scaffold-project

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 <target>

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.