Cross-agent skill quality gate for SKILL.md files.
Validates against the agentskills.io specification, scores description discoverability, checks file references, and warns about cross-platform compatibility.

---

What It Does

skillcheck catches problems in your SKILL.md files before they hit production across Claude Code, VS Code/Copilot, OpenAI Codex, Cursor, and other agents:

• Frontmatter validation -- required fields, character constraints, length limits, reserved words, first/second-person voice, XML tags, unknown fields
• Full name spec compliance -- leading/trailing hyphen checks, consecutive hyphen checks, directory-name matching (required by VS Code or the skill silently fails to load)
• Description quality scoring -- scores 0-100 across action verbs, trigger phrases, keyword density, specificity, and length. Agents use descriptions to decide whether to activate a skill. A bad description means the skill never fires
• File reference validation -- checks that relative file references in the body actually exist on disk and that reference depth stays within one level of SKILL.md
• Progressive disclosure budget -- validates the three-tier token budget (metadata ~100 tokens, body <5000 tokens, resources on demand) and flags bloat patterns like oversized code blocks, large tables, and embedded base64
• Cross-agent compatibility warnings -- flags fields that only work in Claude Code (model, disable-model-invocation, mode, hooks), notes VS Code directory-name requirements, and marks fields with unverified behavior in Codex and Cursor
• CI-friendly -- JSON output, deterministic exit codes, zero config

Install

pip install skillcheck

Or install from source with dev dependencies:

pip install -e ".[dev]"

Quick Start

# Validate a single file
skillcheck path/to/SKILL.md

# Scan a directory recursively for all SKILL.md files
skillcheck skills/

# Machine-readable output for CI pipelines
skillcheck skills/ --format json

# Score description quality with a minimum threshold
skillcheck SKILL.md --min-desc-score 50

# Check compatibility for a specific agent
skillcheck SKILL.md --target-agent vscode --strict-vscode

# Skip checks that need filesystem context (useful in CI)
skillcheck SKILL.md --skip-dirname-check --skip-ref-check

GitHub Action

Add skillcheck to any CI pipeline in three lines:

# .github/workflows/skills.yml
name: Skill Validation
on: [push, pull_request]

jobs:
  skillcheck:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: moonrunnerkc/skillcheck@v0
        with:
          path: .github/skills/

That's it. Failures block the PR, diagnostics appear inline on the diff, and a summary table is added to the job.

Action Inputs

Input	Default	Description
path	.	Path to a SKILL.md file or directory to scan recursively
version	latest	Pin a specific skillcheck version (e.g., 0.2.0)
min-desc-score		Minimum description quality score (0-100)
target-agent	all	Scope compat checks: claude, vscode, or all
strict-vscode	false	Promote VS Code compat issues to errors
skip-dirname-check	false	Skip directory-name matching check
skip-ref-check	false	Skip file reference validation
ignore		Comma-separated rule prefixes to suppress
max-lines	500	Override line-count threshold
max-tokens	8000	Override token-count threshold

Action Outputs

Output	Description
exit-code	0 = pass, 1 = errors, 2 = input error
json	Full JSON output from skillcheck

What You Get

• PR annotations that put errors and warnings inline on the diff
• Job summary with a Markdown results table on the workflow run page
• Exit code gating so the step fails if any skill has errors

Examples

Strict VS Code mode with a description quality floor:

- uses: moonrunnerkc/skillcheck@v0
  with:
    path: skills/
    strict-vscode: true
    min-desc-score: 60

Use the JSON output in a downstream step:

- uses: moonrunnerkc/skillcheck@v0
  id: check
  with:
    path: SKILL.md
- run: echo '${{ steps.check.outputs.json }}' | jq .files_failed

Example Output

✔ PASS  skills/deploy/SKILL.md
            · info              description.quality-score  Description quality score: 85/100.

✗ FAIL  skills/pdf-tool/SKILL.md
  line 2  ✗ error    frontmatter.name.directory-mismatch  Name 'pdf-processor' does not match parent directory 'pdf-tool'.
  line 3  ⚠ warning  frontmatter.field.unknown            Unknown field 'author'.
          · info     description.quality-score            Description quality score: 45/100.
                     Suggestions: Start with an action verb; Add trigger phrases.
          · info     compat.claude-only                   Field 'model' is Claude Code-specific.

