Src App Skill

glyphfield / src/app

Purpose

Own the Next.js route surface: Studio/docs pages, API handlers, metadata/OG, machine-readable resources, and global visual foundations.

Mental model & key files

• studio/page.tsx mounts the interactive client Studio.
• docs/** renders the Fumadocs shell; content comes from content/docs through src/lib/docsSource.ts and src/components/DocsMdx.tsx.
• api/agent, api/catalog, api/labs, api/materials, api/identities, api/elements, and api/generate expose public agent contracts.
• api/docs/[[...slug]] returns one processed Markdown page.
• llms-full.txt/route.ts concatenates the same processed docs corpus.
• openapi.json returns OPENAPI_DOCUMENT; api/search powers docs search.
• globals.css and docs/docs.css own shared chrome and documentation visuals.

Patterns to follow / invariants

• Route handlers are thin. Validation and domain behavior live in src/lib.
• Discovery responses are cacheable briefly; generation is no-store.
• Agent routes use the shared CORS headers and return structured errors.
• Docs Markdown routes and /llms-full.txt must derive from docsSource, not from a separately maintained copy.
• Add accessible titles/descriptions and keep docs navigation metadata in sync.

Common tasks → first action

• Add an endpoint → define the contract/domain behavior in src/lib, add a thin route, update manifest/OpenAPI/docs, then test the contract.
• Add a docs page → create MDX, add it to the section meta.json, use shared media components, run pnpm docs:generate.
• Change docs layout → inspect desktop/mobile behavior and the custom 1px scrollbar.
• Add machine docs → extend docsSource-backed routes instead of copying prose.

Gotchas

Do not import browser-only components into route handlers. Do not add undocumented agent endpoints. Do not make /llms.txt and human docs disagree; the former is a router and /llms-full.txt is the complete derived corpus. Preserve explicit content types and X-Content-Type-Options on text/Markdown responses.