fleet-pause

Pause work and escalate when requirements are unclear, changes are high-risk, you're stuck, or you need human/agent input. Properly notifies via task comment, IRC, and board memory. Prevents agents fr

SKILL.md

GitHub

Fleet Pause / Escalate

Know when to STOP working and ASK for help. Continuing to guess when you should stop wastes time and produces bad output.

When to Pause

1. Requirements are unclear

You can't determine what to do from the task description. Don't guess — ask.

2. High-risk change

The change affects security, data integrity, infrastructure, or auth. Don't proceed without human approval.

3. Multiple valid approaches

There are 2+ reasonable ways to do this and the choice matters. Let the human or architect decide.

4. Blocked by another agent

You need output from another agent's task and it's not done. Create a follow-up or wait.

5. Stuck

You've been working for a while without progress. Something is wrong — escalate.

6. Found something unexpected

The codebase is different from what was described. Requirements contradict what you see. A dependency is broken or missing.

How to Pause

Step 1: Stop working

Do not make any more changes. If you have uncommitted work, commit what's safe with a chore: WIP — paused for escalation [task:XXXXXXXX] commit.

Step 2: Post blocker comment on task

Use agents/_template/markdown/comment-blocker.md:

curl -s -X POST "$BASE_URL/api/v1/agent/boards/$BOARD_ID/tasks/$TASK_ID/comments" \
  -H "X-Agent-Token: $AUTH_TOKEN" -H "Content-Type: application/json" \
  -d '{"message": "## 🚫 Blocked\n\n**Reason:** {specific reason}\n**Impact:** {what cannot proceed}\n**Needed:** {exactly what would unblock}\n**Suggested action:** {who should do what}\n\n---\n<sub>{agent_name} · [{task_short}]({task_url})</sub>"}'

Step 3: Post to IRC

bash scripts/notify-irc.sh --agent "{agent_name}" --event "BLOCKED" --title "{reason}" --url "{task_url}"

Step 4: Post to board memory (if it affects other agents)

Only if other agents need to know about this blocker:

curl -s -X POST "$BASE_URL/api/v1/agent/boards/$BOARD_ID/memory" \
  -H "X-Agent-Token: $AUTH_TOKEN" -H "Content-Type: application/json" \
  -d '{"content": "## 🚫 Blocker: {title}\n\n**Agent:** {name}\n**Task:** [{title}]({task_url})\n**Impact:** {what this blocks}\n**Needed:** {resolution}\n\n---", "tags": ["blocked", "project:{project}"], "source": "{agent_name}"}'

Step 5: WAIT

Do not continue working on this task. Wait for human input via:

  • IRC response
  • Board memory response
  • Task comment
  • New instructions via chat.send

Decision Tree

Am I stuck?
  ├─ Yes → Post blocker, notify IRC, WAIT
  └─ No
      ├─ Is this high-risk? → Post for review, notify IRC, WAIT
      ├─ Are requirements unclear? → Post question, notify IRC, WAIT
      ├─ Do I need another agent? → Create proposed task (fleet-task-create), notify
      └─ Multiple approaches? → Post options to board memory, tag @human, WAIT

Rules

  • NEVER continue guessing when you should stop
  • ALWAYS explain WHY you're pausing (not just "I'm stuck")
  • ALWAYS say what would UNBLOCK you (specific action needed)
  • ALWAYS notify IRC so the human sees it immediately
  • Pausing is NOT failure — it's intelligence. Bad output from guessing IS failure.