Skill Checker

Audit a target SKILL.md against the Agent Skills specification and generate a Chinese HTML report. Use when the user asks to check, audit, review, or optimiz...

Security Scan

The skill's files, runtime instructions, and code align with its stated purpose: it audits SKILL.md files locally and emits a Chinese/HTML report; it does not request secrets, perform network calls, or install arbitrary code.

Purpose & Capability

Name/description match the included code and tests. The project contains a Python checker script, fixtures, a checklist, and docs; all requested functionality (auditing SKILL.md and producing an HTML report) is implemented and proportional to the stated purpose.

Instruction Scope

SKILL.md instructs the agent to run the local Python script against a SKILL.md path or skill directory, resolves SKILL.md, and writes an HTML report. The checker only reads the supplied SKILL.md and its directory name; there are no instructions to read unrelated files, system credentials, or to transmit data to external endpoints.

Install Mechanism

No install spec is provided (instruction-only with accompanying scripts). The shipped code is pure Python and tests; nothing downloads or extracts external archives or adds packages at install time.

Credentials

The skill declares no required environment variables, no credentials, and no config paths. The code performs local file I/O only and does not access environment secrets.

Persistence & Privilege

always is false and the skill does not request persistent or elevated platform privileges. It only writes its own HTML report output when asked and does not modify other skills or global agent configuration.

Assessment

This skill appears coherent and limited: it reads a specified SKILL.md (or skill directory), validates frontmatter and body rules, and writes a local HTML report. Before installing or running it: inspect scripts/check_skill.py if you want to be extra cautious (the code is local Python and easy to review), run tests if you need assurance, and choose an explicit --out path so you know where the HTML report will be written. There are no network calls or secret requirements, but as with any tool that reads files, avoid pointing it at sensitive directories you don't want scanned or logged.

Check only SKILL.md.

Use this skill when the user explicitly wants to check, audit, review, or optimize a SKILL.md file.

Accept either a SKILL.md path or a skill directory path. If the input is a directory, resolve SKILL.md inside it before running any checks. Support absolute paths directly. Prefer absolute paths when auditing skills outside the current workspace.

Run the checker script (PowerShell):

# Auto locate this skill directory (works even if you run from another folder)
$SkillDir = Split-Path -Parent $MyInvocation.MyCommand.Path
Set-Location $SkillDir

# Run
python .\scripts\check_skill.py <target-path> [--out <report-path>]

Example with absolute paths:

$SkillDir = Split-Path -Parent $MyInvocation.MyCommand.Path
Set-Location $SkillDir
python .\scripts\check_skill.py "C:\path\to\skill\SKILL.md" --out ".\reports\skill-report.html"

Use the generated HTML report as the primary output. Summarize the overall result, the severe issue count, and the highest-priority fixes in the final response.

What to check

Validate the target SKILL.md against the Agent Skills specification:

1. Confirm YAML frontmatter exists and is parseable.
2. Confirm required fields such as name and description.
3. Validate optional specification fields when present.
4. Check that name matches the skill directory name when it can be inferred.
5. Review whether description says what the skill does and when to use it.
6. Review whether the body contains enough actionable guidance to help another agent execute the skill.

Treat specification violations as severe issues.

Treat obvious semantic failures as severe issues too, including:

• descriptions that are too short or too vague to support triggering
• bodies that are nearly empty or do not provide actionable guidance

Treat weaker organization or thin guidance as warnings.

Return a failing result when severe issues are 2 or more. Otherwise return a passing result.

Reporting

Always point the user to the generated HTML report path, especially when the target skill lives outside the current repo.

When summarizing results:

1. State whether the result is passing or failing.
2. Mention the severe issue count and warning count.
3. Call out the first fixes the user should make.

References

Read references/specification-checklist.md when you need the exact rules and severity mapping used by this skill.