AI legibility

lantern-ui ships everything a coding agent needs to build with it correctly — a model-readable reference, a rules block for your AGENTS.md, installable skills, and a linter that fails on the mistakes agents make most.

llms.txt

llms.txt is a one-screen rule list plus a one-line-per-component catalog; llms-full.txt adds every attribute table. Regenerated from the component registry, so it cannot drift.

mix lantern.llms           # regenerate (inside the lantern_ui package)
mix lantern.llms --check   # fail CI when stale

Rules agents follow

• Build with lantern components in HEEx — never reach for React or a JS component library.
• One flat list: a status column on each row, filter chips with counts. No group headers.
• Never hand-roll a table or button where table/1, data_table/1 or button/1 exist.
• Semantic tokens only — no palette classes, no arbitrary pixel values.
• Title and actions live in the breadcrumb bar; no tabs as a default grouping mechanism.
• Copy the recipe HEEx; never hand-roll rows, glyphs, rings or rails.

Install into your app

Copies the skills and writes the rules block (with a fresh component catalog) into AGENTS.md.

mix lantern_ui.install_skills
mix lantern_ui.install_skills ../my-app --force

Lint

Fails on bypassed tokens, banned grouped-list markup, hand-rolled tables/buttons and unknown component or attribute names — with did-you-mean hints.

mix lantern.lint
mix lantern.lint --format json

Skills

Four skills ship with the package: lantern-ui-components, lantern-recipes, lantern-migration and phoenix-page-design.