job-guardian

What problem does it solve?

This Skill prevents long-running jobs from silently failing while you’re away, by enforcing evidence-based liveness and progress checks plus safe, reversible recovery.

Core Features & Use Cases

• Launch contract + crash-gating: Forces a written job contract before launch and confirms real health/progress before committing to long polling.
• Stability polling with evidence: Detects hangs by verifying progress changes (e.g., log growth, step/throughput movement) rather than assuming “no error” means “running fine.”
• Recoverable vs. unrecoverable triage: Applies only predefined recovery playbooks with attempt caps, otherwise stops, tears down safely, and notifies.

Quick Start

Give the AI a launch-and-watch request for a batch job, and it will write a guarding contract, start the job, poll every 270s for progress, and either recover or stop-and-notify with the evidence log.