Skills for agents

Snapds can export your component contract as skill docs so coding agents use your design system correctly — without re-reading source or burning context tokens.

Dictionary + on-demand detail

A tiny, always-loaded index lists your components and links to per-component detail that is loaded only when needed. Context windows stay small while the agent still has the full contract a click away.

Supported agents

Select any combination via snapds.skills.formats. Each writes to the location and file format that agent expects, and each keeps a main dictionary/router file that snapds sorts into the first card position (badged router) in the Settings UI.

Context-efficient by design

The structure adapts to how each agent loads context:

• Claude, Augment load a component’s skill only when it’s used (folder-based progressive disclosure).
• Cursor, Windsurf keep a tiny always-on router and load each component rule on demand via its description / model_decision trigger.
• Copilot, Cline have no lazy loading, so snapds writes a single catalog file that inlines every component instead of many separate always-loaded files.

Every layout carries each component’s full contract — description, import, usage, and a complete Props table — including the consolidated catalogs.

Compact consolidated catalogs

Because Copilot and Cline load their single catalog on every request, a large design system can make that file heavy. When one of them is selected, the AI tab shows a Compact catalog toggle (snapds.skills.compactConsolidated): enable it to drop the prop tables from those catalogs — keeping import + usage only — so the always-loaded file stays small. It has no effect on the other agents, whose per-component detail is loaded on demand and always keeps the full props table.

Destinations

Set snapds.skills.destination. The chosen root is shared by every selected agent — each still writes to its own subpath (.claude/skills, .cursor/rules, …) underneath it.

• workspace — the repo root. Committable and team-shared; the right choice for most repos.
• subfolder — a path relative to the repo root via subPath (e.g. apps/web), for monorepos where an agent runs from a specific app/package.
• custom — any absolute folder via customPath — e.g. ~ to write personal agent skills (~/.claude/skills) shared across all your projects, or an ignored scratch folder to keep generated docs out of git.

Generating

• Snapds: Generate Skills — interactive; lets you pick one or more agents and a destination, then writes the full set for every selected component.
• Snapds: Regenerate All Skills — non-interactive; reuses your saved snapds.skills config and rewrites the index and all detail files.

Per-component guidance

Add free-text guidance under snapds.skills.instructions, keyed by component id (pkg#Name). It’s appended verbatim as an Additional guidance section:

{
  "snapds.skills": {
    "instructions": {
      "@acme/ui#Button": "Prefer `variant=\"primary\"` for the main page action."
    }
  }
}

Incremental auto-generation

Enable Auto-generate (autoGenerate: true, the default) and Snapds regenerates the index and only the new detail files as your selection changes — existing files are left untouched, never the whole tree.