paulrberg/coderabbit

CodeRabbit Review Triage

Fetch and analyze CodeRabbit review comments from a GitHub PR, or run a local CodeRabbit CLI review. Default mode fetches from the remote PR. Use --local to run the CLI against local diffs.

Arguments

Parse $ARGUMENTS for:

• <pr-number> — integer PR number
• <pr-url> — GitHub PR URL (e.g., https://github.com/owner/repo/pull/123), extract the PR number from the path
• --local — switch to local review mode (see below)

Mode Dispatch

If $ARGUMENTS contains --local, strip it and forward remaining arguments to references/local-review.md. Follow that workflow instead of continuing here. All flags after --local (--base, --base-commit, --type, --config) are forwarded to the local workflow.

Otherwise, continue with the PR comment analysis workflow below.

Prerequisites

Run these checks in order. Stop at the first failure.

1) GitHub CLI Authenticated

gh auth status

If not authenticated, stop and tell the user to run gh auth login.

2) Detect Repository

gh repo view --json owner,name --jq '"\(.owner.login)/\(.name)"'

Store the result as {owner}/{repo}.

3) Resolve PR Number

• If a PR number was provided as an argument, use it directly.
• If a PR URL was provided, extract the number from the URL path.
• If neither was provided, auto-detect from the current branch:

gh pr view --json number --jq '.number' 2>/dev/null

If auto-detection fails, stop: "No open PR found for the current branch. Provide a PR number or URL, or use --local for local review."

4) Validate PR

gh pr view {pr_number} --json state,url --jq '"\(.state) \(.url)"'

If the command fails, the PR does not exist. Report and stop. Otherwise, note the state (OPEN/MERGED/CLOSED) and URL for the report scope section.

Workflow

1) Fetch CodeRabbit Comments

Fetch from three GitHub API endpoints, filtering for the CodeRabbit bot. Use --paginate for large PRs. The bot's login is coderabbitai but may appear as coderabbitai[bot], so match with startswith.

Walkthrough/summary comments (issue comments where CodeRabbit posts its walkthrough):

gh api "repos/{owner}/{repo}/issues/{pr_number}/comments" \
  --paginate \
  --jq '[.[] | select(.user.login | startswith("coderabbitai"))]'

Review objects (top-level review body and state):

gh api "repos/{owner}/{repo}/pulls/{pr_number}/reviews" \
  --paginate \
  --jq '[.[] | select(.user.login | startswith("coderabbitai"))]'

Inline review comments (file-level comments with path and line info):

gh api "repos/{owner}/{repo}/pulls/{pr_number}/comments" \
  --paginate \
  --jq '[.[] | select(.user.login | startswith("coderabbitai"))]'

If no CodeRabbit comments are found across all three endpoints, report "No CodeRabbit review found on PR #{pr_number}" and stop.

2) Parse Comments into Findings

Process the raw API responses into a normalized list of findings:

• From inline comments (pulls/{pr}/comments): Each comment becomes a finding with path (from .path), line (from .line or .original_line), body (the comment text), diff_hunk (surrounding context), and source: "inline". These are the primary source of actionable findings.
• From review bodies (pulls/{pr}/reviews): Parse each review's body for actionable items. Extract individual suggestions. Mark source: "review".
• From issue comments (issues/{pr}/comments): Parse for file-level summaries and actionable items within the walkthrough. Mark source: "walkthrough".

Inline comments are the primary source of actionable findings. Review summaries and walkthrough comments provide context and may surface additional actionable items not tied to specific lines.

3) Triage Findings

Load references/triage.md and follow the shared triage process — categorize, classify, confirm ambiguous, generate fix plan, and report.

Stop Conditions

Stop and ask for direction when:

• gh CLI is not authenticated.
• The PR does not exist or is not accessible.
• No CodeRabbit comments found on the PR.
• Auto-detection fails and no PR number was provided.
• A finding requires architectural changes beyond the PR's scope.
• Classification confidence is low for code on critical paths (auth, payments, data integrity).