What it does

A repeatable, product-oriented forensic procedure for investigating stalled, looping, or over-recovered issue trees. The skill guides the agent to identify the exact stop-point, collect evidence (run ids, timestamps, transitions), survey recent related work, classify non-progressing issues, and frame a general product contract that preserves three invariants: productive work continues, only real blockers stop work, and no infinite loops. The output is a documented plan (phased, assigned, and approval-gated) rather than code changes.

When to use it

Use this skill when an assignment or issue title/body explicitly asks "why did this work stop", "infinite loop", "liveness — what happened here", or when a user links a stalled/looping issue tree and requests forensics or a plan before any code changes. Do NOT use when the task asks to ship a code fix immediately.

What's included

Scripts: none (has_scripts=false)
References: none (has_references=false)
Instructions: step-by-step forensic procedure: read execution semantics doc, walk the tree node-by-node, quote evidence, survey recent work, classify each non-progressing issue, state a contract-rule preserving invariants, write a phased plan (Phase 0 hygiene, docs update, implementation phases), and open a request_confirmation interaction before creating child issues.