Show HN: RepoReaper – AST-aware, JIT-loading code audit agent (Python/AsyncIO)

Summary

RepoReaper 💀 Harvest Logic. Dissect Architecture. Chat with Code. 基于 AST 深度解析 · 双语适配的自治型代码审计 Agent English • 简体中文 👇 Live Demo Access / 在线体验 👇 ⚠️ Public Demo Limitations: Hosted instances use shared API quotas. If you encounter rate limits (403/429), please deploy locally for the best experience. ⚠️ 演示环境说明: 中国用户请使用 SEOUL SERVER。在线服务使用共享 API 配额。如遇请求受限或响应缓慢，强烈建议 Clone 本地运行 以获取无限制的极速体验。 An intelligent, agentic system for automated architectural analysis and semantic code search. This project transcends traditional "Chat with Code" paradigms by implementing an autonomous Agent that mimics the cognitive process of a Senior Tech Lead. Instead of statically indexing a repository, the system treats the Large Language Model (LLM) as the CPU and the Vector Store as a high-speed Context Cache. The agent dynamically traverses the repository structure, pre-fetching critical contexts into the "cache" (RAG) and performing Just-In-Time (JIT) reads when semantic gaps are detected. 🚀 Core Philosophy: RAG as an Intelligent Cache In traditional code assistants, RAG (Retrieval-Augmented Generation) is often a static lookup table. In this architecture, we redefine RAG as a Dynamic L2 Cache for the LLM: Cold Start (Repo Map): The agent first parses the Abstract Syntax Tree (AST) of the entire repository to build a lightweight symbol map (Classes/Functions). This serves as the "index" to the file system. Prefetching (Analysis Phase): During the initial analysis, the agent autonomously selects the most critical 10-20 files based on architectural relevance, parses them, and "warms up" the vector store (the cache). Cache Miss Handling (ReAct Loop): During user Q&A, if the retrieval mechanism (BM25 + Vector) returns insufficient context, the Agent triggers a Just-In-Time (JIT) file read. It autonomously tools the GitHub API to fetch missing files, updates the cache in real-time, and re-generates the answer. 🏗 System Architecture & Innovations 1. AST-Aware Semantic Chunking Standard text chunkin...