brooks-lint — AI Code Review That Cites the Canon

Author: Björn Schotte (LinkedIn) URL: https://www.linkedin.com/posts/bjoernschotte_aicodereview-harnessengineering-claudecode-share-7469996798825246720-YgXD/

What It Is

An AI code review tool called brooks-lint — every finding cites the book it came from: Brooks, Fowler, Martin, Feathers — twelve software engineering classics in all.

How It Works

It distills those books into:

  • 6 production-code decay risks — cognitive overload, change propagation, dependency disorder, accidental complexity
  • 6 test-suite risks — (matching decay patterns)

Each finding lands in a structured format:

Symptom → Source → Consequence → Remedy

Tagged with a severity label and a 0–100 health score.

Architecture Audit Mode

Draws a Mermaid dependency graph, color-coded by severity.

Platform Support

Ships as:

  • Claude Code plugin
  • Codex CLI skill
  • Gemini CLI extension
  • GitHub Action

License: MIT Stars: 532 in ten weeks

The Benchmark

Across three review scenarios:

  • With brooks-lint: structured and sourced findings 94% of the time
  • Without (same model unguided): 16% of the time

The Insight

“The model already knows the canon. The harness makes it cite the canon — every finding, every run. On AI review the bottleneck was never detection. It’s whether you can trust the same answer twice.”