github.devotd.
NewDeveloper tools

redhat-et/ripwire v0.6.7 — a repository map for coding agents

Covers redhat-et/ripwire v0.6.7's repository map and change-impact analysis features, plus the limits of the evidence behind its claimed reduction rates.

Repositoryredhat-et/ripwire ↗

The ripgrep of AI context: a zero-dependency C++23 CLI + MCP server for coding agents. Find what you want without reading the repo, then check you built what you meant — blast radius, tests-to-run, quality deltas. Signatures at 74.7% fewer bytes than bodies; every guess labelled, every loss published. Paddle out with a map.

stars
★ 2.5K
Forks
157
Language
C++
License
Apache-2.0
Latest release
v0.6.7
Read original source ↗View on GitHub ↗

redhat-et/ripwire is a C++23 CLI and MCP server that helps coding agents find the code they need before reading a repository's body. It builds a repository map with no external dependencies, and the latest release is v0.6.7. The provided release information does not list any items newly changed in this version.

The core of it is adding call relationships and change impact to code search. According to the README, it uses a name-based call graph to propose the scope affected by a change and the tests to run, and after an edit it shows only the metrics that got worse among twelve quality indicators. It also points out files that should be modified together and names that can be interpreted in multiple ways. It bundles 26 third-party grammars, including CUDA kernel launch sites and PHP dynamic dispatch, and states a determinism contract — the same tree producing the same byte sequence every time — as four rules. It is licensed under Apache-2.0.

The README's comparison table reports that signature bundle output uses 74.7% fewer bytes than the body text, and only 5.0% of the tokens compared with grep-then-read. Looking only at questions both approaches answered, that figure is 5.2%. Separate from these numbers, the workflow multipliers and per-query reduction claims cited in the document rest on operators' field reports, with no independent verification. So the repository map and impact analysis can be treated as documented features, but the actual size of the reduction is hard to assert as verified performance.