From ATELIER — Design studio & adversarial UI review
Verify ATELIER's external tool dependencies are installed and reachable. Trigger with /atelier:check (or "check atelier prerequisites"). Runs a fast ✓/✗ probe grouped by tier; advisory by default, --strict to fail on a missing required tool. Reads the canonical manifest skills/check/requirements.tsv.
How this skill is triggered — by the user, by Claude, or both
Slash command
/atelier:checkclaude-haiku-4-5The summary Claude sees in its skill listing — used to decide when to auto-load this skill
ATELIER reviews with built-in vision (the `Read` tool sees PNGs — no API key) and drives live browsers via
ATELIER reviews with built-in vision (the Read tool sees PNGs — no API key) and drives live browsers via
the Playwright MCP. Its tool surface is small; this confirms the few it uses are present. It installs
nothing — it reports and points at install hints.
bash ${CLAUDE_PLUGIN_ROOT}/skills/check/scripts/check.sh # advisory ✓/✗ table
bash ${CLAUDE_PLUGIN_ROOT}/skills/check/scripts/check.sh --strict # exit 1 if a REQUIRED tool is missing
requirements.tsv is the single source of truth. The shipped Playwright MCP
(.mcp.json) and the committed-snapshot crawler both launch via npx/node; a Chromium browser is needed
for live crawls (the MCP downloads its own on first use; the crawler uses the target project's Playwright).
gh is recommended so self-improve can open the improvement PR. User-flow rendering is pressroom's
job via /publish — verify those with /pressroom:check, not here. Companions (pressroom, foundry,
ideator) are plugins, referenced by capability, not probed.
npx claudepluginhub agentic-underground/idea-to-productionProvides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Searches MemPalace before answering questions about past work, people, projects, or prior decisions. Returns verbatim stored content instead of guessing from model memory.