跳转至

skill-lint

所属包: companion

skill-lint

Purpose

Validate a skill directory for structure, frontmatter quality, body guidance quality, and basic script safety.

When to use

Use this skill when the user asks to lint skill, check skill, validate skill, review skill quality, or asks whether a skill is valid.

Typical targets:

  • a single skill directory
  • a pack's .opencode/skills/ tree
  • a skill that fails to trigger reliably
  • a skill that is about to be published or committed

Workflow

  1. Identify the target skill directory or root directory to scan.
  2. Run scripts/lint_skill.py on the target.
  3. If the user wants a machine-readable report, use --json.
  4. If the user wants batch validation, use --recursive.
  5. If the user wants low-noise output, use --severity error or --severity warn.
  6. If the user asks for safe auto-fixes, run --fix first to show the dry-run plan, then --fix-apply only after confirmation.

Output rules

Report findings with these severities:

  • ERROR: must fix
  • WARN: should fix
  • INFO: suggestion

For every finding, provide:

  • rule ID
  • severity
  • short explanation
  • concrete fix suggestion

Always summarize:

  • total score
  • error count
  • warning count
  • info count

Must do

  • Run lint_skill.py on the requested target instead of manually improvising checks.
  • Use the reported rule IDs and severities exactly.
  • Give a concrete fix suggestion for each finding.
  • Call out whether --fix is dry run or applied.

Must not do

  • Do not invent extra blocking rules beyond the linter output.
  • Do not treat INFO as release-blocking.
  • Do not apply fixes silently; use dry run first unless the user explicitly wants changes applied.

Reference

See references/lint-rules.md for the full rule catalog and examples.