| ai_api_key | API key for your AI provider. Store as a repository secret. | yes | — |
| ai_model | Model identifier for your AI provider. Examples: claude-sonnet-4-6 (Anthropic), gpt-4o (OpenAI), gemini-2.0-flash (Gemini).
| yes | — |
| ai_provider | AI provider adapter: anthropic, openai, gemini, or github-models. Use github-models for zero-cost review with GITHUB_TOKEN — no extra account needed. Use openai for any OpenAI-compatible endpoint (Groq, Azure OpenAI, Mistral, Ollama) with ai_base_url.
| yes | — |
| ai_base_url | Optional base URL override for OpenAI-compatible endpoints. Example: https://api.groq.com/openai for Groq. Leave empty to use the provider's default endpoint.
| no | "" |
| pr_number | Pull request number to review. | yes | — |
| pr_title | Pull request title. | yes | — |
| pr_body | Pull request description body. | no | "" |
| diff_lines | Lines-per-chunk threshold for diff processing. When chunk_large_diffs is true in .github/reviewsentry.yml, this is the maximum lines per review pass. When chunk_large_diffs is false or absent (default), diffs exceeding this limit are truncated and the review lists any files that were not reviewed. Default 1500.
| no | 1500 |
| review_criteria | Additional review criteria to append to the standard checklist. Plain text, one criterion per line.
| no | "" |
| custom_rules | Custom sensitive data scan rules (one per line). Applied in addition to built-in sensitive data checks. Use for domain-specific terms, product names, or internal identifiers.
| no | "" |
| system_context | Optional project-specific context appended to the AI system prompt. Use to tell the reviewer about your stack, conventions, or constraints. Example: "TypeScript/React project. No class components. Tailwind CSS only."
| no | "" |
| review_drafts | Whether to review draft pull requests. Default true — reviews fire on drafts by default. Set to false to skip review until the PR is marked ready for review. Accepted values: true/1/yes or false/0/no. Any other value triggers a workflow warning and defaults to true.
| no | true |
| github_token | GitHub token for posting the review comment. | yes | — |
| fail_on | When to fail the workflow based on the AI verdict. Accepted values: never (default) — informational only, no CI failure; request_changes — exit non-zero when the AI verdict is REQUEST CHANGES. Any other value triggers a workflow warning and defaults to never.
| no | never |
| show_passing_criteria | Whether to include passing criteria (no issues found) in the review output. Default true. Set to false to show only criteria with findings, keeping the review concise for large PRs. Accepted values: true/1/yes or false/0/no. Any other value triggers a workflow warning and defaults to true.
| no | true |
| max_tokens | Maximum tokens for the AI response. Default 4096. Raise if reviews are being cut off mid-criterion; lower to reduce cost on very small PRs. Values below 256 are clamped to 256.
| no | 4096 |
| pr_body_chars | Maximum characters of the pull request description to include in the review context. Default 2000. Raise for very detailed PR descriptions; lower to reduce token usage on simple PRs. Descriptions longer than this limit are truncated and the AI is notified so it does not flag the description as incomplete. Values below 50 are clamped to 50.
| no | 2000 |