feat(08-01): return council task ids without blocking

- make athena_council launch-only and remove internal polling/formatting

- return JSON payload with running task mappings and launch failures

- update tool tests for task-id visibility, filtering, failure reporting, and dedup
This commit is contained in:
ismeth
2026-02-13 11:19:25 +01:00
committed by YeonGyu-Kim
parent 9a69478d8e
commit 5816cdddc6
4 changed files with 204 additions and 104 deletions

View File

@@ -1,9 +1,9 @@
export const ATHENA_COUNCIL_TOOL_DESCRIPTION_TEMPLATE = `Execute Athena's multi-model council. Sends the question to all configured council members in parallel and returns their collected responses.
export const ATHENA_COUNCIL_TOOL_DESCRIPTION_TEMPLATE = `Execute Athena's multi-model council. Launches council members as background tasks and returns their task IDs immediately.
Optionally pass a members array of member names or model IDs to consult only specific council members. If omitted, all configured members are consulted.
{members}
Returns council member responses with status, response text, and timing. Use this output for synthesis.
Use background_output(task_id=...) to retrieve each member's response. The system will notify you when tasks complete.
IMPORTANT: This tool is designed for Athena agent use only. It requires council configuration to be present.`

View File

@@ -2,7 +2,7 @@
import { describe, expect, test } from "bun:test"
import type { BackgroundManager } from "../../features/background-agent"
import { ATHENA_COUNCIL_TOOL_DESCRIPTION_TEMPLATE } from "./constants"
import type { BackgroundTask } from "../../features/background-agent/types"
import { createAthenaCouncilTool, filterCouncilMembers } from "./tools"
const mockManager = {
@@ -25,6 +25,27 @@ const configuredMembers = [
{ model: "google/gemini-3-pro" },
]
function createRunningTask(id: string): BackgroundTask {
return {
id,
parentSessionID: "session-1",
parentMessageID: "message-1",
description: `Council member task ${id}`,
prompt: "prompt",
agent: "athena",
status: "running",
}
}
function parseLaunchResult(result: unknown): {
launched: number
members: Array<{ task_id: string; name: string; model: string; status: string }>
failed: Array<{ name: string; model: string; error: string }>
} {
expect(typeof result).toBe("string")
return JSON.parse(result as string)
}
describe("filterCouncilMembers", () => {
test("returns all members when selection is undefined", () => {
// #given
@@ -151,4 +172,140 @@ describe("createAthenaCouncilTool", () => {
// #then
expect(result).toBe("Unknown council members: unknown-model. Available members: Claude, GPT, google/gemini-3-pro.")
})
test("returns launched task_ids and member mapping for configured council", async () => {
// #given
let launchCount = 0
const launchManager = {
launch: async () => {
launchCount += 1
return createRunningTask(`bg-${launchCount}`)
},
getTask: () => undefined,
} as unknown as BackgroundManager
const athenaCouncilTool = createAthenaCouncilTool({
backgroundManager: launchManager,
councilConfig: { members: configuredMembers },
})
// #when
const result = await athenaCouncilTool.execute({ question: "How should we proceed?" }, mockToolContext)
