菜单
Audit installed skills and agents against best practices. Use when reviewing skill quality, checking for common problems, or before upgrading existing skills. Read-only — produces a report, changes nothing.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
When the user wants to codify a repeatable code generation pattern into a jig recipe — design variables, choose operations (create/inject/patch/replace), write Jinja templates, and validate the result. Use when someone says "I keep writing this same boilerplate", "make a recipe for X", or wants to turn a manual edit into an automated pattern.
When the user needs to chain multiple jig recipes into a multi-step workflow with conditional execution and shared variables. Use when a pattern spans multiple files that need coordinated creation and modification, like scaffolding an entire endpoint or a full feature slice.
Scaffold queryset/manager/selector layers and wire view reads through selector entrypoints with jig.
Generate two-step rollout-safe Django migrations plus model-field patching using jig.
Insert consistent method/step structured logging lines into target functions using jig.
Write deterministic service tests with stable fixtures and boundary mocks without using jig.
| name | audit |
| description | Audit installed skills and agents against best practices. Use when reviewing skill quality, checking for common problems, or before upgrading existing skills. Read-only — produces a report, changes nothing. |
| allowed-tools | Read Glob Grep |
Read-only audit of installed skills and agents. Grades each one against ${CLAUDE_SKILL_DIR}/../improve/references/best-practices.md and produces a structured report. Changes nothing.
/audit — audit all skills and agents
/audit event-logging — audit a specific skill by name
/audit .claude/agents/ — audit all agents in a directory
If $ARGUMENTS names a specific skill, audit only that one. If empty, audit everything.
Find all SKILL.md files and agent definitions:
skills/*/SKILL.md
.claude/skills/*/SKILL.md
.claude/agents/*.md
For each file, read the full content. You need the actual text to audit, not just the path.
Read ${CLAUDE_SKILL_DIR}/../improve/references/best-practices.md for the grading criteria.
For each skill or agent, evaluate these dimensions:
references/ for progressive disclosure?disable-model-invocation: true?Output a structured report. For each skill:
### <skill-name> (<location>)
Lines: <count> | References: <yes/no> | Gotchas: <yes/no>
| Dimension | Grade | Finding |
|------------------------|-------|--------------------------------------------|
| Description | ✓/✗ | <specific finding> |
| Structure & Size | ✓/✗ | <specific finding> |
| Gotchas | ✓/✗ | <specific finding> |
| Hardcoded Paths | ✓/✗ | <specific finding> |
| Content Quality | ✓/✗ | <specific finding> |
| Side Effect Protection | ✓/✗ | <specific finding or N/A> |
**Priority fix**: <the single most impactful improvement for this skill>
After all individual reports, add a summary:
## Summary
Audited: <count> skills, <count> agents
Passing all checks: <count>
Most common issues:
1. <issue> — <count> skills affected
2. <issue> — <count> skills affected
3. <issue> — <count> skills affected
Recommended upgrade order:
1. <skill> — <why this one first>
2. <skill> — <why>
3. <skill> — <why>
The upgrade order should prioritize skills that are heavily used but poorly built — high impact, low quality. If you can't determine usage, prioritize by how many checks failed.
/upgrade <skill-name> for fixing.disable-model-invocation. They do have tools restrictions and model selection — audit those instead.