Brooks-Lint

Brooks-Lint is an MIT-licensed AI code-review tool with an unusual, opinionated angle: rather than counting lines or cyclomatic complexity, it diagnoses software-quality "decay risks" (six for production code, six for test suites) and grounds each finding in twelve classic engineering books (Brooks, McConnell, Fowler, Martin, Hunt & Thomas, Evans, Ousterhout, Winters et al., Osherove, Whittaker et al., Feathers, Meszaros). Output is structured as "Symptom → Source → Consequence → Remedy" with severity labels, book citations, and a 0–100 Health Score. Six modes: PR Review, Architecture Audit, Tech Debt Assessment, Test Quality Review, Health Dashboard, and Full Sweep (auto-fix). It installs through the Agent Skills standard across Claude Code, Gemini CLI, Codex CLI, OpenCode, Cursor, Windsurf, Antigravity, pi, GitHub Copilot, Kiro, and Factory Droid. Claude Code, Gemini CLI, and Codex CLI are maintainer-verified; the other listed platforms are documented and installer/file-layout tested but not all maintainer-run end to end. CI runs through GitHub Actions with fail-below, severity, and regression gates plus SARIF output for GitHub Code Scanning; project behavior is configured with .brooks-lint.yaml. Language-agnostic. The project documents "~$0.05–0.15 per PR run" as a vendor estimate that depends on diff size and model. Vendor-published benchmark claims, not independently reproduced by Aicoolies: an illustrative three-scenario table reports a "94%" overall pass rate for Brooks-Lint versus "16%" for Claude alone, and 100% structured-finding compliance versus 0%. Separately, the maintainer publishes a frozen 30-report parser benchmark with 30/30 exact severity-count matches and valid SARIF output; that corpus measures report-parsing fidelity, not whether model findings are correct. Brooks-Lint documents no hosted service or product telemetry; analysis content is sent through the configured coding agent/model provider, so that provider's data-handling terms apply.