const parsed = parseLaunchResult(result)
// #then
expect(parsed.launched).toBe(3)
expect(parsed.failed).toEqual([])
expect(parsed.members).toEqual([
{
task_id: "bg-1",
name: "Claude",
model: "anthropic/claude-sonnet-4-5",
status: "running",
},
{
task_id: "bg-2",
name: "GPT",
model: "openai/gpt-5.3-codex",
status: "running",
},
{
task_id: "bg-3",
name: "google/gemini-3-pro",
model: "google/gemini-3-pro",
status: "running",
},
])
})
test("returns task_ids length matching selected members", async () => {
// #given
let launchCount = 0
const launchManager = {
launch: async () => {
launchCount += 1
return createRunningTask(`bg-${launchCount}`)
},
getTask: () => undefined,
} as unknown as BackgroundManager
const athenaCouncilTool = createAthenaCouncilTool({
backgroundManager: launchManager,
councilConfig: { members: configuredMembers },
})
// #when
const result = await athenaCouncilTool.execute(
{
question: "Who should investigate this?",
members: ["GPT", "google/gemini-3-pro"],
},
mockToolContext
)
const parsed = parseLaunchResult(result)
// #then
expect(parsed.launched).toBe(2)
expect(parsed.members).toHaveLength(2)
expect(parsed.members.map((member) => member.name)).toEqual(["GPT", "google/gemini-3-pro"])
})
test("returns failed launches inline while keeping successful task mappings", async () => {
// #given
let launchCount = 0
const launchManager = {
launch: async () => {
launchCount += 1
if (launchCount === 2) {
throw new Error("provider outage")
}
return createRunningTask(`bg-${launchCount}`)
},
getTask: () => undefined,
} as unknown as BackgroundManager
const athenaCouncilTool = createAthenaCouncilTool({
backgroundManager: launchManager,
councilConfig: { members: configuredMembers },
})
// #when
const result = await athenaCouncilTool.execute({ question: "Any concerns?" }, mockToolContext)
const parsed = parseLaunchResult(result)
// #then
expect(parsed.launched).toBe(2)
expect(parsed.members).toHaveLength(2)
expect(parsed.failed).toHaveLength(1)
expect(parsed.failed[0]).toEqual({
name: "GPT",
model: "openai/gpt-5.3-codex",
error: "Launch failed: Error: provider outage",
})
})
test("returns dedup error when council is already running in same session", async () => {
// #given
let resolveLaunch: ((task: BackgroundTask) => void) | undefined
const pendingLaunch = new Promise<BackgroundTask>((resolve) => {
resolveLaunch = resolve
})
const launchManager = {
launch: async () => pendingLaunch,
getTask: () => undefined,
} as unknown as BackgroundManager
const athenaCouncilTool = createAthenaCouncilTool({
backgroundManager: launchManager,
councilConfig: { members: [{ model: "openai/gpt-5.3-codex" }] },
})
// #when
const firstExecution = athenaCouncilTool.execute({ question: "First run" }, mockToolContext)
const secondExecution = await athenaCouncilTool.execute({ question: "Second run" }, mockToolContext)
resolveLaunch?.(createRunningTask("bg-dedup"))
const firstResult = parseLaunchResult(await firstExecution)
// #then
expect(secondExecution).toBe("Council is already running for this session. Wait for the current council execution to complete.")
expect(firstResult.launched).toBe(1)
})
})

