version: "1.0.0" name: coderabbit description: 'Review pending changes with the CodeRabbit CLI: review the branch diff and working tree, fix findings judged real, dismiss false positives with justification, then run the quality gate. Use for CodeRabbit reviews of local changes.' metadata: aidd-category: runtime

CodeRabbit Review

Run the CodeRabbit CLI against the target application's pending changes, triage every finding, fix the ones that are real, and dismiss the rest with a written justification. CodeRabbit is an external cloud review service: the CLI sends the diff to CodeRabbit for analysis, so only use this skill on projects whose owner is comfortable with that.

Usage

coderabbit [app] [base-ref]

If [app] is omitted, infer from the current working directory. [base-ref] is the branch to compare against; when omitted, use main if it exists (locally or as origin/main), otherwise master locally or as origin/master.

aidd's execution intent controls whether confirmed remediations are applied or only described. The external review still uploads the diff under either intent.

Setup

1. Check if the CodeRabbit CLI is available:

• Run coderabbit --version to test.
• On Windows the official CLI does not run natively; if the native probe fails, try WSL:

wsl -e coderabbit --version. If WSL has it, run every subsequent CodeRabbit command through WSL against the project's mount path, e.g. wsl -e sh -c "cd /mnt/d/path/to/project && coderabbit review --plain" (translate D:\path\to\project to /mnt/d/path/to/project).

1. If neither probe succeeds, STOP and report exactly what is missing instead of failing

cryptically. Setup guidance to include in the report:

• Install (macOS/Linux/WSL): curl -fsSL https://cli.coderabbit.ai/install.sh | sh
• Log in: coderabbit auth login (interactive; the user must do this themselves)
• On Windows, both steps happen inside WSL.

1. Verify the CLI is logged in by checking coderabbit auth status (through WSL if applicable).

If not logged in, STOP with the same guidance; never attempt the interactive login yourself.

Workflow

Step 1: Resolve scope

• Determine the base ref: use the [base-ref] argument if provided; otherwise main (local or

origin/main), else master (local or origin/master). If none exists, review only uncommitted changes.

• Confirm there is something to review (git status --porcelain and

git log <base>..HEAD --oneline). If both are empty, report "nothing to review" and stop.

Step 2: Run the review

• With a resolved base ref, run coderabbit review --plain --base <base>.
• With no base ref, run coderabbit review --plain so the review covers the uncommitted changes

without passing a nonexistent ref.

• This is a long-running command: reviews commonly take several minutes. Run it in the

foreground with a generous timeout (at least 15 minutes) and wait for it to complete; do not background it.

• Capture the full output. If it must be persisted for reference, write it to the OS temp

directory, never into the reviewed working tree.

• If the CLI reports a rate limit (the free tier has daily limits), report that and stop; do not

retry in a loop.

Step 3: Triage every finding

For each finding, judge it against the actual code (read the file, do not trust the excerpt alone):

• Real: the issue exists and matters. Queue it for a fix.
• False positive or not applicable: dismiss it with a one-line justification (e.g. "guarded

two lines above", "intentional per project convention X").

Respect the repository's own conventions when judging: a pattern the project uses deliberately is not a finding.

Step 4: Fix the real findings

• Apply fixes for every finding judged real, smallest change that resolves the issue.
• Do not refactor beyond the finding's scope.

Step 5: Quality gate

• If the project's package.json defines a smoke:qc script, run bun run smoke:qc.
• Otherwise run the project's relevant build, typecheck, lint, format-validation, and test scripts

as defined in its package.json.

• Fix any breakage your changes introduced before finishing.

Step 6: Summary

Report:

• Findings fixed (or proposed under review-only intent), with the file and a one-line description

for each

• Findings dismissed, each with its justification
• Quality-gate result

Notes

• CodeRabbit's free tier has daily review limits and paid plans meter usage per reviewed file;

surface any limit errors to the user rather than working around them.

• This skill never pushes, opens PRs, or posts review comments. Apart from uploading review input

to CodeRabbit as disclosed above, its only project mutation is editing the working tree.