fix: query 超时保护 + 进程清理，防止 PM2 重启后残留孤儿进程

[lishuceo]

Summary

• executor 接入 CLAUDE_TIMEOUT 配置，通过 abortController 强制终止超时 query（默认 300s）
• 启动时扫描并 kill 残留的 claude 孤儿子进程（PM2 SIGKILL / 服务崩溃场景）
• shutdown handler 加 3s 延迟等待子进程退出 + 重复信号去重
• 新增 PM2 ecosystem 配置，kill_timeout 从默认 1600ms 提升到 10s

Test plan

• 200 个现有测试全部通过
• 验证 PM2 restart 后无残留 claude 进程
• 验证 query 超过 CLAUDE_TIMEOUT 后被正确终止

🤖 Generated with Claude Code

[claude]

Claude finished @lishuceo's task in 1m 33s —— View job

---

PR Review: Timeout Protection + Process Cleanup

• Check for previous review comments
• Read project conventions (CLAUDE.md)
• Review PR diff and all changed files
• Post inline comments for issues found
• Post summary verdict

---

⚠️ Issues Found

Overall assessment: The PR is well-structured and addresses a real operational pain point (orphan processes after PM2 restarts). The timeout implementation in executor.ts is clean — proper AbortController usage, clearTimeout on both success and error paths, and correct timeout-vs-error differentiation. The shutdown dedup and PM2 ecosystem config are solid additions.

Two moderate issues found:

• 🟡 Process cleanup too broad (confidence: 82) — killOrphanedClaudeProcesses() kills any process with basename claude, without verifying it's actually orphaned (PPID=1). On a shared server, this could kill legitimate claude processes belonging to other users or services. See inline comment with suggested fix.

• 🟡 Interval not cleared on shutdown (confidence: 76) — The 30-minute cleanup interval can fire during the 3-second shutdown grace period, calling sessionManager.cleanup() / pipelineStore.cleanExpired() on already-closed resources. See inline comment.

What looks good:

• Timeout implementation with AbortController + clearTimeout in both paths is correct
• Shutdown signal deduplication via shuttingDown flag
• PM2 kill_timeout: 10000 aligns well with the 3s shutdown delay
• Test mock updated to include timeoutSeconds
• ESM patterns and .js import extensions followed consistently

[claude]

Claude finished @lishuceo's task in 3m 11s —— View job

---

PR Re-review: Timeout Protection + Process Cleanup

• Check previous review comments against current code
• Resolve fixed issues from previous review
• Read all changed files in full
• Review current diff for new issues
• Post summary verdict

---

✅ Approved — All Previous Issues Fixed, No New Issues Found

2 previous issues fixed:

1. Process cleanup now checks PPID=1 — killOrphanedClaudeProcesses() at src/utils/process-cleanup.ts:37-43 now verifies the process is a genuine orphan (reparented to init) before killing it. Threads resolved.
2. Interval cleared on shutdown — cleanupInterval is stored (src/index.ts:42) and cleared via clearInterval(cleanupInterval) at the start of shutdown() (src/index.ts:57). Thread resolved.

Review of current state:

• Timeout implementation (src/claude/executor.ts:130-137) — Clean AbortController + setTimeout pattern. clearTimeout(timer) correctly called in both the catch block (line 283) and the success path (line 301). The timedOut flag properly differentiates timeout errors from other failures.
• Process cleanup (src/utils/process-cleanup.ts) — Safe: pid is parseInt-parsed (line 25) before interpolation into execSync (line 39), eliminating command injection risk. Triple filtering (basename check → PPID=1 check → skip self) is thorough.
• Shutdown deduplication (src/index.ts:50-68) — shuttingDown flag prevents duplicate signal handling. clearInterval + claudeExecutor.killAll() + 3s grace period is well-structured. PM2 kill_timeout: 10000 in ecosystem.config.cjs gives adequate headroom.
• Test mock (src/claude/__tests__/executor.test.ts:15) — Updated to include timeoutSeconds: 300, matching the config shape.
• ESM patterns — .js import extensions followed consistently throughout.

No new issues found. The PR is clean and ready to merge.