index = 1; index <= 25; index += 1) {
+        mockSessionMessages.push({
+          info: { role: "assistant" },
+          parts: [{ type: "text", text: `Post-completion assistant output ${index}` }],
+        })
+      }
+
+      const hook = createRalphLoopHook(createMockPluginInput(), {
+        getTranscriptPath: () => join(TEST_DIR, "nonexistent.jsonl"),
+      })
+      hook.startLoop("session-123", "Build something", { completionPromise: "DONE" })
+
+      // when - session goes idle
+      await hook.event({
+        event: { type: "session.idle", properties: { sessionID: "session-123" } },
+      })
+
+      // then - loop should complete despite large trailing output
+      expect(promptCalls.length).toBe(0)
+      expect(toastCalls.some((t) => t.title === "Ralph Loop Complete!")).toBe(true)
+      expect(hook.getState()).toBeNull()
     })
 
     test("should allow starting new loop while previous loop is active (different session)", async () => {
@@ -992,7 +1075,7 @@ Original task: Build something`
       expect(promptCalls.length).toBe(0)
       expect(hook.getState()).toBeNull()
       // API should NOT be called since transcript found completion
-      expect(messagesCalls.length).toBe(0)
+      expect(messagesCalls.length).toBe(1)
     })
 
     test("should show ultrawork completion toast", async () => {

src/hooks/ralph-loop/loop-state-controller.ts

@@ -23,6 +23,7 @@ export function createLoopStateController(options: {
 			loopOptions?: {
 				maxIterations?: number
 				completionPromise?: string
+				messageCountAtStart?: number
 				ultrawork?: boolean
 				strategy?: "reset" | "continue"
 			},
@@ -34,6 +35,7 @@ export function createLoopStateController(options: {
 					loopOptions?.maxIterations ??
 					config?.default_max_iterations ??
 					DEFAULT_MAX_ITERATIONS,
+				message_count_at_start: loopOptions?.messageCountAtStart,
 				completion_promise:
 					loopOptions?.completionPromise ??
 					DEFAULT_COMPLETION_PROMISE,
@@ -93,5 +95,19 @@ export function createLoopStateController(options: {
 
 			return state
 		},
+
+		setMessageCountAtStart(sessionID: string, messageCountAtStart: number): RalphLoopState | null {
+			const state = readState(directory, stateDir)
+			if (!state || state.session_id !== sessionID) {
+				return null
+			}
+
+			state.message_count_at_start = messageCountAtStart
+			if (!writeState(directory, state, stateDir)) {
+				return null
+			}
+
+			return state
+		},
 	}
 }

src/hooks/ralph-loop/ralph-loop-event-handler.ts

@@ -25,6 +25,8 @@ export function createRalphLoopEventHandler(
 	ctx: PluginInput,
 	options: RalphLoopEventHandlerOptions,
 ) {
+	const inFlightSessions = new Set<string>()
+
 	return async ({ event }: { event: { type: string; properties?: unknown } }): Promise<void> => {
 		const props = event.properties as Record<string, unknown> | undefined
 
@@ -32,115 +34,128 @@ export function createRalphLoopEventHandler(
 			const sessionID = props?.sessionID as string | undefined
 			if (!sessionID) return
 
-			if (options.sessionRecovery.isRecovering(sessionID)) {
-				log(`[${HOOK_NAME}] Skipped: in recovery`, { sessionID })
+			if (inFlightSessions.has(sessionID)) {
+				log(`[${HOOK_NAME}] Skipped: handler in flight`, { sessionID })
 				return
 			}
 
