cc-haha/src/query/stopHooks.ts
程序员阿江(Relakkes) 71ce980f1c fix: make goal completion durable through stop hooks
The goal command now uses a session Stop hook as the durable completion mechanism and removes the separate evaluator path. Completion output is persisted as local command output so CLI streams, transcripts, and desktop history all see the same completed state.

Desktop keeps active goals in a compact header strip and lets completed goals render in the conversation flow, which avoids the old duplicate pinned card behavior while preserving visibility for running goals.

Constraint: /goal state has to survive transcript replay and desktop session restore.
Rejected: Keep the legacy evaluator alongside Stop hooks | duplicate completion paths made desktop state drift and hid the final card.
Confidence: high
Scope-risk: moderate
Directive: Keep goal completion events on the local_command transcript path unless desktop history restore is changed at the same time.
Tested: bun test src/goals/goalState.test.ts src/query/stopHooks.test.ts src/commands/goal/goal.test.tsx src/commands/headless.test.ts src/server/__tests__/ws-memory-events.test.ts src/server/__tests__/sessions.test.ts
Tested: bun run check:server
Tested: bun run check:desktop
Tested: real /tmp /goal CLI run 98fb4bb2-52da-4850-bc8e-e5e1cc0e4c51 with deepseek-v4-pro and WebUI screenshot verification
Not-tested: Remote release workflow
2026-05-18 22:30:18 +08:00

510 lines
18 KiB
TypeScript

import { feature } from 'bun:bundle'
import { getShortcutDisplay } from '../keybindings/shortcutFormat.js'
import { getSessionId } from '../bootstrap/state.js'
import { isExtractModeActive } from '../memdir/paths.js'
import {
type AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS,
logEvent,
} from '../services/analytics/index.js'
import type { ToolUseContext } from '../Tool.js'
import {
ensureThreadGoalHookFromTranscript,
isGoalPromptHookCommand,
} from '../goals/goalState.js'
import type { HookResult } from '../types/hooks.js'
import type { HookProgress } from '../types/hooks.js'
import type {
AssistantMessage,
Message,
RequestStartEvent,
StopHookInfo,
StreamEvent,
TombstoneMessage,
ToolUseSummaryMessage,
} from '../types/message.js'
import { createAttachmentMessage } from '../utils/attachments.js'
import { logForDebugging } from '../utils/debug.js'
import { errorMessage } from '../utils/errors.js'
import type { REPLHookContext } from '../utils/hooks/postSamplingHooks.js'
import {
executeStopHooks,
executeTaskCompletedHooks,
executeTeammateIdleHooks,
getStopHookMessage,
getTaskCompletedHookMessage,
getTeammateIdleHookMessage,
} from '../utils/hooks.js'
import {
createStopHookSummaryMessage,
createCommandInputMessage,
createSystemMessage,
createUserInterruptionMessage,
createUserMessage,
} from '../utils/messages.js'
import type { SystemPrompt } from '../utils/systemPromptType.js'
import { getTaskListId, listTasks } from '../utils/tasks.js'
import { getAgentName, getTeamName, isTeammate } from '../utils/teammate.js'
/* eslint-disable @typescript-eslint/no-require-imports */
const extractMemoriesModule = feature('EXTRACT_MEMORIES')
? (require('../services/extractMemories/extractMemories.js') as typeof import('../services/extractMemories/extractMemories.js'))
: null
const jobClassifierModule = feature('TEMPLATES')
? (require('../jobs/classifier.js') as typeof import('../jobs/classifier.js'))
: null
/* eslint-enable @typescript-eslint/no-require-imports */
import type { QuerySource } from '../constants/querySource.js'
import { executeAutoDream } from '../services/autoDream/autoDream.js'
import { executePromptSuggestion } from '../services/PromptSuggestion/promptSuggestion.js'
import { isBareMode, isEnvDefinedFalsy } from '../utils/envUtils.js'
import {
createCacheSafeParams,
saveCacheSafeParams,
} from '../utils/forkedAgent.js'
type StopHookResult = {
blockingErrors: Message[]
preventContinuation: boolean
}
export function shouldLetGoalPromptHookContinue(
result: Pick<HookResult, 'blockingError' | 'preventContinuation'>,
): boolean {
return Boolean(
result.preventContinuation &&
isGoalPromptHookCommand(result.blockingError?.command),
)
}
export async function* handleStopHooks(
messagesForQuery: Message[],
assistantMessages: AssistantMessage[],
systemPrompt: SystemPrompt,
userContext: { [k: string]: string },
systemContext: { [k: string]: string },
toolUseContext: ToolUseContext,
querySource: QuerySource,
stopHookActive?: boolean,
): AsyncGenerator<
| StreamEvent
| RequestStartEvent
| Message
| TombstoneMessage
| ToolUseSummaryMessage,
StopHookResult
> {
const hookStartTime = Date.now()
const stopHookContext: REPLHookContext = {
messages: [...messagesForQuery, ...assistantMessages],
systemPrompt,
userContext,
systemContext,
toolUseContext,
querySource,
}
// Only save params for main session queries — subagents must not overwrite.
