Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
44 lines (31 loc) · 2.44 KB

CONTRIBUTING.md

File metadata and controls

44 lines (31 loc) · 2.44 KB

Contributing to Second Pass

Thanks for considering a contribution. This is a small project — the contribution surface is intentionally narrow.

What contributions are most welcome

  1. Compatibility reports. Tested Second Pass against a skill not in the README compatibility table? Open an issue with the skill name, source repo, output type, and a paragraph on whether the judge grades it sensibly. These reports are the single most valuable contribution.
  2. Skill rule-set translations. If you maintain a popular skill and want Second Pass to grade it well, the highest-impact thing is making sure your SKILL.md clearly describes what success looks like. The judge reads that file. If you want help auditing your skill for "Second-Pass legibility," open an issue.
  3. Bug fixes. If the loop misbehaves (mis-grading, infinite iteration, status line wrong), file an issue with the exact prompt, skill, output, and observed behavior.
  4. Rubric improvements. The A–F rubric in agents/self-judge.md is a single point of judgment. If you have evidence that a specific clause is mis-grading a class of outputs, propose a change with reasoning.

What's out of scope (for now)

  • Multi-skill orchestration. Tracked as a public issue.
  • External-grader (different model judging) support. Tracked as a public issue.
  • Per-skill custom rubrics. Deliberately deferred — the universal rubric is the value prop.
  • Caching of judge output. Not planned.

Filing an issue

Use the templates in .github/ISSUE_TEMPLATE/. Bug reports need:

  • The originating skill name + source link
  • The user request that produced the artifact
  • The artifact (output) being graded
  • Self-judge response (grade + feedback)
  • Expected vs actual

Pull requests

  1. Fork, branch, edit, push.
  2. Keep the PR scope tight — one change per PR.
  3. Update CHANGELOG.md under ## [Unreleased] with a one-line description.
  4. If you change rubric behavior, include a before/after example showing the grade shift.
  5. Don't break existing slash command interfaces or config schema without a major version bump.

Code style

  • Markdown skill files: keep frontmatter minimal, lead body with overview.
  • Slash command .md files: imperative voice, terse, do not explain what the skill is.
  • JSON config: 2-space indent, no trailing commas.

Questions

Open a GitHub Discussion (if enabled) or an issue tagged question. There is no Discord, no mailing list — keep the conversation in the repo.