What does the SKILL.md Linter check?

Required sections (description, modes, iron laws, examples), quality of iron laws (are they specific constraints or vague suggestions?), mode definitions (are they actionable?), example quality, and overall completeness.

What's a good SKILL.md score?

80+ is good. 90+ is excellent. Below 60 usually means missing required sections or vague iron laws. The linter tells you exactly what to fix.

Can I use this with any AI agent skill format?

It's optimized for SKILL.md format as used by Claude/WireClaw agents, but works on any structured skill definition in Markdown.

Yes, completely free. No account required.

Free Tool · No account · Instant

SKILL.md Linter

Paste your SKILL.md file. Get a scored quality report across 8 dimensions — frontmatter, description, trigger, output format, examples, iron laws, token budget, error handling. 100% algorithmic. No CLI needed.

8 Dimensions scored

1 · 20 pts

Frontmatter Completeness

Required fields: name (or skill), description, trigger (or when/argument-hint). Optional: version, modes, author, model, context.

2 · 15 pts

Description Quality

50–300 chars is ideal. Must describe what it does and when. Vague ("does tasks") or overlong descriptions penalized.

3 · 15 pts

Trigger Clarity

Trigger should specify the exact phrase or condition that activates the skill. "When the user asks to X" format is gold standard.

4 · 15 pts

Output Format

Output Format or Response Format section present. Specifies structure: JSON, markdown, plain text. Examples in the section are a bonus.

5 · 12 pts

Iron Laws / Constraints

Iron Laws, Constraints, or Rules section present. At least 3 constraints listed. Explicit "never" or "always" rules score higher.

6 · 10 pts

Examples

At least one concrete example (input/output pair, usage sample, or worked example). Examples dramatically improve agent adherence.

7 · 8 pts

Token Budget

Under 800 tokens: optimal. 800–2000: good. 2000–8000: acceptable. Over 8000: too long — agents hit saturation and skip sections.

8 · 5 pts

Error Handling

Error recovery, edge cases, or failure modes section present. What should the agent do when the input is ambiguous or the task fails?