// Outside the prompt-suggestion gate: the REPL /btw command and the
// side_question SDK control_request both read this snapshot, and neither
// depends on prompt suggestions being enabled.
if (querySource === 'repl_main_thread' || querySource === 'sdk') {
saveCacheSafeParams(createCacheSafeParams(stopHookContext))
}
// Template job classification: when running as a dispatched job, classify
// state after each turn. Gate on repl_main_thread so background forks
// (extract-memories, auto-dream) don't pollute the timeline with their own
// assistant messages. Await the classifier so state.json is written before
// the turn returns — otherwise `claude list` shows stale state for the gap.
// Env key hardcoded (vs importing JOB_ENV_KEY from jobs/state) to match the
// require()-gated jobs/ import pattern above; spawn.test.ts asserts the
// string matches.
if (
feature('TEMPLATES') &&
process.env.CLAUDE_JOB_DIR &&
querySource.startsWith('repl_main_thread') &&
!toolUseContext.agentId
) {
// Full turn history — assistantMessages resets each queryLoop iteration,
// so tool calls from earlier iterations (Agent spawn, then summary) need
// messagesForQuery to be visible in the tool-call summary.
const turnAssistantMessages = stopHookContext.messages.filter(
(m): m is AssistantMessage => m.type === 'assistant',
)
const p = jobClassifierModule!
.classifyAndWriteState(process.env.CLAUDE_JOB_DIR, turnAssistantMessages)
.catch(err => {
logForDebugging(`[job] classifier error: ${errorMessage(err)}`, {
level: 'error',
})
})
await Promise.race([
p,
// eslint-disable-next-line no-restricted-syntax -- sleep() has no .unref(); timer must not block exit
new Promise<void>(r => setTimeout(r, 60_000).unref()),
])
}
// --bare / SIMPLE: skip background bookkeeping (prompt suggestion,
// memory extraction, auto-dream). Scripted -p calls don't want auto-memory
// or forked agents contending for resources during shutdown.
if (!isBareMode()) {
// Inline env check for dead code elimination in external builds
if (!isEnvDefinedFalsy(process.env.CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION)) {
void executePromptSuggestion(stopHookContext)
}
if (
feature('EXTRACT_MEMORIES') &&
!toolUseContext.agentId &&
isExtractModeActive()
) {
// Fire-and-forget in both interactive and non-interactive. For -p/SDK,
// print.ts drains the in-flight promise after flushing the response
// but before gracefulShutdownSync (see drainPendingExtraction).
void extractMemoriesModule!.executeExtractMemories(
stopHookContext,
toolUseContext.appendSystemMessage,
)
}
if (!toolUseContext.agentId) {
void executeAutoDream(stopHookContext, toolUseContext.appendSystemMessage)
}
}
// chicago MCP: auto-unhide + lock release at turn end.
// Main thread only — the CU lock is a process-wide module-level variable,
// so a subagent's stopHooks releasing it leaves the main thread's cleanup
// seeing isLockHeldLocally()===false → no exit notification, and unhides
// mid-turn. Subagents don't start CU sessions so this is a pure skip.
