PR Auto Review Multi-Agent System

基于 uv + LangGraph + FastAPI + Celery 的 PR 自动化 Code Review 系统。

Quick Start

1. 安装依赖

uv sync

2. 配置环境变量

cp .env.example .env

3. 启动 API

uv run uvicorn apps.api.main:app --reload --host 0.0.0.0 --port 8000

4. 启动 Worker

uv run celery -A apps.worker.celery_app:celery_app worker -l INFO

5. 回放 webhook（本地联调）

uv run python scripts/replay_webhook.py --payload samples/pull_request_opened.json --secret "$GITHUB_WEBHOOK_SECRET"

Architecture

FastAPI(Webhook) -> Celery -> LangGraph -> GitHub Comment/Label/Check

Tool agents (security/readability/performance/test) now run in a ReAct loop with bind_tools + ToolNode for tool-calling orchestration. Reflexion memory is persisted in SQLite and injected into future agent runs. Runtime control adds trace/span logging, per-step latency metrics, token budget, LLM concurrency quotas, and dynamic prompt trimming.

Notes

• Worker 只扫描 PR 变更的 Python 文件，避免全仓扫描造成噪音和性能问题。
• Parser / Supervisor / Consensus 不执行 PR 代码，仅做 API/LLM 逻辑。