Skip to content

method.md

Latest commit

 

History

History
30 lines (23 loc) · 1.6 KB

method.md

File metadata and controls

30 lines (23 loc) · 1.6 KB
Raw

The method, in any stack

The code in this repository is TypeScript, but the method works in any stack:

  1. Give the API one source of truth and generate the rest from it. A check regenerates the files and fails if anything changed.
  2. Write each rule the agent must follow as a lint rule or a test, and make its message say what to do instead. An agent can skip a rule that only lives in a Markdown file.
  3. Put numbers on size and complexity. Agents write long functions unless something fails.
  4. Test what review tends to miss: queries per request, accessibility, CSP violations, page performance.
  5. Keep one fast command for all of it, run it on every commit, and have CI run the same thing.
  6. Keep the agent instructions short and point them at the checks. Put multi-step workflows in skills.
  7. Measure changes to the instructions with tasks that have hidden tests.

Tools that do the same jobs elsewhere

Stack Lint and types Boundaries Query budgets
Python Ruff, Pyright import-linter pytest-django's django_assert_num_queries
Rails RuboCop Packwerk Prosopite or Bullet
Go golangci-lint depguard

AGENTS.md and agents/gates.md show how each check is set up here, which is the part worth copying.

React and TypeScript on another stack

The minimal branch has only the gates, on a small client-side app with TanStack Router and Tailwind, and lists the files to copy for each one. It assumes React 19 with the React Compiler and TypeScript 6 or 7.