What problem does it solve? After installing or modifying the Mycelium plugin, manually invoking a hook script cannot prove that Claude Code or Codex actually discovered, trusted, and dispatched the hook. This Skill runs a genuine end-to-end lifecycle audit through a real host CLI task, so you can verify SessionStart injection, PostToolUse tracking, Stop enforcement, and handoff generation behave correctly in production-like conditions. ## Core Features & Use Cases - Black-box host verification: Launches a fresh Claude Code or Codex task with a canonical child-task prompt and observes only automatic lifecycle behavior, never invoking hooks manually. - Positive and negative probes: Exercises a harmless analysis command, an argv-rewrite near-neighbor that must stay silent, edit tracking with a disposable file, and the expected Stop-blocking retry with a five-section handoff. - Failure classification: Sorts every unexpected result into artifact identity, host dispatch, hook logic, environment, agent compliance, or inconclusive categories using a structured evidence table. - Use Case: After merging cross-host lifecycle changes, run this audit against a disposable initialized repository to confirm the installed plugin artifact matches the source commit and that nested-subagent isolation, lineage recording, and handoff structure all pass before release. ## Quick Start Use the lifecycle-audit skill to run a black-box Mycelium lifecycle smoke test against a fresh disposable repository in Claude Code and classify any failures.