if (!toolUseContext.agentId) {
try {
const { cleanupComputerUseAfterTurn } = await import(
'../utils/computerUse/cleanup.js'
)
await cleanupComputerUseAfterTurn(toolUseContext)
} catch {
// Failures are silent — this is dogfooding cleanup, not critical path
}
}
try {
const blockingErrors = []
const appState = toolUseContext.getAppState()
const permissionMode = appState.toolPermissionContext.mode
if (!toolUseContext.agentId) {
ensureThreadGoalHookFromTranscript(
toolUseContext,
getSessionId(),
[...messagesForQuery, ...assistantMessages],
)
}
const generator = executeStopHooks(
permissionMode,
toolUseContext.abortController.signal,
undefined,
stopHookActive ?? false,
toolUseContext.agentId,
toolUseContext,
[...messagesForQuery, ...assistantMessages],
toolUseContext.agentType,
)
// Consume all progress messages and get blocking errors
let stopHookToolUseID = ''
let hookCount = 0
let preventedContinuation = false
let stopReason = ''
let hasOutput = false
const hookErrors: string[] = []
const hookInfos: StopHookInfo[] = []
let goalCompleted = false
for await (const result of generator) {
if (result.message) {
yield result.message
// Track toolUseID from progress messages and count hooks
if (result.message.type === 'progress' && result.message.toolUseID) {
stopHookToolUseID = result.message.toolUseID
hookCount++
// Extract hook command and prompt text from progress data
const progressData = result.message.data as HookProgress
if (progressData.command) {
hookInfos.push({
command: progressData.command,
promptText: progressData.promptText,
})
}
}
// Track errors and output from attachments
if (result.message.type === 'attachment') {
const attachment = result.message.attachment
if (
'hookEvent' in attachment &&
(attachment.hookEvent === 'Stop' ||
attachment.hookEvent === 'SubagentStop')
) {
if (attachment.type === 'hook_non_blocking_error') {
hookErrors.push(
attachment.stderr || `Exit code ${attachment.exitCode}`,
)
// Non-blocking errors always have output
hasOutput = true
} else if (attachment.type === 'hook_error_during_execution') {
hookErrors.push(attachment.content)
hasOutput = true
} else if (attachment.type === 'hook_success') {
if (isGoalPromptHookCommand(attachment.command)) {
goalCompleted = true
}
// Check if successful hook produced any stdout/stderr
if (
(attachment.stdout && attachment.stdout.trim()) ||
(attachment.stderr && attachment.stderr.trim())
) {
hasOutput = true
}
}
// Extract per-hook duration for timing visibility.
// Hooks run in parallel; match by command + first unassigned entry.
if ('durationMs' in attachment && 'command' in attachment) {
const info = hookInfos.find(
i =>
i.command === attachment.command &&
i.durationMs === undefined,
)
if (info) {
info.durationMs = attachment.durationMs
}
}
}
}
}
if (result.blockingError) {
const userMessage = createUserMessage({
content: getStopHookMessage(result.blockingError),
isMeta: true, // Hide from UI (shown in summary message instead)
})
blockingErrors.push(userMessage)
yield userMessage
hasOutput = true
// Add to hookErrors so it appears in the summary
hookErrors.push(result.blockingError.blockingError)
}
// Check if hook wants to prevent continuation
if (result.preventContinuation) {
if (!shouldLetGoalPromptHookContinue(result)) {
preventedContinuation = true
stopReason = result.stopReason || 'Stop hook prevented continuation'
// Create attachment to track the stopped continuation (for structured data)
yield createAttachmentMessage({
type: 'hook_stopped_continuation',
message: stopReason,
hookName: 'Stop',
toolUseID: stopHookToolUseID,
hookEvent: 'Stop',
})
}
}
// Check if we were aborted during hook execution
if (toolUseContext.abortController.signal.aborted) {
logEvent('tengu_pre_stop_hooks_cancelled', {
queryChainId: toolUseContext.queryTracking
?.chainId as AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS,
queryDepth: toolUseContext.queryTracking?.depth,
})
yield createUserInterruptionMessage({
toolUse: false,
})
return { blockingErrors: [], preventContinuation: true }
}
}
// Create summary system message if hooks ran
if (hookCount > 0) {
yield createStopHookSummaryMessage(
hookCount,
hookInfos,
hookErrors,
preventedContinuation,
stopReason,
hasOutput,
'suggestion',
stopHookToolUseID,
)
// Send notification about errors (shown in verbose/transcript mode via ctrl+o)
if (hookErrors.length > 0) {
const expandShortcut = getShortcutDisplay(
'app:toggleTranscript',
'Global',
'ctrl+o',
)
toolUseContext.addNotification?.({
key: 'stop-hook-error',
text: `Stop hook error occurred \u00b7 ${expandShortcut} to see`,
priority: 'immediate',
})
}
}
if (goalCompleted) {
yield createCommandInputMessage(
'<local-command-stdout>Goal marked complete.</local-command-stdout>',
)
}
if (preventedContinuation) {
return { blockingErrors: [], preventContinuation: true }
}
// Collect blocking errors from stop hooks
if (blockingErrors.length > 0) {
return { blockingErrors, preventContinuation: false }
}
// After Stop hooks pass, run TeammateIdle and TaskCompleted hooks if this is a teammate
if (isTeammate()) {
const teammateName = getAgentName() ?? ''
const teamName = getTeamName() ?? ''
const teammateBlockingErrors: Message[] = []
let teammatePreventedContinuation = false
let teammateStopReason: string | undefined
// Each hook executor generates its own toolUseID — capture from progress
// messages (same pattern as stopHookToolUseID at L142), not the Stop ID.
