agentleFS
w2urw2ur/claude-code-setup/skills/design-direction/SKILL.md

design-direction

Skill2 starsChanged 2 months agoMITView on GitHub

Load before designing a new site or page, or redesigning one, when no DESIGN.md exists yet.

8 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If the project already has a DESIGN.md or a design section in its CLAUDE.md, it has a direction: follow it, and use design-critique instead.

    line 15
  2. Very specific:

    Write these down first, in this order, before a colour, a font or a layout is considered (the headings match …

    line 28
  3. Very specific:

    Read (its best for column).

    line 42
  4. Very specific:

    Run references/dna.js: pass the file's whole text as the function argument of browser_evaluate.

    line 51
  5. Distinctive:

    Every proposed direction must differ from each neighbour on at least one named axis: ground, type family, layout structure, signature.

    line 79
  6. Very specific:

    Write DESIGN.md at the repo root from references/design-md-template.md, filling every section with the chosen direction, the references of …

    line 105

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Design direction — a written direction before any code

A model with no direction lands on its own defaults, and a portfolio of projects designed that way converges on one look. This skill makes the direction a written, reviewable decision: read the subject, study real sites, check the portfolio for convergence, propose 2-3 directions, and record the chosen one in DESIGN.md.

If the project already has a DESIGN.md or a design section in its CLAUDE.md, it has a direction: follow it, and use design-critique instead.

1. Load the ban list

Load design-tells. Its ids are the vocabulary of every step below: read the ids from design-tells itself, never from a copy, and refer to a trap by id, never by restating it. With no stated direction, never default to anything section 1 of design-tells names, nor to B9 — and never pick a direction silently: step 5 always proposes 2-3 and waits.

2. Read the subject before any visual decision

Write these down first, in this order, before a colour, a font or a layout is considered (the headings match references/design-md-template.md):

  • Read — one line: Reading this as: <page kind> for <audience>, with a <vibe> language.
  • Domain — at least 5 concepts, objects or gestures of the subject.
  • Colour world — at least 5 colours that exist in the subject, each with where it is seen. Tokens come later, from these.

A colour or a font that cannot be traced back to this list is a default wearing a reason.

3. Reference study

Pick 2-4 sources from the table in references/inspiration.md that match the Read (its best for column); never one from its Excluded list (zero-cost rule). For each site studied:

  1. Open it with the Playwright MCP (browser_navigate; follow the row's how to read it, including any wait), then browser_take_screenshot. The MCP writes screenshots only under its allowed roots (~/Dev and ~/Dev/.playwright-mcp); with no filename they land in ~/Dev/.playwright-mcp/. Do not request a path outside those roots — the call is refused.
  2. Run references/dna.js: pass the file's whole text as the function argument of browser_evaluate. It returns the computed fonts, sizes, weights, heading tracking, ground and text colours by area, radii, section spacing and content width. For a local file or a build the MCP will not open, run ~/.claude/scripts/design/dna-harness.mjs <url|file> [light|dark] (exit 2 = could not run; never read that as an empty DNA).
  3. Note one thing to take and one to refuse, each tied to the screenshot or a DNA value.

fonts is the computed font-family stack, not proof the face loaded; check the screenshot. A clip-text: entry under text colours is gradient text (B9).

4. Convergence check

List every direction already in the portfolio — DESIGN.md files, and design or identity entries of project CLAUDE.md files (a heading, or a bold bullet such as the hub's "Deliberate identity: the paper palette"):

find ~/Dev -maxdepth 2 -name DESIGN.md
find ~/Dev -maxdepth 2 -name CLAUDE.md -exec grep -H -n -i -E \
  '^(#+ .*(design|identity|palette)|[-*] \*\*[^*]*(identity|palette|design system|design direction))' {} + || true

Both commands print nothing, without error, when there is no match. Read each hit in its file; skip one that is a rule list rather than a project's own direction, and say why. Then:

  • Every proposed direction must differ from each neighbour on at least one named axis: ground, type family, layout structure, signature.
  • Where several neighbours share a value on one axis, differ on that axis first: a shared value is the portfolio's convergence, not a safe choice.
  • Then ask of each direction: would a similar prompt for another project land here too? If yes, revise that part — it came from the model, not the subject.

5. Propose 2-3 directions, then wait

Each direction carries:

  • Tokens — ground, ink, muted, rule, accent; hex, light and dark.
  • Type — families with their role, and the scale.
  • First screen — an ASCII sketch of the layout at desktop width.
  • Signature — one owned device from the Domain list, used once (C18).
  • Refused defaults — at least 3 design-tells ids it refuses.
  • Convergence — one line per neighbour from step 4: the axis it differs on.

The directions differ from each other on at least one axis too; three shades of one idea are one direction. A warm light ground is allowed only as an explicit, named choice the owner can refuse (G1). Stop and wait for the owner's pick. No code, no DESIGN.md, before it.

6. Record the direction

Write DESIGN.md at the repo root from references/design-md-template.md, filling every section with the chosen direction, the references of step 3 and the convergence lines of step 4 (Differs from). Then load claude-md-hygiene first (it governs every CLAUDE.md edit), and add one line to the design section of the project CLAUDE.md (create the section if there is none) pointing to DESIGN.md as the project's identity. From then on, design-tells and design-critique treat what it names as sanctioned.

Motion standards: read ~/.claude/skills/review-animations/SKILL.md directly; it is not model-invocable, so do not try to load it as a skill.

7. Build, then critique

Build the page, then load design-critique in fix mode on the rendered result: one fix round, one re-check, stop. Report what the critique found, what was fixed, and what remains.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.