Checked 2 files: 1 passed, 1 failed, 1 warning

Options

Flag	Description
--format json	Machine-readable JSON output
--max-lines N	Override line-count threshold (default: 500)
--max-tokens N	Override token-count threshold (default: 8000)
--ignore PREFIX	Suppress rules matching a prefix (repeatable)
--no-color	Disable colored output
-q, --quiet	Suppress all output; exit code only (for CI)
--skip-dirname-check	Skip directory-name matching (useful for CI temp paths)
--skip-ref-check	Skip file reference validation
--min-desc-score N	Minimum description quality score (0-100); below triggers a warning
--target-agent NAME	Scope compat checks: claude, vscode, or all (default: all)
--strict-vscode	Promote VS Code compat issues to errors
--version	Show version

Examples

# Override sizing thresholds
skillcheck skills/ --max-lines 800 --max-tokens 6000

# Suppress specific rule categories
skillcheck SKILL.md --ignore frontmatter.description

# Require description quality above 60
skillcheck SKILL.md --min-desc-score 60

# Check only Claude Code compatibility
skillcheck SKILL.md --target-agent claude

# Strict VS Code mode (name must match directory or exit 1)
skillcheck SKILL.md --strict-vscode

# Pipe-friendly plain output
skillcheck SKILL.md --no-color

# Silent mode for CI (exit code only)
skillcheck SKILL.md --quiet

Exit Codes

Code	Meaning
0	No errors (warnings and info are allowed)
1	One or more errors found
2	Input error (missing file, empty directory)

Rules

Rules marked spec are derived from the agentskills.io specification or agent-specific documentation. Rules marked advisory are best-practice recommendations enforced by skillcheck.

Rule ID	Severity	Source	What it checks
frontmatter.name.required	error	spec	name field must exist
frontmatter.name.type	error	advisory	name must be a string (catches YAML coercion of true, 123, null)
frontmatter.name.max-length	error	spec	Name must be 64 characters or fewer
frontmatter.name.invalid-chars	error	spec	Lowercase, numbers, hyphens only
frontmatter.name.leading-trailing-hyphen	error	spec	No leading or trailing hyphens
frontmatter.name.consecutive-hyphens	error	spec	No consecutive hyphens
frontmatter.name.reserved-word	error	advisory	Not a reserved word (claude, anthropic)
frontmatter.name.directory-mismatch	error	spec	Name must match parent directory (VS Code requirement)
frontmatter.description.required	error	spec	description field must exist
frontmatter.description.type	error	advisory	description must be a string (catches YAML coercion)
frontmatter.description.empty	error	spec	Description must not be blank
frontmatter.description.max-length	error	spec	Description must be 1024 characters or fewer
frontmatter.description.xml-tags	error	advisory	No XML/HTML tags in description
frontmatter.description.person-voice	error	advisory	No first/second-person pronouns
frontmatter.field.unknown	warning	advisory	Flags fields not in the spec
frontmatter.yaml-anchors	warning	advisory	YAML anchors/aliases can silently copy values
description.quality-score	info	advisory	Scores description 0-100 for agent discoverability
description.min-score	warning	advisory	Description score below --min-desc-score threshold
sizing.body.line-count	warning	spec	File exceeds line threshold
sizing.body.token-estimate	warning	spec	File exceeds token threshold
disclosure.metadata-budget	warning	spec	Frontmatter exceeds ~100 token metadata budget
disclosure.body-budget	warning	spec	Body exceeds 5000 token instruction budget
disclosure.body-bloat	info	advisory	Large code blocks, tables, or base64 in body
references.broken-link	error	advisory	Referenced file does not exist
references.escape	error	advisory	Reference resolves outside skill directory (CWE-59)
references.depth-exceeded	warning	spec	Reference deeper than one level from SKILL.md
compat.claude-only	info	spec	Field only works in Claude Code
compat.vscode-dirname	info/error	spec	Name does not match parent directory (VS Code)
compat.unverified	info	advisory	Field behavior unverified in Codex/Cursor

Case Study

The Skill That Silently Disappeared in VS Code: A deploy skill works in Claude Code but never loads in VS Code/Copilot. No error. No warning. It just isn't there. This walkthrough shows how skillcheck catches the name/directory mismatch that causes silent failures, with verified source links to the spec and VS Code documentation.

Limitations

• Token counts are estimates. The heuristic fallback has ~15% error; install tiktoken for ~5% error. Neither matches Claude's exact tokenizer (not publicly available).
• Cross-agent compatibility data for Codex and Cursor is based on available documentation as of early 2026. Fields marked "unverified" may work, may be ignored, or may cause issues. File bugs if you find discrepancies.
• Description quality scoring uses heuristics, not an LLM. It catches common patterns (missing action verbs, no trigger phrases, vague words) but cannot evaluate semantic quality.
• Directory-name matching compares against the immediate parent directory. If your CI clones into a temp path, use --skip-dirname-check.
• File reference validation only checks references extractable from markdown link syntax and source:/file: directives. Arbitrary path strings in prose are not detected.

Testing

pip install -e ".[dev]"
python3 -m pytest tests/ -v

License

MIT