claude-reviewAI-powered GitHub PR review agent

Automated code review as a GitHub App, backed by Anthropic Claude, OpenAI, or any compatible LLM.

Features

Multi-Provider LLM Support

Use Anthropic Claude (hosted or API-compatible), OpenAI, Ollama, vLLM, Together, Groq, LiteLLM, or any OpenAI-compatible endpoint.

Anti-Instruction Injection

Randomized fence delimiters, canary token verification, content sanitization, and suspicious pattern detection defend against prompt injection attacks.

Rule Engine

A full expression language with boolean logic, comparisons, string operations, and regex matching drives per-file and per-PR review directives.

Sandbox Isolation

Linux bubblewrap (bwrap) and Windows Job Objects restrict subprocess execution with configurable filesystem and network policies.

MCP Tool Integration

Connect external tools via the Model Context Protocol over stdio, HTTP, SSE, or WebSocket transports.

Rate Limiting and Whitelists

Per-repo token bucket rate limiting with user bypass, plus organization, user, and repository whitelists control access.