Brooks Lint 🤖 Agent Open source

What it does

Brooks Lint is an AI code-review pack that grades code against twelve classic software-engineering books rather than counting lines or cyclomatic complexity. It diagnoses six production-code "decay risks" (cognitive overload, change propagation, knowledge duplication, accidental complexity, dependency disorder, and domain-model distortion) and six matching test-suite risks. Every finding comes back in a fixed Symptom, Source, Consequence, Remedy shape, with a book citation, a severity label, and a 0 to 100 health score, so reviews stay consistent and traceable instead of varying run to run.

What is inside

The pack is a set of Agent Skills, one SKILL.md per mode plus a shared framework folder, exposing six commands: brooks-review for diffs, brooks-audit for architecture with a Mermaid dependency graph, brooks-debt for tech-debt triage, brooks-test for test quality, brooks-health for a composite score, and brooks-sweep for a scan that also applies safe fixes. Behaviour is tuned through a .brooks-lint.yaml file (strictness presets, disabled risk codes, severity overrides, ignore globs, custom risks). A GitHub Action can run it on pull requests, post the review as a comment, emit SARIF for Code Scanning, and fail below a health threshold.

Works with

Claude Code (as a plugin) plus any agent that loads Agent Skills, with an installer covering Codex, Gemini CLI, Cursor, Windsurf, OpenCode, Copilot, Kiro, and more. Claude Code, Gemini CLI, and Codex are maintainer-verified; other platforms are documented but not yet run end to end.

How to install

/plugin marketplace add hyhmrright/brooks-lint
/plugin install brooks-lint@brooks-lint-marketplace

Maintenance and safety

MIT licensed and actively developed. The CI action sends your diff to the Anthropic API (or a custom endpoint you configure), so point it only at hosts you trust with source. Runs cost roughly a few cents per pull request.

Who should use it

Teams that want architectural and test-quality review with cited, repeatable reasoning, layered on top of an ordinary linter rather than replacing it.