julien777z/Code Review Action
Review pull requests with Claude or Cursor — inline severity-rated comments and an approval verdict.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Activelast commit Jul 11, 2026
- License
- MIT
Pinned Snippet
uses: julien777z/code-review-action@cd88357f145af4c31d6aff33384b8c1696f3fe88 # v0.3.2tags can be moved; commit SHAs can't. why a SHA?
Inputs
| name | description | required | default |
|---|---|---|---|
| github-token | Token used to read the diff and post reviews and check runs. | — | ${{ github.token }} |
| resolve-token | Token used to resolve the action's own review threads. The default GITHUB_TOKEN cannot resolve threads (GitHub rejects it with "Resource not accessible by integration"), so set this to a GitHub App installation token (or a fine-grained PAT) with pull-request write to enable auto-resolution. Empty falls back to github-token. | no | "" |
| anthropic-api-key | Anthropic API key for the Claude backend and PR-summary generation. | no | "" |
| cursor-api-key | Cursor API key for the Cursor backend. | no | "" |
| review-model | Backend to use — auto, claude, or cursor. | — | auto |
| first-review-model | Backend for the PR's first review (opened/ready). Empty falls back to review-model. | — | "" |
| claude-model | Anthropic model id for Claude Managed Agents reviews and Claude PR-summary generation. | — | claude-opus-4-8 |
| cursor-model | Cursor model id for the Cursor backend. | — | composer-2.5 |
| additional-context | Extra context injected into the review prompt. | — | "" |
| approval-include | Comma-separated severities that, if open, request changes (otherwise comment only). | — | critical |
| approval-disable | Post review comments only and skip the approval verdict and check run. | — | false |
| pr-review-summary | Append an AI-generated summary (bullets, risk note, overview) to the PR description on open/ready events. | — | true |
| enforce-project-rules | Enforce the repository's own coding rules as part of the review. Each backend loads the rule files it understands from the checked-out repo. No-op when the repository defines none. | — | true |
| project-rules-severity | Fixed severity to report project-rule violations at (critical, high, medium, or low). Empty lets the review rate each violation itself. | — | "" |
| simplify-suggest | Also suggest code simplifications using the code-simplify skill, as optional suggestions. | — | false |
| simplify-suggest-severity | Severity to report simplification suggestions at (critical, high, medium, or low). Empty defaults to low. | — | "" |
| simplify-nearby-code | Extend simplification suggestions to weigh nearby and related code as review criteria. | — | false |
| min-severity | Lowest severity worth posting and counting — critical, high, medium, or low. | — | low |
| low-findings-cap | Maximum number of low-severity findings to post per review; the runner selects the most important lows once the review stream ends. | — | 3 |
| max-findings | Overall cap on inline comments per review (empty = uncapped). | — | "" |
| include-paths | Comma/newline-separated globs to restrict the review to. | — | "" |
| exclude-paths | Comma/newline-separated globs to skip (generated files, lockfiles, vendored code). | — | "" |
| trigger-phrase | Comment phrase that triggers a manual review. | — | agent review |
| review-drafts | Review draft pull requests. | — | true |
| pr-number | PR number to review on a workflow_dispatch run. | — | "" |
| review-timeout-minutes | Overall cap on the review agent's runtime in minutes. The agent streams findings incrementally, and a few minutes before the cap the runner interrupts it and asks it to immediately emit any findings it has identified but not yet reported, so a timed-out review still posts what was found. Set to 0 to disable. | — | 15 |
Outputs
no outputs