Skill Audit
Audit the requested installed skills or plugins. Resolve editable owners; installed caches, memory, and session archives are evidence, not edit targets. For a portfolio request, include the requested scope, including this skill when it falls within that scope.
An audit alone is read-only. If the user requests fixes, complete the findings and proceed under that authorization using the owning maintenance workflow; use skill-creator for substantial skill reshapes. Never steer or mutate a monitored task without separate authorization.
Evidence routes
Read states to classify target and evidence. Resolve unclear cache paths with cache resolution. Then select the target overlay: standalone skills, plugins, or bundled skills.
- Static wording, triggers, or ownership: inspect target instructions and relevant consumers directly. No history helper is needed.
- Prior runtime behavior: historical evidence.
- Current task monitoring: live monitoring. Do not substitute archives for unavailable live evidence.
- Portfolio overlap or cost: portfolio hygiene.
- Instruction density: writing review.
Use scripts/session-evidence only for invocation or behavioral history;
scripts/portfolio-health only for inventory, budget, or usage questions.
Neither proves a runtime defect without representative evidence.
Return the applicable output format, scaled to the findings. Distinguish static concerns from observed failures and report missing evidence. Recommend new surfaces only when existing owners cannot serve the need.