agentleFS
danielmachado60danielmachado60/claude-setup/.claude/skills/design-direction/SKILL.md

design-direction

Skill1 starsChanged 4 months agoMITView on GitHub

Art direction before the first screen — real references, 3 DISTINCT visual directions for the user to choose from, tokens with personality and anti-generic-design rules. Use PROACTIVELY on a new project with UI (or a redesign) BEFORE creating tokens or designing the first screen; also when the user asks for "art direction", "site visual identity", or complains the design looks generic / "AI-looking".

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

5 rules. 0 are common, 4 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask the user: is there a brand?

    line 17
  2. Distinctive:

    No default-AI purple/blue gradient as identity.

    line 65
  3. Very specific:

    No display type in Inter/Roboto/system without a justification recorded in direction.md.

    line 66
  4. Distinctive:

    No emoji as icons on a client site.

    line 67
  5. Distinctive:

    No grid of 3 identical cards with soft shadows as the default solution for every section.

    line 68

“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 — kill the generic before the first screen

Claude without direction produces competent, forgettable design: purple gradient, Inter, identical cards with a soft shadow. This skill decides the LOOK of the project before any token or screen. Output: docs/design/direction.md + tokens — the ui-designer and the dev execute on top of them.

Step 1 — Inputs

  1. ABOUT.md: segment, audience, tone of voice (the visual direction translates the verbal tone).
  2. docs/discovery.md if it exists (what the audience is like, what competitors look like).
  3. Ask the user: is there a brand? (logo, mandatory colors, brand font, brand manual?) Constraints outrank preferences — record what is non-negotiable.

Step 2 — Real references (not from memory)

Collect 4–6 VISIBLE references (Playwright for screenshots; without Playwright, WebFetch + honest description):

  • 2–3 from the niche that perform (find via discovery/competitors — including to DIVERGE from them).
  • 2–3 from outside the niche with high craft — search galleries (awwwards, godly.website, land-book, minimal.gallery) for adjacent segments.

From each one note: what to steal (type pairing, spacing rhythm, use of color, a layout pattern) and what to avoid. A reference without notes is decoration.

Step 3 — Propose 3 DISTINCT directions

Each direction is a coherent package. The three must genuinely diverge — three variations of blue are not three directions. Format per direction:

  • Name (e.g. "Sober editorial", "Soft brutalism", "Classic premium")
  • Intent — 1 paragraph: what feeling it delivers and why it fits THIS audience/tone
  • Palette — 3–5 colors with hex and role (dominant, signature, surface, text). Rule: 1 dominant + 1 memorable signature color. No rainbow.
  • Type pairing — display + text, real and available fonts (Google Fonts is fine). Display with personality: Inter/Roboto/Open Sans/Lato/Montserrat only as TEXT, never as display without a written justification.
  • Distinctive layout rule — the structural signature: giant hero typography, grid with a deliberate break, full-bleed photo, hard borders with no shadow, alternating solid-color sections... ONE clear rule the dev can follow.
  • Anchor reference — which of the Step 2 references anchors this direction.

Present via AskUserQuestion (one option per direction, with a textual preview). Hybrids are allowed ("palette from 1 with typography from 3") — record the final choice.

Step 4 — Materialize the chosen direction

  1. docs/design/direction.md: the full direction (intent, palette, typography, layout rule, references with notes, what is FORBIDDEN in this project).
  2. Tokens in the active pack's format (custom properties in static, project tokens in nextjs): colors by role, type scale with the display font, spacing, radius, shadow — all consistent with the direction (a "hard borders" direction with border-radius: 12px is incoherence, not a detail).
  3. From here on, precedence applies: direction.md > one-off screen preference. Direction changed mid-project? Update the document first, screens after.

Anti-generic rules (apply to ANY direction)

  • No default-AI purple/blue gradient as identity.
  • No display type in Inter/Roboto/system without a justification recorded in direction.md.
  • No emoji as icons on a client site.
  • No grid of 3 identical cards with soft shadows as the default solution for every section.
  • Minimum 1 personality element PER PAGE (expressive typography, signature color used boldly, characteristic shape/pattern) — without turning into a circus: 1 is the floor, 3 is the ceiling.
  • AA contrast always (personality never justifies illegible text).
  • Generic stock photos (handshake, smiling call center) don't go in; prefer real product, real results, or custom illustration/composition.

Process anti-patterns

  • Jumping straight to the screen "because the deadline is tight" — direction takes 1 hour and saves 3 rounds of "I don't like it".
  • 3 directions that are the same one with a swapped palette.
  • Choosing the direction yourself without presenting all 3 — the choice belongs to the user/client.
  • Copying the anchor reference 1:1 — a reference is a starting point, not a template.

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.