let teammateHookToolUseID = ''
// Run TaskCompleted hooks for any in-progress tasks owned by this teammate
const taskListId = getTaskListId()
const tasks = await listTasks(taskListId)
const inProgressTasks = tasks.filter(
t => t.status === 'in_progress' && t.owner === teammateName,
)
for (const task of inProgressTasks) {
const taskCompletedGenerator = executeTaskCompletedHooks(
task.id,
task.subject,
task.description,
teammateName,
teamName,
permissionMode,
toolUseContext.abortController.signal,
undefined,
toolUseContext,
)
for await (const result of taskCompletedGenerator) {
if (result.message) {
if (
result.message.type === 'progress' &&
result.message.toolUseID
) {
teammateHookToolUseID = result.message.toolUseID
}
yield result.message
}
if (result.blockingError) {
const userMessage = createUserMessage({
content: getTaskCompletedHookMessage(result.blockingError),
isMeta: true,
})
teammateBlockingErrors.push(userMessage)
yield userMessage
}
// Match Stop hook behavior: allow preventContinuation/stopReason
if (result.preventContinuation) {
teammatePreventedContinuation = true
teammateStopReason =
result.stopReason || 'TaskCompleted hook prevented continuation'
yield createAttachmentMessage({
type: 'hook_stopped_continuation',
message: teammateStopReason,
hookName: 'TaskCompleted',
toolUseID: teammateHookToolUseID,
hookEvent: 'TaskCompleted',
})
}
if (toolUseContext.abortController.signal.aborted) {
return { blockingErrors: [], preventContinuation: true }
}
}
}
// Run TeammateIdle hooks
const teammateIdleGenerator = executeTeammateIdleHooks(
teammateName,
teamName,
permissionMode,
toolUseContext.abortController.signal,
)
for await (const result of teammateIdleGenerator) {
if (result.message) {
if (result.message.type === 'progress' && result.message.toolUseID) {
teammateHookToolUseID = result.message.toolUseID
}
yield result.message
}
if (result.blockingError) {
const userMessage = createUserMessage({
content: getTeammateIdleHookMessage(result.blockingError),
isMeta: true,
})
teammateBlockingErrors.push(userMessage)
yield userMessage
}
// Match Stop hook behavior: allow preventContinuation/stopReason
if (result.preventContinuation) {
teammatePreventedContinuation = true
teammateStopReason =
result.stopReason || 'TeammateIdle hook prevented continuation'
yield createAttachmentMessage({
type: 'hook_stopped_continuation',
message: teammateStopReason,
hookName: 'TeammateIdle',
toolUseID: teammateHookToolUseID,
hookEvent: 'TeammateIdle',
})
}
if (toolUseContext.abortController.signal.aborted) {
return { blockingErrors: [], preventContinuation: true }
}
}
if (teammatePreventedContinuation) {
return { blockingErrors: [], preventContinuation: true }
}
if (teammateBlockingErrors.length > 0) {
return {
blockingErrors: teammateBlockingErrors,
preventContinuation: false,
}
}
}
return { blockingErrors: [], preventContinuation: false }
} catch (error) {
const durationMs = Date.now() - hookStartTime
logEvent('tengu_stop_hook_error', {
duration: durationMs,
queryChainId: toolUseContext.queryTracking
?.chainId as AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS,
queryDepth: toolUseContext.queryTracking?.depth,
})
// Yield a system message that is not visible to the model for the user
// to debug their hook.
yield createSystemMessage(
`Stop hook failed: ${errorMessage(error)}`,
'warning',
)
return { blockingErrors: [], preventContinuation: false }
}
}