View File

@@ -1,14 +1,10 @@
import { tool, type ToolDefinition } from "@opencode-ai/plugin"
import { executeCouncil } from "../../agents/athena/council-orchestrator"
import type { CouncilConfig, CouncilMemberConfig, CouncilMemberResponse } from "../../agents/athena/types"
import type { BackgroundManager, BackgroundTask, BackgroundTaskStatus } from "../../features/background-agent"
import type { CouncilConfig, CouncilMemberConfig } from "../../agents/athena/types"
import type { BackgroundManager } from "../../features/background-agent"
import { ATHENA_COUNCIL_TOOL_DESCRIPTION_TEMPLATE } from "./constants"
import { createCouncilLauncher } from "./council-launcher"
import type { AthenaCouncilToolArgs } from "./types"
const WAIT_INTERVAL_MS = 500
const WAIT_TIMEOUT_MS = 600000
const TERMINAL_STATUSES: Set<BackgroundTaskStatus> = new Set(["completed", "error", "cancelled", "interrupt"])
import type { AthenaCouncilLaunchResult, AthenaCouncilToolArgs } from "./types"
/** Tracks active council executions per session to prevent duplicate launches. */
const activeCouncilSessions = new Set<string>()
@@ -17,90 +13,6 @@ function isCouncilConfigured(councilConfig: CouncilConfig | undefined): councilC
return Boolean(councilConfig && councilConfig.members.length > 0)
}
async function waitForTasksToSettle(
taskIds: string[],
manager: BackgroundManager,
abortSignal: AbortSignal
): Promise<Map<string, BackgroundTask>> {
const settledIds = new Set<string>()
const latestTasks = new Map<string, BackgroundTask>()
const startedAt = Date.now()
while (settledIds.size < taskIds.length && Date.now() - startedAt < WAIT_TIMEOUT_MS) {
if (abortSignal.aborted) {
break
}
for (const taskId of taskIds) {
if (settledIds.has(taskId)) {
continue
}
const task = manager.getTask(taskId)
if (!task) {
continue
}
latestTasks.set(taskId, task)
if (TERMINAL_STATUSES.has(task.status)) {
settledIds.add(taskId)
}
}
if (settledIds.size < taskIds.length) {
await new Promise((resolve) => setTimeout(resolve, WAIT_INTERVAL_MS))
}
}
return latestTasks
}
function mapTaskStatus(status: BackgroundTaskStatus): CouncilMemberResponse["status"] {
if (status === "completed") {
return "completed"
}
if (status === "cancelled" || status === "interrupt") {
return "timeout"
}
return "error"
}
function refreshResponse(response: CouncilMemberResponse, task: BackgroundTask | undefined): CouncilMemberResponse {
if (!task) {
return response
}
const status = mapTaskStatus(task.status)
const durationMs =
task.startedAt && task.completedAt
? Math.max(0, task.completedAt.getTime() - task.startedAt.getTime())
: response.durationMs
return {
...response,
status,
response: status === "completed" ? task.result : undefined,
error: status === "completed" ? undefined : (task.error ?? `Task status: ${task.status}`),
durationMs,
}
}
function formatCouncilOutput(responses: CouncilMemberResponse[], totalMembers: number): string {
const completedCount = responses.filter((item) => item.status === "completed").length
const lines = responses.map((item, index) => {
const model = item.member.name ?? item.member.model
const content = item.status === "completed"
? (item.response ?? "(no response)")
: (item.error ?? "Unknown error")
return `${index + 1}. ${model}\n Status: ${item.status}\n Result: ${content}`
})
return `${completedCount}/${totalMembers} council members completed.\n\n${lines.join("\n\n")}`
}
interface FilterCouncilMembersResult {
members: CouncilMemberConfig[]
error?: string
@@ -201,18 +113,30 @@ export function createAthenaCouncilTool(args: {
parentAgent: toolContext.agent,
})
const taskIds = execution.responses
.map((response) => response.taskId)
.filter((taskId) => taskId.length > 0)
const launchResult: AthenaCouncilLaunchResult = {
launched: execution.responses.filter((response) => response.taskId.length > 0).length,
members: execution.responses
.filter((response) => response.taskId.length > 0)
.map((response) => ({
task_id: response.taskId,
name: response.member.name ?? response.member.model,
model: response.member.model,
status: "running",
})),
failed: execution.responses
.filter((response) => response.taskId.length === 0)
.map((response) => ({
name: response.member.name ?? response.member.model,
model: response.member.model,
error: response.error ?? "Launch failed",
})),
}
const latestTasks = await waitForTasksToSettle(taskIds, backgroundManager, toolContext.abort)
const refreshedResponses = execution.responses.map((response) =>
response.taskId ? refreshResponse(response, latestTasks.get(response.taskId)) : response
)
return formatCouncilOutput(refreshedResponses, execution.totalMembers)
} finally {
activeCouncilSessions.delete(toolContext.sessionID)
return JSON.stringify(launchResult)
} catch (error) {
activeCouncilSessions.delete(toolContext.sessionID)
throw error
}
},
})

View File

@@ -2,3 +2,22 @@ export interface AthenaCouncilToolArgs {
question: string
members?: string[]
}
export interface AthenaCouncilLaunchedMember {
task_id: string
name: string
model: string
status: "running"
}
export interface AthenaCouncilFailedMember {
name: string
model: string
error: string
}
export interface AthenaCouncilLaunchResult {
launched: number
members: AthenaCouncilLaunchedMember[]
failed: AthenaCouncilFailedMember[]
}