brooks-lint
brooks-lint is a set of Agent Skills for AI code review. It diagnoses code and architecture through six decay-risk dimensions, returns structured findings, and cites the engineering sources behind each rule. It supports DeepSeek Harness through the standard Skills layout, while its maintainers note that not every supported platform has been run end to end.
Source repository: hyhmrright/brooks-lint
Browse code-review assets See EvoMap integration routes
Reviewing Engineering Quality and Maintainability Risks
Despite the name, brooks-lint is not a syntax linter. It targets maintainability, architecture, test health, debt, and related decay risks. A typical finding follows Symptom > Source > Consequence > Remedy, with a severity label and book citation.
Try requests such as:
Review this pull request for maintainability risks and cite the rule behind every finding.
Review the module boundaries and identify circular dependencies before proposing a fix.
What brooks-lint Actually Checks For
The project provides six main modes: code review, architecture review, debt classification, test review, health scan, and sweep. The sweep can apply safe changes, request confirmation for broader edits, and leave architectural choices for manual judgment.
That scope differs from ESLint or Pylint. Keep conventional linters in the pipeline for syntax and language-specific rules. Use brooks-lint as an AI-assisted review layer, then verify every proposed change with tests and human review.
Compare J-Space for delivery gates and Vibe-Skills for routing review work.
Running It as Part of an Agent Turn
Use a review loop that keeps diagnosis separate from modification:
- Select the narrowest mode that matches the question.
- Provide the diff, files, and architectural context.
- Inspect each finding for a real source location and consequence.
- Approve only the changes you want applied.
- Run the repository's own tests and linters afterward.
Example output contract:
Return a table with severity, file and line, symptom, engineering source, consequence, remedy, and verification command. Do not edit files.
This makes a review easier to challenge. A Health Score is a project-specific summary from the Skill, not an industry certification.
Installing brooks-lint and Configuring Your Rules
For DeepSeek Harness, the installer supports the dsh platform and places Skills under ~/.dsh/skills. Clone or download hyhmrright/brooks-lint, then run the installer from the repository root:
./scripts/install.sh dsh
Use --project if you want the documented project-local behavior. After installation, ask DSH for a code review or invoke the matching command from its / menu. The maintainers verify the installer layout but explicitly distinguish that from end-to-end testing on every platform.
Review-path status checked 2026-09-28: DeepSeek Harness is still in developer preview, with compatibility-breaking changes expected. Recheck Skill discovery before relying on a production review.
Publish Reusable Review Knowledge Through EvoMap
brooks-lint supplies review guidance, while EvoMap separately organizes verified agent assets. After a rule, repair, and test result are approved, a team can explicitly package that experience for reuse. The EvoMap Wiki describes the asset model; CloudBase-AI-Toolkit covers a related cloud workflow.
FAQ
Does brooks-lint replace ESLint or Pylint?
No. The project focuses on engineering and decay risks rather than language syntax and style rules.
Can it change code automatically?
The sweep mode can apply safe fixes and request confirmation for broader work. Review changes, run normal tests, and use J-Space when the task also needs delivery gates.
Is DeepSeek Harness fully maintainer-verified?
The installer and file layout are documented and tested, while the maintainers say only some platforms have been verified end to end. Validate the workflow in a disposable project before relying on it.
Preserve a Verified Engineering Review Rule
Convert a recurring engineering rule into reusable experience only after the finding, repair, and verification result are all traceable.