SuperAPI AI Directory

spin-design

Design & creation

View repository

Updated 10-15 20:11

About this skill

spin-design

Shorthand for "prepare a design subfolder within this project." It packages everything a Claude Design session needs — a basic design package plus a README and a paste-ready brief — with the specific objectives inferred from the current conversation, then points the user at it to start an open-ended design session.

The model for the output folder is the proven CIA/publishing/Guide Abrege/_claude-design-prep/ package (README + DESIGN-PROMPT + manuscript/asset references + reference images).

When invoked

/spin-design [optional: subfolder name or short topic]

What to do

1. Resolve the project + target

  • The project = the folder the conversation is about (the cwd, or the path most recently discussed). Confirm if ambiguous.
  • The design target = what we're designing (cover, book interior, landing page, brand system, a single page, a deck…). Infer it from the conversation — do not ask if the chat already makes it obvious.

2. Create the subfolder

  • Default location: <project>/design-<slug>/ where <slug> comes from the arg or the inferred target (e.g. design-cover, design-interior, design-landing). If the project already has a design-prep convention (_claude-design-prep/, design/), match it.
  • Never overwrite an existing folder — if it exists, add a dated sibling or ask.

3. Populate it (the "basic design package")

Write these files, filling them with specifics inferred from the conversation — not generic boilerplate:

  • README.md — what this folder is; the design target in one line; the inferred objectives (a tight bulleted list pulled from what the user actually said in chat — aesthetic direction, what to keep, what to change, constraints, audience, format/trim); what's in the folder; and the one-line "how to start" (open a Claude Design session here / paste DESIGN-PROMPT.md).
  • DESIGN-PROMPT.md — the paste-ready, open-ended kickoff prompt for the Claude Design session. State the goal, the inferred objectives, the constraints, the references included, and "what to produce" — but leave room for the designer to explore (open-ended, not a rigid spec). End with the deliverables list.
  • Copy in the real package — duplicate (don't symlink) the relevant existing assets so the folder is self-contained:
    • current design artifacts if any (existing cover/cover.jpg, *.css, current PDF/page, brand files),
    • the content the design serves (manuscript excerpt, copy, product description),
    • reference images discussed in the chat (e.g. a photo to incorporate, a style reference),
    • any house style / typography notes. If an asset is large or lives elsewhere, copy it in anyway (self-contained beats tidy) unless it's huge (>50MB) — then reference its path in the README.

4. Hand off

Report: the folder path, a one-line summary of the inferred objectives, the list of files created, and the next step ("Start a Claude Design session pointed at this folder, or paste DESIGN-PROMPT.md"). Offer to open the README in Typora.

When the design comes back (you're handed a claude_design project to implement): see memory reference_claude_design_mcp_roundtrip — render_preview→serve_url→headless-Chrome --print-to-pdf for a faithful PDF; read_file output is entity-escaped + <x-dc>/<helmet>/support.js-wrapped (unescape once + strip to edit locally); and write_files base64 friction means edit + render LOCALLY for the deliverable, don't try to sync the full design back.

Rules

  • Infer, don't interrogate. Pull objectives from the conversation. Ask at most one clarifying question, and only if the target is genuinely unclear.
  • Self-contained. The folder should let a fresh Claude Design session start cold with no other context.
  • Open-ended brief. The DESIGN-PROMPT states goals + constraints + references, then invites exploration — it is not a pixel spec.
  • Don't design here. This skill assembles the handoff; the actual design happens in the Claude Design session it kicks off.