-			const state = options.loopState.getState()
-			if (!state || !state.active) {
-				return
-			}
-
-			if (state.session_id && state.session_id !== sessionID) {
-				if (options.checkSessionExists) {
-					try {
-						const exists = await options.checkSessionExists(state.session_id)
-						if (!exists) {
-							options.loopState.clear()
-							log(`[${HOOK_NAME}] Cleared orphaned state from deleted session`, {
-								orphanedSessionId: state.session_id,
-								currentSessionId: sessionID,
-							})
-							return
-						}
-					} catch (err) {
-						log(`[${HOOK_NAME}] Failed to check session existence`, {
-							sessionId: state.session_id,
-							error: String(err),
-						})
-					}
-				}
-				return
-			}
-
-			const transcriptPath = options.getTranscriptPath(sessionID)
-			const completionViaTranscript = detectCompletionInTranscript(transcriptPath, state.completion_promise)
-			const completionViaApi = completionViaTranscript
-				? false
-				: await detectCompletionInSessionMessages(ctx, {
-					sessionID,
-					promise: state.completion_promise,
-					apiTimeoutMs: options.apiTimeoutMs,
-					directory: options.directory,
-				})
-
-			if (completionViaTranscript || completionViaApi) {
-				log(`[${HOOK_NAME}] Completion detected!`, {
-					sessionID,
-					iteration: state.iteration,
-					promise: state.completion_promise,
-					detectedVia: completionViaTranscript
-						? "transcript_file"
-						: "session_messages_api",
-				})
-				options.loopState.clear()
-
-				const title = state.ultrawork ? "ULTRAWORK LOOP COMPLETE!" : "Ralph Loop Complete!"
-				const message = state.ultrawork ? `JUST ULW ULW! Task completed after ${state.iteration} iteration(s)` : `Task completed after ${state.iteration} iteration(s)`
-				await ctx.client.tui?.showToast?.({ body: { title, message, variant: "success", duration: 5000 } }).catch(() => {})
-				return
-			}
-
-			if (state.iteration >= state.max_iterations) {
-				log(`[${HOOK_NAME}] Max iterations reached`, {
-					sessionID,
-					iteration: state.iteration,
-					max: state.max_iterations,
-				})
-				options.loopState.clear()
-
-				await ctx.client.tui?.showToast?.({
-					body: { title: "Ralph Loop Stopped", message: `Max iterations (${state.max_iterations}) reached without completion`, variant: "warning", duration: 5000 },
-					}).catch(() => {})
-				return
-			}
-
-			const newState = options.loopState.incrementIteration()
-			if (!newState) {
-				log(`[${HOOK_NAME}] Failed to increment iteration`, { sessionID })
-				return
-			}
-
-			log(`[${HOOK_NAME}] Continuing loop`, {
-				sessionID,
-				iteration: newState.iteration,
-				max: newState.max_iterations,
-			})
-
-			await ctx.client.tui?.showToast?.({
-				body: {
-					title: "Ralph Loop",
-					message: `Iteration ${newState.iteration}/${newState.max_iterations}`,
-					variant: "info",
-					duration: 2000,
-				},
-				}).catch(() => {})
+			inFlightSessions.add(sessionID)
 
 			try {
-				await continueIteration(ctx, newState, {
-					previousSessionID: sessionID,
-					directory: options.directory,
-					apiTimeoutMs: options.apiTimeoutMs,
-					loopState: options.loopState,
-				})
-			} catch (err) {
-				log(`[${HOOK_NAME}] Failed to inject continuation`, {
+
+				if (options.sessionRecovery.isRecovering(sessionID)) {
+					log(`[${HOOK_NAME}] Skipped: in recovery`, { sessionID })
+					return
+				}
+
+				const state = options.loopState.getState()
+				if (!state || !state.active) {
+					return
+				}
+
+				if (state.session_id && state.session_id !== sessionID) {
+					if (options.checkSessionExists) {
+						try {
+							const exists = await options.checkSessionExists(state.session_id)
+							if (!exists) {
+								options.loopState.clear()
+								log(`[${HOOK_NAME}] Cleared orphaned state from deleted session`, {
+									orphanedSessionId: state.session_id,
+									currentSessionId: sessionID,
+								})
+								return
+							}
+						} catch (err) {
+							log(`[${HOOK_NAME}] Failed to check session existence`, {
+								sessionId: state.session_id,
+								error: String(err),
+							})
+						}
+					}
+					return
+				}
+
+				const transcriptPath = options.getTranscriptPath(sessionID)
+				const completionViaTranscript = detectCompletionInTranscript(transcriptPath, state.completion_promise)
+				const completionViaApi = completionViaTranscript
+					? false
+					: await detectCompletionInSessionMessages(ctx, {
+						sessionID,
+						promise: state.completion_promise,
+						apiTimeoutMs: options.apiTimeoutMs,
+						directory: options.directory,
+						sinceMessageIndex: state.message_count_at_start,
+					})
+
+				if (completionViaTranscript || completionViaApi) {
+					log(`[${HOOK_NAME}] Completion detected!`, {
+						sessionID,
+						iteration: state.iteration,
+						promise: state.completion_promise,
+						detectedVia: completionViaTranscript
+							? "transcript_file"
+							: "session_messages_api",
+					})
+					options.loopState.clear()
+
+					const title = state.ultrawork ? "ULTRAWORK LOOP COMPLETE!" : "Ralph Loop Complete!"
+					const message = state.ultrawork ? `JUST ULW ULW! Task completed after ${state.iteration} iteration(s)` : `Task completed after ${state.iteration} iteration(s)`
+					await ctx.client.tui?.showToast?.({ body: { title, message, variant: "success", duration: 5000 } }).catch(() => {})
+					return
+				}
+
+				if (state.iteration >= state.max_iterations) {
+					log(`[${HOOK_NAME}] Max iterations reached`, {
+						sessionID,
+						iteration: state.iteration,
+						max: state.max_iterations,
+					})
+					options.loopState.clear()
+
+					await ctx.client.tui?.showToast?.({
+						body: { title: "Ralph Loop Stopped", message: `Max iterations (${state.max_iterations}) reached without completion`, variant: "warning", duration: 5000 },
+						}).catch(() => {})
+					return
+				}
+
+				const newState = options.loopState.incrementIteration()
+				if (!newState) {
+					log(`[${HOOK_NAME}] Failed to increment iteration`, { sessionID })
+					return
+				}
+
+				log(`[${HOOK_NAME}] Continuing loop`, {
 					sessionID,
-					error: String(err),
+					iteration: newState.iteration,
+					max: newState.max_iterations,
 				})
+
+				await ctx.client.tui?.showToast?.({
+					body: {
+						title: "Ralph Loop",
+						message: `Iteration ${newState.iteration}/${newState.max_iterations}`,
+						variant: "info",
+						duration: 2000,
+					},
+					}).catch(() => {})
+
+				try {
+					await continueIteration(ctx, newState, {
+						previousSessionID: sessionID,
+						directory: options.directory,
+						apiTimeoutMs: options.apiTimeoutMs,
+						loopState: options.loopState,
+					})
+				} catch (err) {
+					log(`[${HOOK_NAME}] Failed to inject continuation`, {
+						sessionID,
+						error: String(err),
+					})
+				}
+				return
+			} finally {
+				inFlightSessions.delete(sessionID)
 			}
-			return
 		}
 
