Stuck — loop, then ping me

The overnight failure mode is silent thrashing: trying the same fix over and over. This skill is the circuit breaker.

Trigger

You hit the same error, or make no real progress, after 3 genuine attempts on one problem. (Genuine = a different hypothesis each time, not the same edit retried.)

Steps

1. Stop. Do not attempt a 4th variation of the same fix.
2. Write a STUCK report — concise and specific:
   • The exact error / failing check (paste the key lines).
   • The 2-3 approaches you already tried and why each failed.
   • Your current best hypothesis and what you'd need to verify it.
   • The specific decision or input you need from the operator.
3. Ping the operator via the messaging tool (Telegram is the overnight channel). Send the STUCK report as a single clear message ending in a concrete question.
4. Park the task at a clean checkpoint and, if other independent tasks exist, move on rather than burning cycles.

Before pinging, consider escalation

If the blocker is a genuinely-hard but well-scoped subproblem (not a missing decision), the workflow-escalate skill may resolve it without human input — but only if escalation is enabled. A missing decision always goes to the human; only hard reasoning is a candidate for escalation.

Why

This is the operator's explicit "loop then ping me" contract. A specific blocker with attempted approaches is actionable; silent grinding wastes the unattended hours that are this system's main advantage.