What problem does it solve?
Long-running or silent background tasks (such as probes, deployment scripts, fetch loops, browser automation, or dispatched sub-agents) can hang without emitting any completion event, leaving the AI idle indefinitely with no signal that work has stalled. This invisible failure mode leads to wasted time, missed deadlines, and unhandled errors that only surface when a human notices the task is stuck.
Core Features & Use Cases
- Liveness guarantee enforcement: Requires all long/silent work to be launched with at least one liveness safeguard: in-turn completion with a hard timeout, a bounded wrapper with incremental progress logging, or an explicit BLOCKED-AT return for unbounded waits.
- Hang diagnosis and recovery: Provides a clear protocol to distinguish slow-but-progressing work from actual hangs by checking activity (log lines, file mtime, byte growth) instead of relying on 0-byte file existence, plus step-by-step recovery steps to stop stalled work, re-run with safeguards, and clean up residue.
- Zero-trust sub-agent rules: Mandates proactive polling of sub-agent liveness via git progress or process activity (not completion pings) and verification of final output before trusting results, eliminating risks from missing or false sub-agent completion signals.
- Use case example: If you are dispatching a sub-agent to run a multi-step code review or launching a long deployment pipeline, this skill ensures you will detect stalls immediately and recover instead of waiting indefinitely for a completion event that never comes.
Quick Start
Apply the liveness-discipline skill to attach a bounded timeout and incremental progress logging to any long-running background task before you launch it.
