Repowise
Repowise is a self-hosted platform that reads a codebase and produces the kind of institutional knowledge a new hire would otherwise spend weeks piecing together: which files are risky to touch, who actually owns them, what documentation is stale, and why a given design decision was made. It targets two audiences at once — human engineers doing code review, and AI coding agents that need grounded context before making a change.
The tool scores every file for defect risk, maintainability, and performance using 49 deterministic checks, covering things like excessive method complexity, low class cohesion, "god classes," and repeated database query patterns across files. Rather than stopping at a warning, it proposes concrete refactoring moves — extracting a class, moving a method, breaking a dependency cycle — along with an estimate of how much of the codebase that change would touch.
A second layer mines git history rather than source code. It tracks which files change together, who has effectively "owned" a file based on commit activity, how concentrated that ownership is (a bus-factor signal), and which files are hotspots for bug fixes. This behavioral data feeds into a risk score for incoming changes.
A third layer generates and maintains documentation automatically, rebuilding module and file-level wiki pages on every commit and marking them with a freshness indicator so stale docs are visible rather than silently rotting.
For AI agents specifically, Repowise exposes an MCP server with a get_why() function that answers "why was this built this way," tracing each answer back to a specific, verbatim piece of source evidence and labeling how confident that trace is.
Repowise runs locally — pip install repowise followed by repowise serve starts an MCP server and a local dashboard, and source code never leaves the machine it runs on. A hosted managed version exists for teams that would rather not operate it themselves.
Repowise is placed in Assess. It is a legitimate, actively maintained project — over 6,000 GitHub stars and commits as recent as the day this entry was written — and the local-first, self-hosted deployment model removes the usual "will our code leave our servers" objection that stalls evaluation of this category of tool. But placement here reflects the wiki's ring gate, not the tool's maturity: no one has run Repowise against a real codebase and confirmed its risk scores or refactor suggestions were actually useful. That first-person trial is the specific gap between Assess and Trial. Position is set to center — trying it costs one pip install and pointing it at an existing repository, which is a low-friction way to validate the claims.
