skill-audit 技能树体检

You can detect dead Claude Code skills by computing real usage rankings from local execution logs. This process identifies archive candidates by flagging skills with zero triggers over 30 days combined with missing hook bindings and CLAUDE.md references.

Finding overlapping trigger conditions requires scanning skill descriptions for common phrases to diagnose conflicts. This approach detects likely competing trigger word groups across your skill tree, highlighting overlapping logic that may cause unintended execution paths.

Rules that never fire indicate a rule-execution gap where iron-rule skills are referenced in text but show zero actual executions. Auditing local execution logs identifies these gaps, turning written rules into measurable operational evidence.

Yes, tracking Claude Code skill invocations requires a PreToolUse hook configured to write invocation data to a local log file. The audit process reads this execution log to compute zero-invocation indicators and generate accurate usage rankings.

Preparing safe archive decisions involves inventorying skills and computing real usage rankings with a human review loop. By combining zero-invocation signals, missing hook bindings, and conflict diagnostics, you can safely identify unused candidates for archiving.

Auditing a Claude Code skill tree requires reading local execution logs to compute real usage rankings and zero-invocation indicators. Without local logging enabled via a PreToolUse hook, the audit cannot detect dead skills or measure rule-execution gaps accurately.