CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Common Commands

Build: pnpm run build - Compiles TypeScript and packages with ncc
Package: pnpm run package - Alias for build command
Lint: pnpm run lint - ESLint with TypeScript support, must have 0 warnings
Type Check: pnpm run type-check - TypeScript compilation check without output
Test: pnpm run test - Runs Jest test suite with full coverage
Manual Review: pnpm run review - Runs ts-node on ai-review.ts for local testing

This is a GitHub Action that provides AI-powered code review for Pull Requests using multiple LLM providers (OpenAI, Anthropic, Google Gemini).

src/index.ts: GitHub Action entry point that validates inputs, generates git diffs, and delegates to ai-review.ts
src/ai-review.ts: Main review engine with hunk-based diff processing, LLM integration via Vercel AI SDK, and GitHub API comment posting
src/system-prompt.ts: Configurable system prompt builder with default code review guidelines and XML response format enforcement
types.ts: Type definitions for Finding objects (path, line, message, severity, suggestion, code)

Provider Agnostic: Uses Vercel AI SDK (ai package) for unified LLM access across OpenAI, Anthropic, and Google
Hunk-Based Processing: Splits diffs into individual hunks rather than arbitrary size limits for focused reviews
Structured XML Output: Enforces consistent XML response format with severity levels (required, optional, nit, fyi)
Line Resolution: Maps XML findings to actual GitHub line numbers through diff parsing
Retry Logic: Built-in exponential backoff for LLM API failures
Custom Prompts: Supports custom review prompts while preserving XML response format

The action expects these environment variables (set automatically by GitHub Actions):