| name | Dev10x:skill-audit |
| description | Audit a session's skill usage, compliance, and extract lessons learned. Default (lightweight): analyzes visible context inline and presents a disposition gate — no separate terminal needed for most audits. Forensic (--full or escalation): dispatches parallel subagents for deep transcript analysis — run from a separate terminal. TRIGGER when: session is complete and user wants usage review, or a skill didn't behave as expected. DO NOT TRIGGER when: mid-session during active work, or user is asking about a specific skill's documentation.
|
| user-invocable | true |
| invocation-name | Dev10x:skill-audit |
| allowed-tools | ["Agent","AskUserQuestion","Read(~/.claude/**)","Read(~/.claude/skills/**)","Bash(/tmp/Dev10x/bin/mktmp.sh:*)","Read(/tmp/Dev10x/skill-audit/**)","Edit(~/.claude/**)","Edit(/tmp/Dev10x/skill-audit/**)","Bash(${CLAUDE_PLUGIN_ROOT}/skills/skill-audit/scripts/:*)","mcp__plugin_Dev10x_cli__audit_extract_session","mcp__plugin_Dev10x_cli__audit_analyze_actions","mcp__plugin_Dev10x_cli__audit_analyze_permissions","Bash(ls -t ~/.claude/:*)","Bash(wc:*)","Bash(git config --list:*)","Bash(ls ~/.config/fish/functions/:*)","Bash(ls ~/.claude/tools/:*)","Bash(find ~/.claude/skills:*)","Skill(Dev10x:ticket-create)"] |
Skill Audit
Analyze a Claude Code session transcript for skill compliance,
missed invocations, user corrections, and process improvements
worth persisting into skill definitions.
Strategies
Lightweight (default): Works from visible conversation
context. No transcript extraction, no subagent fan-out. Presents
inline findings and a structured disposition gate. Runs in the
current session — no separate terminal needed.
Forensic (--full or escalation): Full transcript extraction
and Wave 1/2 subagent fan-out. Use when the lightweight path
cannot answer, or when the supervisor explicitly requests deep
analysis. Run from a separate terminal.
Instructions
The full workflow — strategy selection, task creation, session
resolution, wave orchestration, phase references, and reporting —
lives in instructions.md.
When this skill is invoked, Read instructions.md now and
follow it end-to-end. TaskCreate calls, AskUserQuestion
gates, and Agent dispatches documented there are REQUIRED.
