Enables automatic CodeRabbit reviews for pull requests targeting main.Description

[jpascucci-nv]

Addresses #2137 rollout steps 1 and 2.

Description

The new .coderabbit.yaml is based on CCCL’s low-noise configuration and adapts its review guidance to cuda-python’s packages and workflows. It:

• Uses the chill review profile with brief, severity-prefixed comments.
• Disables walkthrough/status noise, finishing-touch suggestions, and pre-merge checks.
• Skips drafts and PRs authored by repository automation bots.
• Enables Gitleaks, markdownlint, and ShellCheck.
• Adds focused review instructions for cuda-bindings, cuda-core, cuda-pathfinder, the cuda-python metapackage, benchmarks, CI, GitHub automation, and Cython sources.
• Uses the repository’s existing AGENTS.md, contribution guides, package documentation, and release-note indexes as review context.

Automatic reviews are intentionally limited to main. CodeRabbit and Greptile will initially run together so their signal, noise, and overlap can be evaluated on real PRs before expanding CodeRabbit to backport branches or changing either reviewer’s configuration.

Testing

• Parsed .coderabbit.yaml successfully.
• Ran YAML linting.
• Verified the SPDX header.
• Verified every configured knowledge-base file exists.
• Verified auto-review is enabled only for main.
• Ran git diff --check.

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[github-actions]

Doc Preview CI
Preview removed because the pull request was closed or merged.

[mdboom]

For deterministic checks (gitleaks, markdownlint, shellcheck) what is the advantage of doing that with coderabbit vs. precommit hooks? precommit hooks are great because they run quickly and locally before committing... it seems like a step back to do them later in CodeRabbit instead.

[jpascucci-nv]

For deterministic checks (gitleaks, markdownlint, shellcheck) what is the advantage of doing that with coderabbit vs. precommit hooks? precommit hooks are great because they run quickly and locally before committing... it seems like a step back to do them later in CodeRabbit instead.

@mdboom I think your comment should be directed to @rparolin, who specified those settings, I don't have an answer .

[rparolin]

For deterministic checks (gitleaks, markdownlint, shellcheck) what is the advantage of doing that with coderabbit vs. precommit hooks? precommit hooks are great because they run quickly and locally before committing... it seems like a step back to do them later in CodeRabbit instead.

@mdboom I think your comment should be directed to @rparolin, who specified those settings, I don't have an answer .

When my agent analyzed CCCLs setup thats where those settings came from. So if there is a specific reason why they work for CCCL and wouldn't work for cuda-python then fine lets remove them. The issue write up was intended to be a starting place and not overly prescriptive of what the landed solution should be. However, I would love to see an articulated argument of why we are deviating from CCCLs setup and why that doesn't work for us?

[mdboom]

For deterministic checks (gitleaks, markdownlint, shellcheck) what is the advantage of doing that with coderabbit vs. precommit hooks? precommit hooks are great because they run quickly and locally before committing... it seems like a step back to do them later in CodeRabbit instead.

@mdboom I think your comment should be directed to @rparolin, who specified those settings, I don't have an answer .

When my agent analyzed CCCLs setup thats where those settings came from. So if there is a specific reason why they work for CCCL and wouldn't work for cuda-python then fine lets remove them. The issue write up was intended to be a starting place and not overly prescriptive of what the landed solution should be. However, I would love to see an articulated argument of why we are deviating from CCCLs setup and why that doesn't work for us?

No objection to putting these linters in cuda-python. It just seems nonsensical to put deterministic linting into a non-deterministic agent-driven review. It's a step back -- precommit runs locally and quickly on every commit. CodeRabbit is slow and non-deterministic. cuda-python has historically put linters into pre-commit, and I think we should continue to do that.

[mdboom]

Thanks, LGTM