计划执行与检查节点
executing-plans
按既定实施计划完成工作,保持过程检查和反馈。
agent-status-pings【效率与协作】Emit standardized progress pings during long-running work so the coordinator and the user know whether you are alive, stuck, or done — never go dark.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Long-running agent work (research, refactors, multi-step planning) MUST surface a heartbeat. Silence is the worst possible signal in a multi-agent system because the coordinator cannot tell silence from a stalled process.
parallel-agent-coordination, swarm-coordination, or crew-manager-pattern.crew-planner-executor once you cross the first executor step.Use a short, structured message (Slack thread, planning task comment, or assistant message) with three fields:
status: <starting|working|blocked|waiting-for-human|done|failed>
progress: <one short sentence — what you finished, what you're doing next>
next_eta: <ISO timestamp or "<N> minutes">
Examples:
status: working
progress: parsed 3 of 7 repos; next is monorepo X
next_eta: 4 minutes
status: blocked
progress: terraform plan errors on missing IAM role poly-deploy
next_eta: human input needed; paused pending Slack reply
status: starting).status: working ping every 3–5 minutes, OR after every meaningful sub-step.status: done (with a brief result summary) or status: failed (with the failure reason and what you tried).blocked ping must include WHAT would unblock you — not just that you are stuck.planning_list_tasks → comment write.blocked ping that does not say what unblocks it.