toriihq/Claude Code Review
AI-powered code review using Claude. Handles diff capture, re-review reconciliation, cost tracking, and configurable review authority.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Stale
- License
- None
Inputs
| name | description | required | default |
|---|---|---|---|
| anthropic-api-key | Anthropic API key | yes | — |
| review-guide-path | Path to repo review guide markdown. Empty = no guide. | no | "" |
| critical-rules | Multiline string of BLOCKER-level rules injected into prompt | no | "" |
| extra-prompt | Appended to end of prompt (custom instructions) | no | "" |
| include-pr-description | Feed PR title+body into review prompt | no | true |
| context-intro | Opening line of prompt (repo identity/context) | no | You are a code reviewer. |
| max-files | Skip review if PR exceeds this many files | no | 50 |
| max-diff-lines | Truncate diff after N lines | no | 3000 |
| max-diff-bytes | Truncate diff after N bytes | no | 80000 |
| max-turns | Claude conversation turn limit | no | 30 |
| timeout-minutes | Job timeout (informational — consumer workflow controls actual timeout) | no | 20 |
| model | Claude model to use | no | claude-sonnet-4-6 |
| allowed-tools | Tools Claude can use during review | no | Bash,Read,Write,Grep,Glob |
| review-authority | Review authority level: comment-only, request-changes, or full | no | request-changes |
| approve-threshold | For full mode: strict (zero MEDIUM+) or normal (zero HIGH+) | no | strict |
| approve-max-files | For full mode: only approve PRs with <= N files | no | 50 |
| review-label | Label name passed to claude-code-action label_trigger. Must match consumer workflow if: condition. | no | claude-review |
| trigger-phrase | Comment trigger phrase (used to exclude trigger comments from author feedback) | no | @claude |
| default-branch | Base branch for guide fetch and config restore. Empty = auto-detect from repo settings. | no | "" |
| skip-if-already-reviewed | Skip on label trigger if no new commits since last review | no | true |
| include-previous-review | Enable re-review reconciliation with previous findings | no | true |
| track-cost | Append cost/turns/model to review comment | no | true |
| dismiss-previous-reviews | Dismiss old Claude reviews before posting new one | no | true |
| post-check-run | Also post the review verdict as a GitHub Check Run (in addition to the PR review). Requires checks:write in the consumer workflow permissions. | no | false |
| blocking | Check Run only. false = neutral conclusion (advisory). true = failure conclusion when changes are requested. NOTE: never gates merge unless the consumer marks the check required in branch protection. | no | false |
| check-run-name | Name of the posted Check Run (also the name to select in branch-protection required checks). | no | Claude Code Review |
Outputs
no outputs