What it does
OrcaCode Review checks each pull request automatically and posts its findings as comments on the affected lines. Findings are graded P0 to P3; P0 and P1 fail the check and block the merge, while lower severities pass. Which severities block and which are posted is configurable, and a /orcacode-review comment requests another run. Models are routed through OrcaRouter, where you choose the reviewer model, review mode, custom rubric and guardrails such as secret and PII detection. The Action needs an OrcaRouter API key stored as a repository secret. The review engine is Open Code Review, which the page says is Apache-2.0 licensed and ships inside the package. One command also installs skills for 36 coding-agent platforms, including Claude Code, Cursor and Codex. These let your agent review local changes or a pull request on demand with the model it already has, without the Action, an OrcaRouter account or an API key. The page says it does not execute pull request code, and that optional run reports carry only metadata such as severity counts and gate result.
Capabilities
- Automatic review on every push, with inline findings
- P0 to P3 severity levels and a merge gate for P0 and P1
- Reviewer model, review mode and rubric set in the OrcaRouter console
- Local review through a skill for 36 coding-agent platforms
- Optional OrcaRouter guardrails such as secret and PII detection
- Re-run on demand with a /orcacode-review comment
When to use it
- Gating merges on serious review findings in a GitHub repository
- Reviewing uncommitted changes or a branch locally through your existing coding agent
- Teams that want to choose their own review model and pay only for inference
