SuperAPI AI Directory

scaffold-project

Frontend & websites

View repository

Updated 10-16 08:06

About this skill

Step 0 — is the kit here? This skill reads files from the kit. Check once: ls ${CLAUDE_SKILL_DIR}/../../../tokens >/dev/null 2>&1 && echo KIT_OK || echo KIT_MISSING On KIT_MISSING only the skill folders were installed, which is what npx skills add does. Say so plainly, point the user at npx ux-ui-agent-skills init or the plugin install, and stop. Do not guess the contents of a file you could not open.

Generate a new design-project skeleton from ${CLAUDE_SKILL_DIR}/../../../templates/product-design/ into a target directory the user names.

The layout to produce (exactly the reference structure):

<target>/
  # CONTEXT CLAUDE LOADS
  CLAUDE.md              the brief Claude reads every session
  CLAUDE.local.md        your personal prefs, gitignored
  .mcp.json              Figma, Notion, Drive connections
TEAM TOOLKIT

.claude/ rules/ conventions, loaded only when relevant skills/ repeatable workflows, off main context commands/ your custom slash commands settings.json shared permissions, checked into git

YOUR PROJECT

design-tokens.json source of truth: color, type, spacing src/components/ the real UI Claude reads and edits public/images/ real images so prototypes don't break reference/ real screens Claude studies for context


The fast path, when the kit is installed or reachable via npx:

npx ux-ui-agent-skills new


That does steps 2 and 3 below in one call (template copied, `CLAUDE.local.md`
un-suffixed, engine areas installed without overwriting the lean brief). Then pick
up at step 4. The manual steps are spelled out so the same result can be produced
by hand from a clone.

Steps:
1. Ask for the target directory if not given. Create it if it does not exist.
2. Copy every file of `${CLAUDE_SKILL_DIR}/../../../templates/product-design/` into it, preserving the tree.
   Two rules on the copy:
   - `CLAUDE.local.md.template` lands as `CLAUDE.local.md` (the suffix exists
     only so the kit's own gitignore cannot swallow the file).
   - `.gitkeep` stays in `.claude/skills/`, `.claude/commands/`,
     `src/components/`, `public/images/`, and `reference/` so the empty folders
     survive the first commit.
   The template already ships `.gitignore` (which ignores `CLAUDE.local.md`),
   `.mcp.json`, `.claude/settings.json`, and three rules files
   (`components.md`, `tokens.md`, `accessibility.md`).
3. Install the engine next to it so the gates the brief references actually run:

npx ux-ui-agent-skills add tokens components taste accessibility
workflows content frameworks design-systems scripts skills


Use `add`, not `init`: `init` also writes the full engine `CLAUDE.md` over the
lean project brief. The project brief stays short by design, and the engine
knowledge loads from the copied folders when a task needs it.
4. Install the render gates' one dependency in the target: `npm i -D playwright`.
Skipping this is silent: every render gate prints `SKIPPED` and exits 0.
5. Fill in the placeholders in `CLAUDE.md` with the user: product, primary user,
stack, icon set, vocabulary. Leave nothing in angle brackets behind.
6. Point `design-tokens.json` at the brand: replace the `primitive.brand` ramp,
keep the semantic and component tiers, then prove it still passes:

python3 ${CLAUDE_SKILL_DIR}/../../../scripts/validate_contrast.py


7. Remind the user to fill `reference/` with real screens and `public/images/`
with real imagery, and to set the MCP env vars (`FIGMA_API_KEY`, ...) in their
own shell. Never write a secret into `.mcp.json`.

Verify before handing over: the target tree matches the layout above, and

python3 ${CLAUDE_SKILL_DIR}/../../../scripts/validate_template.py


passes against the source template.