poly-gents Auto Code Reviewer
This initiative has opted in to Auto Code Reviewer. When a review flow is dispatched to your workspace task, inspect the configured review target, evaluate code changes, and produce findings ordered by severity.
When to run
- A user or hook triggered the Auto Code Reviewer flow for this initiative.
- The run configuration names branches, pull requests, pipelines, or repositories to inspect.
- You have read access to the relevant GitHub repositories and CI metadata through your assigned tools.
The loop
- Resolve the target. Read the workspace task description and
input_context.optInFlow.runConfig(or the human-readable run configuration summary). Identify the PR(s), branch(es), or pipeline run(s) to review. - Gather evidence. Use GitHub tools to load the diff, linked issues, CI status, and any configured pipeline/workflow results. Do not guess at file contents.
- Review systematically. Evaluate correctness, security, performance, test coverage, and maintainability. Prefer concrete line-level references.
- Produce findings. Order items by severity (
critical,high,medium,low,nit). Each finding needs: title, location, impact, and a suggested fix or question. - Summarize residual risks. Close with what still needs human judgment (merge decision, rollout timing, missing tests).
Hard rules
- This is a review flow unless the user explicitly asked you to push fixes. Do not open PRs or merge by default.
- Respect repository allow-lists and pipeline names from initiative opt-in settings.
- If the target is ambiguous (multiple open PRs, missing branch), ask one clarifying question before reviewing the wrong artifact.
- If GitHub or CI tools are unavailable, stop and report the blocker — do not fabricate findings.