메뉴
Diagnose and fix AO lifecycle-worker backfill failures (stale worktrees, branch conflicts, claim_failed loops)
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/jleechanorg/claude-commands --skill diagnose-lifecycle-worker명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | diagnose-lifecycle-worker |
| description | Diagnose and fix AO lifecycle-worker backfill failures (stale worktrees, branch conflicts, claim_failed loops) |
Use this when lifecycle.backfill.claim_failed errors appear in lifecycle-worker logs, or when open PRs have no active sessions and are stalling.
pgrep -af "lifecycle-worker"
If nothing: lifecycle-worker is not running. Check launchd state:
launchctl print gui/$(id -u)/ai.agento.lifecycle-all 2>&1 | grep "state ="
If state ≠ running, bootstrap it:
launchctl bootstrap gui/$(id -u) ~/Library/LaunchAgents/ai.agento.lifecycle-all.plist
# Check common log locations
ls ~/.openclaw/logs/ao-lifecycle-*.log 2>/dev/null | tail -5
ls ~/.agent-orchestrator/*/logs/lifecycle-*.log 2>/dev/null | tail -5
Tail the most recent log:
tail -50 ~/.openclaw/logs/ao-lifecycle-<project>.log
grep -n "claim_failed\|refusing to fetch\|Session not found" ~/.openclaw/logs/ao-lifecycle-<project>. | -20
Extract the error message and branch/worktree path from the output.
This means a worktree has a branch checked out that the lifecycle-worker is trying to fetch into.
Identify the worktree and branch:
# From the error, note the path and branch name
# e.g.: fatal: refusing to fetch into branch 'refs/heads/feat/foo' checked out at '/path/to/worktree'
Check if the worktree's tmux session is alive:
# Extract session name from the worktree path (e.g. ao-123 from /path/to/agent-orchestrator/ao-123)
tmux has-session -t <session-name> 2>/dev/null && echo "ALIVE" || echo "DEAD"
If dead — remove the ghost worktree:
# Derive the git repo dir from the worktree path (parent of .git worktrees list entry)
WORKTREE="/path/to/worktree"
GIT_DIR="$(dirname "$WORKTREE")" # e.g. ~/.worktrees/jleechanclaw
# Remove the ghost worktree (operate from the parent repo, not inside the worktree)
git -C "$GIT_DIR" worktree remove --force "$WORKTREE"
git -C "$GIT_DIR" branch -d <branch-name> # safe if merged upstream
git -C "$GIT_DIR" branch -D <branch-name> # force if not yet merged
Verify the fix works:
GIT_DIR="$(dirname "/path/to/worktree")"
git -C "$GIT_DIR" fetch --force origin +refs/pull/<PR_NUMBER>/head:<branch-name>
The lifecycle-worker lost track of a session.
Identify the repo from the log context (look for project: in the lifecycle log):
grep -n "project:" ~/.openclaw/logs/ao-lifecycle-<project>.log | tail -5
Find orphaned session metadata:
ls ~/.agent-orchestrator/*/sessions/archive/ 2>/dev/null | head -20
ls ~/.agent-orchestrator/*/sessions/ 2>/dev/null | head -20
Find orphaned worktrees — always scope to the correct worktreeDir:
# Determine worktreeDir for the project from agent-orchestrator.yaml
python3 -c "
import yaml
cfg = yaml.safe_load(open('$HOME/.openclaw/agent-orchestrator.yaml'))
proj = cfg['projects']['<project-name>']
print('worktreeDir:', proj.get('worktreeDir', '~/.worktrees/' + proj.get('name','').lower().replace(' ', '-')))
print('repo:', proj.get('repo'))
"
WORKTREE_DIR="~/.worktrees/jleechanclaw-main" # substitute from above
git -C "$WORKTREE_DIR" worktree list
Find orphaned worktrees:
git worktree list | grep <project-name>
If you find a worktree with no corresponding live tmux session, it's orphaned — remove it per Path A.
If the error path points to the main repo (not a worktree):
git -C /path/to/repo branch --show-current
If it's not main:
git -C /path/to/repo checkout main && git -C /path/to/repo pull --ff-only
# Watch the log for ~60 seconds after cleanup
tail -f ~/.openclaw/logs/ao-lifecycle-<project>.log
# Press Ctrl+C when done
Look for new entries — the lifecycle-worker should resume processing within a few minutes.
If the same PR has failed 3+ times after cleanup, something systemic is wrong.
Send MCP mail alert:
Use the MCP mail tool (or mcp__mcp-agent-mail__send_message):
project_key: "jleechanclaw" (or relevant project)sender_name: "claude"subject: "lw-stall: PR #"body_md: "3+ consecutive claim_failed for PR # on after cleanup attempts. Manual intervention required. Last error: <error snippet>"Alternatively via curl:
# Requires: OPENCLAW_SLACK_BOT_TOKEN (from ~/.bashrc) and JLEECHAN_DM_CHANNEL (your DM channel ID)
# Bot token posts as openclaw bot; user token ($SLACK_USER_TOKEN from ~/.profile) posts as $USER
OPENCLAW_SLACK_BOT_TOKEN="${OPENCLAW_SLACK_BOT_TOKEN:-}" # set in ~/.bashrc
JLEECHAN_DM_CHANNEL="${JLEECHAN_DM_CHANNEL:-}" # set in ~/.bashrc
curl -s -X POST "https://slack.com/api/chat.postMessage" \
-H "Authorization: Bearer ${OPENCLAW_SLACK_BOT_TOKEN:-$SLACK_USER_TOKEN}" \
-H "Content-Type: application/json" \
-d "{\"channel\": \"${JLEECHAN_DM_CHANNEL:-C0AKALZ4CKW}\", \"text\": \"*[AO Alert]* lifecycle-worker stalled on <project> PR #<N>. 3+ consecutive failures. Manual intervention required.\"}"
| Error substring | Likely cause | Fix |
|---|---|---|
refusing to fetch into branch + worktree path | Ghost worktree | git worktree remove --force <path> |
refusing to fetch + main repo path | Main repo on wrong branch | git checkout main && git pull |
Session not found | Orphaned session metadata | Find+remove orphaned worktrees |
already exists on worktree add | Duplicate worktree | git worktree remove --force <path> then retry |
| Rate limit | gh API exhausted | Wait ~1hr; check gh api rate_limit |