宜博 / lemonboss

  • fix(graph-engine): 卡在 running 且背后进程已死的节点持续性存活扫描（45 号文档 Part A-C） ...
    a4b8b87b 
    真实事故：h3video 公司 h3-story-draft 节点卡在 status=running，背后 CLI 会话已死但服务器
进程没重启，开机一次性孤儿扫描/goal-task 巡检/wallclock 看门狗均看不到，ceo 被转发问题后
绕过引擎自己拉会话，陷入死循环。

- sweepStaleRunningAgentCalls + startStaleAgentCallWatchdog：运行时长超阈值（默认 25 分钟）
  的 agent_call 节点判 infra_flake 原地重跑，5 分钟一次巡检，挂进 server/index.js
- ceo/tools.md 补 force-fail 端点（已有，未改代码）的使用文档 + 禁止自拉会话的规则
- ceo/heartbeat.md 同步更新"不用管的"一节

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WjFL1nfPDqZnJeuskYHwvs
    yibo authored
     
    Browse File »
  • fix(graph-engine): commit T-mtgxcnfz fix — stopRun orphan-run race + /next auto-mode guard ...
    364788ae 
    Snapshot-then-stop race in stopRun() left child runs created after the
snapshot outside the stop scope; add auto-mode guard rejecting manual
/next on execution_mode='auto' runs (source of the concurrent-advance race).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
    yibo authored
     
    Browse File »