Agents

How coding agents (and humans pasting into agents) should load Sometic context without inventing a visual UI kit.

Copy Prompt

Surface docs expose a Copy Prompt button (dashed outline). It copies a plain-text scaffold brief for that area:

SurfacePagePrompt file
EcosystemHome (hero CTA)ecosystem.txt
App ShellApp Shellapp-shell.txt
AuthAuthenticationauth.txt
HTTPHTTPhttp.txt
HeadHead / SEOhead.txt
QueryQueryquery.txt
FormsFormsforms.txt
StoresStoresstores.txt
ThemingThemingtheming.txt
FoundationPrimitivesfoundation.txt

Paste the prompt into your agent, then point it at the matching docs URL and package names.

llms.txt

  • Curated index: /llms.txt (also linked in site head as rel="alternate")
  • Full consumer export: /llms-full.txt (regenerated via python scripts/generate-llms-full.py)
  • Short pointer file: /ai.txt

Prefer llms.txt for routing; use llms-full.txt for offline / IDE ingestion.

Package graph (mental model)

text
Adapters (@sometic/react, vue, elements, …)
    → Composition (@sometic/app-shell)
        → Features (forms, auth, http, query, theme, head, dom, …)
            → Foundation (core, events, store, styling, accessibility, …)
  • App Shell composes System packages behind one session epoch and dispose().
  • Query caches server data; store holds client/UI state; do not mix.
  • HTTP owns transport and 401 refresh; auth owns session; query refetches after re-auth.
  • Prefer subpath imports (@sometic/react/query, @sometic/react/button).
  • Custom elements use the sometic-* prefix.
  • Publishable packages are font-agnostic; brand fonts are docs and demo surfaces only.

Start paths

  1. Architecture: layers and guarantees
  2. What’s included: honest inventory
  3. Comparison: when not to use Sometic
  4. System hubs: App Shell, Auth, HTTP, Query, Forms, Stores, Head