leek/AI Commit Review

Review a single commit with Claude, GPT, and Gemini. Files findings as a GitHub Issue and optionally opens a draft fix PR.

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
commit-shaCommit SHA to review. Caller must have run actions/checkout with fetch-depth: 0.yes
anthropic-api-keyAnthropic API key. Provider runs only when set.no""
openai-api-keyOpenAI API key. Provider runs only when set.no""
gemini-api-keyGemini API key. Provider runs only when set.no""
claude-authClaude provider auth mode: auto, api, or cli. auto uses claude-code-oauth-token when set, otherwise anthropic-api-key.noauto
openai-authOpenAI provider auth mode: auto, api, or cli. auto uses Codex CLI auth when codex-access-token or codex-auth-json is set, otherwise openai-api-key.noauto
claude-code-oauth-tokenClaude Code OAuth token from `claude setup-token`. Enables Claude CLI mode in auto auth.no""
codex-access-tokenCodex access token for CODEX_ACCESS_TOKEN. Enables Codex CLI mode in auto auth.no""
codex-auth-jsonOptional contents of a Codex auth.json file for Codex CLI mode. Use only on trusted private runners.no""
codex-homeOptional CODEX_HOME path for Codex CLI mode. Useful for self-hosted runners with persistent Codex auth.no""
install-cli-toolsInstall missing Claude Code or Codex CLI tools when a CLI mode is selected.notrue
claude-cli-pathClaude Code CLI command path used in Claude CLI mode.noclaude
codex-cli-pathCodex CLI command path used in OpenAI CLI mode.nocodex
codex-sandboxCodex sandbox mode used in OpenAI CLI mode. GitHub-hosted runners may require danger-full-access because Codex read-only sandboxing depends on Linux user namespaces.nodanger-full-access
claude-modelAnthropic model id.noclaude-opus-4-8
openai-modelOpenAI model id.nogpt-5.5
gemini-modelGemini model id.nogemini-3.5-flash
claude-context-fileProject context file injected into the Claude prompt. Empty disables.no""
openai-context-fileProject context file injected into the OpenAI prompt. Empty disables.no""
gemini-context-fileProject context file injected into the Gemini prompt. Empty disables.no""
prompt-filePath to a custom prompt template, relative to the workspace. Overrides the bundled generic prompt. Useful for project-specific reviewer rules.no""
exclude-pathsNewline-separated git pathspecs excluded from the diff. Each line passed verbatim to git diff. Use the :!path syntax.no""
max-diff-linesSkip review if filtered diff exceeds this many added/changed lines.no5000
skip-message-patternsNewline-separated bash glob patterns matched against the commit subject. If any matches, the commit is skipped.noMerge*
skip-author-patternsNewline-separated bash glob patterns matched against the commit author name. If any matches, the commit is skipped.no""
min-severity-for-issueLowest severity that triggers issue creation. One of: critical, warning, info.nowarning
min-models-for-fix-prNumber of providers that must agree on a high-confidence fix before a draft fix PR is opened. Set to 0 to disable fix PRs.no2
issue-labelLabel applied to created issues. Created on demand.noai-review
issue-title-prefixPrefix used in issue titles. The short SHA is appended.no[AI Review]
fix-pr-title-prefixPrefix used in fix PR titles. The short SHA is appended.no[AI Fix] Suggested fixes for
fix-branch-prefixBranch prefix for fix PRs. Short SHA is appended.noai-fix/
base-branchBase branch fix PRs target.nomain
github-tokenToken used to create issues, comments, branches, and PRs.no${{ github.token }}
node-versionNode.js version used to run the review script.no20
namedescription
reviewedtrue if the commit was reviewed, false if skipped.
skip-reasonReason the commit was skipped, if any.
diff-line-countAdded/changed line count of the filtered diff.
critical-countNumber of critical findings after dedup.
warning-countNumber of warning findings after dedup.
info-countNumber of info findings after dedup.
issue-urlURL of the created issue, if any.
fix-pr-urlURL of the created draft fix PR, if any.
provider-failuresComma-separated provider names that failed to produce a valid review.
provider-successesComma-separated provider names that produced a valid review.
provider-skipsComma-separated provider names skipped because credentials or supported auth modes were not provided.