Contributing¶
Summary
RelyLoop welcomes contributions under Apache 2.0. The authoritative guide
is CONTRIBUTING.md
in the repo; this page is the orientation for first-time contributors.
Contributing with an AI agent?
RelyLoop is built agent-first and spec-driven. If you use Claude Code (or another agent), see Contributing with AI agents for onboarding prompts, the skills, the spec → plan → implement pipeline, and how cross-model review works.
First-time path¶
- Set up the dev environment. You'll need Docker 24+ (Compose v2),
Python 3.12+ (managed by
uv), Node 20 LTS + pnpm 9, and ~16 GB RAM. - Pick something small first. A docs fix, a test, or a clearly-scoped bug is the best way to learn the review loop.
- Branch and commit. Short-lived feature branches off
main, named<type>/<short-description>. RelyLoop uses Conventional Commits. - Sign your commits (DCO). RelyLoop uses the Developer Certificate of
Origin instead of a CLA — every commit needs a
Signed-off-by:trailer. Just commit with-s: A CI check and a localcommit-msghook both enforce it. - Run the tests.
make testruns unit, integration, and contract layers. - Open a PR against
main. The PR template prompts for what changed, why, and how you tested. CI runs lint, type-check, tests, secret scanning, and the frontend build; a maintainer reviews and squash-merges.
Adding an adapter¶
RelyLoop's engine, LLM-provider, and Git-provider adapters are built for community extension. Each implements a Protocol, passes the contract test suite, and documents its auth flow and quirks. See the adapters architecture doc and the spec's adapter sections.
Code of Conduct¶
A short kindness ask, not a long rulebook —
CODE_OF_CONDUCT.md.
Reporting issues¶
- Bugs / feature requests — use the issue templates on GitHub Issues.
- Security vulnerabilities — do not file a public issue; see Security.
- Questions / design discussions — use GitHub Discussions.