redhat-et / ripwire

项目摘要

Ripwire is a zero-dependency C++23 CLI and MCP server that provides AI coding agents with a fast, deterministic map of a codebase — call graphs, blast radius analysis, test-to-code mapping, and quality deltas — enabling agents to navigate and modify code without reading entire repositories. It compresses signatures at 74.7% fewer bytes than full bodies and labels every inference, publishing all uncertainty. Designed as the 'ripgrep of AI context,' it serves as both a pre-edit research accelerator and a post-edit verification gate.

为什么值得关注

Ripwire addresses a critical bottleneck in AI-assisted coding: the token cost and error rate of agents reading entire codebases. By providing ranked call graphs, blast radius prediction, forgotten co-change detection, and per-edit quality deltas through a lightweight zero-dependency binary, it transforms agent workflows from brute-force reading to targeted navigation. Its field report from a 20-agent orchestration on a 1,500-file C++/Metal codebase claims 2× whole-workload speedups and 10–25× per-lookup savings, demonstrating load-bearing value at scale.

优势

• Zero runtime dependencies with a single C++23 binary — trivial to install and distribute
• Dual CLI + MCP server design integrates with any coding agent that speaks MCP
• Deterministic call graph with blast radius, forgotten co-changes, and quality deltas provides actionable agent guidance
• Signature compression at 74.7% fewer bytes than bodies reduces token spend for agent context
• Active development: 20 releases in 6 months, 29 contributors, CI, and tests

局限性

• Very new project (created July 2026) with limited production track record
• No Docker support or examples directory may slow adoption for some teams
• Quality delta analysis limited to eleven kinds — may not cover all code quality dimensions teams care about

使用场景

• Pre-edit research: agents query call graphs and blast radius before modifying code to understand impact
• Post-edit verification: agents check quality deltas, forgotten co-changes, and test coverage after writing code
• Multi-agent orchestration: fleet coordination where shared context maps prevent redundant research and silent divergence
• Token budget optimization: compressed signatures reduce context window consumption for large repos