Skills

The same concepts as the library, written for your coding agent. They are plain files: your agent reads them while it builds and reviews, and nothing runs until it does. They work in Claude Code, Codex, Cursor and most other agents.

Install

npx skills add kausthubh-coder/critly -g

npx skills add kausthubh-coder/critly

npx skills update -g

Install globally so every project gets updates. A copy vendored inside one project goes stale; the skill notices and tells you once.

The skills

• critly

  Build and review. Picks the concepts that match the work, applies their exact values, runs a taste pass on every screen it touches, and reviews with a table: what it was, what it should be, and the concept behind each change.

  Ask: “Review the settings page with critly.”

• critly-audit

  Audit a whole app from a URL, the code, or both: the system, the composition, the details and the speed. It writes the app's design policy and sets up the check.

  Ask: “Audit my app at localhost:3000 and fix what you find.”

Keep your agent on it

Agents read a skill once and rarely come back to it. AGENTS.md is in context for every task, whether or not the agent picks the skill, and it survives when a long session is compacted. In our tests, agents with these two lines ran the check every time and finished with no errors; agents with only the skill never ran it. The check's setup adds them for you.

- UI work: use the `critly` skill. Read `.critly/policy.json` before choosing spacing, type or color, use only its values, and never edit it to pass.
- After any UI change, run `npm run lint:ui` and fix every error until it exits 0. Exit 2 means it couldn't run; say so, it is not a pass. Then run `npm run taste:ui`, open the squint and overlay images for each changed screen, and fix each look or say why the screen needs it before saying it's done.