skill-create

Skill compliance audit

When to use

When the user provides or does not provide a specific path, perform a review of a skill directory coveringstructure + metadata + maintainability, against Cursor's official "Creating Skills" conventions (consistent with the built-in create-skill guide).

Default: if the user uses @ to point to a directory (for example skills/a-share-stock-picker/), use that directory as the audit target; otherwise confirm the path first before checking.

Audit steps

1. List the directory: confirm the existence of a root-level SKILL.md; record subdirectories such as references/、scripts/、agents/ .
2. Read the frontmatter（SKILL.md top-level --- block):
   • name: exists, all lowercase, only alphanumeric characters and hyphens, ≤64 characters.
   • description: non-empty, ≤1024 characters,third person(avoid "I/you can..."), and clearly state both what it does (WHAT) and when to use it (WHEN), including searchable trigger words.
3. Read the body size：SKILL.md The main body should be ≤500 lines; if too long, split it into references/*.md etc. (progressive disclosure).
4. Links and references: when pointing from SKILL.md to attached files,one level deep is sufficient (for example references/foo.md); avoid critical steps that can only be read through deep nesting.
5. Scripts: if scripts/exists, check whether SKILL.md explains the purpose, how to invoke it, and whether it should beexecutedor only used as reference; paths should use forward slashes.
6. Anti-pattern scan: Windows backslash paths, listing too many equivalent tools with no default option, easily outdated hardcoded dates (not examples), inconsistent terminology.
7. Repository extension items(common in this repo): ifagents/openai.yaml etc. exist, check whether they are consistent with the SKILL.md of name/capability description, and whether they are mentioned in the body (if not mentioned, record it as "optional metadata, suggest adding a sentence in SKILL").

Output format (required)

Give a concise report in Chinese, with fixed sections:

1. Conclusion: compliant / basically compliant with minor changes needed / needs major revision (one sentence).
2. Compliant items: list of items.
3. Issues and suggestions: divided into severe(affects discoverability, misuse, safety) / general(readability, duplication, structure).
4. Optional follow-ups: if aiming for full marks, the top 1–3 things to do first.

Do not edit the draft for the user unless the user explicitly asks to modify that skill.

Quick self-check checklist (can be checked inline)

• name / description complies with field rules
• description includes WHAT + WHEN + trigger words
• SKILL.md line count and splitting are reasonable
• references are one level deep, attached files can be read on demand
• scripts are documented, path style is correct
• no typical anti-patterns