 		if (event.type === "session.deleted") {

src/hooks/ralph-loop/ralph-loop-hook.ts

@@ -13,6 +13,7 @@ export interface RalphLoopHook {
     options?: {
       maxIterations?: number
       completionPromise?: string
+      messageCountAtStart?: number
       ultrawork?: boolean
       strategy?: "reset" | "continue"
     }
@@ -23,6 +24,19 @@ export interface RalphLoopHook {
 
 const DEFAULT_API_TIMEOUT = 5000 as const
 
+function getMessageCountFromResponse(messagesResponse: unknown): number {
+  if (Array.isArray(messagesResponse)) {
+    return messagesResponse.length
+  }
+
+  if (typeof messagesResponse === "object" && messagesResponse !== null && "data" in messagesResponse) {
+    const data = (messagesResponse as { data?: unknown }).data
+    return Array.isArray(data) ? data.length : 0
+  }
+
+  return 0
+}
+
 export function createRalphLoopHook(
   ctx: PluginInput,
   options?: RalphLoopOptions
@@ -51,7 +65,25 @@ export function createRalphLoopHook(
 
 	return {
 		event,
-		startLoop: loopState.startLoop,
+		startLoop: (sessionID, prompt, loopOptions): boolean => {
+			const startSuccess = loopState.startLoop(sessionID, prompt, loopOptions)
+			if (!startSuccess || typeof loopOptions?.messageCountAtStart === "number") {
+				return startSuccess
+			}
+
+			ctx.client.session
+				.messages({
+					path: { id: sessionID },
+					query: { directory: ctx.directory },
+				})
+				.then((messagesResponse: unknown) => {
+					const messageCountAtStart = getMessageCountFromResponse(messagesResponse)
+					loopState.setMessageCountAtStart(sessionID, messageCountAtStart)
+				})
+				.catch(() => {})
+
+			return startSuccess
+		},
 		cancelLoop: loopState.cancelLoop,
 		getState: loopState.getState as () => RalphLoopState | null,
 	}

src/hooks/ralph-loop/reset-strategy-race-condition.test.ts

@@ -36,7 +36,7 @@ async function waitUntil(condition: () => boolean): Promise<void> {
 }
 
 describe("ralph-loop reset strategy race condition", () => {
-  test("should continue iteration when old session idle arrives before TUI switch completes", async () => {
+  test("should skip duplicate idle while reset iteration handling is in flight", async () => {
     // given - reset strategy loop with blocked TUI session switch
     const promptCalls: Array<{ sessionID: string; text: string }> = []
     const createSessionCalls: Array<{ parentID?: string }> = []
@@ -85,7 +85,7 @@ describe("ralph-loop reset strategy race condition", () => {
           },
         },
       },
-    } as Parameters<typeof createRalphLoopHook>[0])
+    } as unknown as Parameters<typeof createRalphLoopHook>[0])
 
     hook.startLoop("session-old", "Build feature", { strategy: "reset" })
 
@@ -100,14 +100,12 @@ describe("ralph-loop reset strategy race condition", () => {
       event: { type: "session.idle", properties: { sessionID: "session-old" } },
     })
 
-    await waitUntil(() => selectSessionCalls > 1)
-
     selectSessionDeferred.resolve()
     await Promise.all([firstIdleEvent, secondIdleEvent])
 
-    // then - second idle should not be skipped during reset transition
-    expect(createSessionCalls.length).toBe(2)
-    expect(promptCalls.length).toBe(2)
-    expect(hook.getState()?.iteration).toBe(3)
+    // then - duplicate idle should be skipped to prevent concurrent continuation injection
+    expect(createSessionCalls.length).toBe(1)
+    expect(promptCalls.length).toBe(1)
+    expect(hook.getState()?.iteration).toBe(2)
   })
 })