nashif/AI PR Review

AI-powered pull request review, code suggestions, and vulnerability scanning via OpenRouter.

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
openrouter_api_keyOpenRouter API key. Store it as a repository secret.yes
github_tokenToken used to read the pull request diff and post comments.no${{ github.token }}
modelOpenRouter model slug, e.g. anthropic/claude-sonnet-4.5, openai/gpt-4.1, google/gemini-2.5-pro, deepseek/deepseek-chat.noanthropic/claude-sonnet-4.5
fallback_modelsComma-separated model slugs tried in order if the primary model fails.no""
base_urlOpenRouter API base URL.nohttps://openrouter.ai/api/v1
temperatureSampling temperature for the review model.no0.1
max_tokensMaximum tokens in each model response.no8000
excludeNewline- or comma-separated glob patterns for files to omit from the diff sent to the model (lockfiles, vendored code, generated output).no**/*.lock **/*.min.js **/*.min.css **/package-lock.json **/yarn.lock **/pnpm-lock.yaml **/poetry.lock **/Cargo.lock **/go.sum **/*.svg **/*.png **/*.jpg **/*.jpeg **/*.gif **/*.pdf **/*.ico **/vendor/** **/node_modules/** **/dist/** **/build/** **/__snapshots__/** **/*.generated.* **/*_pb2.py
max_diff_charsTotal diff characters sent to the model before chunking kicks in.no180000
chunk_charsApproximate size of each diff chunk when a PR is too large for one request.no60000
max_chunksCap on the number of chunks reviewed for very large PRs.no8
include_file_contextShow the model the real source surrounding each changed hunk, read from the pull request's head commit. Turn this off to review the raw diff alone.notrue
include_repo_overviewPrepend a short map of the repository: description, layout, manifests, README excerpt.notrue
context_linesLines of real source shown above and below each changed hunk.no30
max_context_charsCharacter budget for surrounding-code context per diff slice.no60000
context_sourceWhere to read surrounding code from. `api` fetches each file at the head commit and is always correct. `workspace` reads the checked-out tree, which saves API calls but is only safe when the workflow checked out the pull request's head.noapi
skip_labelsComma-separated labels that suppress the review when present on the PR.nono-ai-review,skip-review
required_labelsIf set, the review only runs when the PR carries one of these comma-separated labels.no""
skip_draftSkip review while the pull request is a draft.notrue
post_commentPost (and update in place) a summary comment on the pull request.notrue
post_inline_commentsAlso attach findings as inline review comments on the changed lines.nofalse
comment_tagHidden marker used to find and update this action's previous comment.noai-pr-review
fail_on_severityFail the step when a finding at or above this severity survives review. One of: none, low, medium, high, critical.nonone
min_scoreFail the step when the overall score (0-100) falls below this value. Empty disables the check.no""
project_contextExtra repository context handed to the model (conventions, architecture, review priorities).no""
languageNatural language for the review prose.noEnglish
namedescription
scoreOverall score from 0 to 100.
verdictOne of approve, comment, request_changes.
findings_countTotal number of findings across all categories.
highest_severityHighest severity observed, or none.
comment_urlURL of the posted summary comment.
skippedtrue when the review was skipped by label, draft, or empty-diff gating.