Skip to content

Latest commit

 

History

History
43 lines (35 loc) · 1.84 KB

File metadata and controls

43 lines (35 loc) · 1.84 KB

Contributing

Principles

  • Each skill in skills/ is the single source of truth for its workflow. Edit it directly — there is no build step.
  • Keep skills customer-safe end to end.
  • Ask for one artifact at a time in diagnostic workflows.
  • Prefer exact config and CI snippets in recommendation workflows when enough context is available.
  • Installed skills must not fetch remote content directly.
  • Installed skills must not include direct download commands such as curl or wget.
  • Hosted metadata URLs are human-facing artifacts only unless the user manually converts them into a local file path or pasted contents.
  • Separate the minimal technical fix from the safer recommended fix whenever untraced, externals, or similar suppression is involved.

Skill Structure

skills/<name>/
  SKILL.md          Main instructions and frontmatter (required)
  template.md       Intake template for users to fill in
  examples/         Example outputs showing expected format
  reference/        Detailed reference docs loaded on demand
  evaluations/      JSON test scenarios
  agents/           openai.yaml for Codex UI metadata
  assets/           Icons and visual assets

Updating a Skill

  1. Edit files directly in skills/<name>/.
  2. Keep SKILL.md under 500 lines. Move detailed content to reference/.
  3. If you change the intake workflow, update template.md to match.
  4. If you add a new reference file, link it from the ## References and examples section in SKILL.md.

Adding a New Skill

  1. Scaffold the skill:
    npx skills init skills/<new-name>
  2. Fill in SKILL.md with name, description, and instructions.
  3. Replace template.md with the intake form users should fill in.
  4. Add examples/, reference/, and evaluations/ as needed.
  5. Update README.md with a short description of the new skill.