Contributing¶
Contributions — by people and by coding agents — are welcome. The process is the one in Agent Workflow; this page is the short form.
Development here is specified agent-first: the canonical guide is
AGENTS.md
at the repository root, routing to the same
Agent Workflow, Code Index
and Conventions this site hosts. Point a coding
agent at a checkout and it reads exactly the rules a human contributor
does — one rule set, no drift.
Contributions are accepted under the Apache License 2.0 that covers the
OpenCDA-MARL code (see
LICENSE).
Files from third parties keep their own terms; see
THIRD_PARTY_NOTICES.md.
- Fork
radar-lab/OpenCDA-MARLand branch frommain(git switch -c feat/short-summary). - Make one verified change. Tests ship with the change;
pixi run pytest test -qis green; lifecycle, launcher and monitor changes get a live smoke run. Docs are updated on the page that owns the feature. - Open a pull request against
mainwith the PR template filled in — what changed, how it was verified, what you could not verify. Maintainers review and squash-merge.
Where things go:
- A new algorithm does not need a PR at all: package it as described in Add an Algorithm and it plugs into the framework from outside. PR it only if you want it shipped as a built-in.
- A new scenario or profile is a YAML file under
opencda_marl/_bundled/configs/<family>/that states only what it changes from the family default and scenario base. - A framework fix keeps the seams listed in the workflow page: one
episode loop, registration instead of name dispatch, every
opencda/patch recorded in the patch ledger.
Bugs and proposals go to issues; include the scenario reference, the command, and the log lines around the failure.