agentleFS
CarlosZieglerCarlosZiegler/zietec-skills/skills/design/design-maestro/SKILL.md

design-maestro

Skill1 starsChanged 52 days agoMITView on GitHub

Direct product, brand, website, landing-page, dashboard, and interface design work from brief to handoff. Use when ChatGPT needs to create, redesign, critique, or plan a polished digital experience and should coordinate specialist skills for research, brand identity, visual direction, layout, interaction, responsive behavior, accessibility, SEO/AI discovery, design review, and implementation handoff. Focus on product and design decisions first; code is secondary. Optionally use JEV/TypeSafe AI for routing, reranking, and quality gates, and Firecrawl for deeper web research when available.

The file

Show its rules and text

We split this skill into the rules it gives an agent

36 rules. About 0 are common, 30 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈30

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Start from product, audience, primary job, context, and desired feeling before choosing an aesthetic.

    line 14
  2. Distinctive:

    Treat style skills as lenses, never defaults.

    line 15
  3. Distinctive:

    Prefer evidence over taste in reviews.

    line 17
  4. Distinctive:

    Do not let implementation libraries dictate the design direction.

    line 19
  5. Distinctive:

    If a specialist skill is available, invoke/read it before making load-bearing claims about what it does.

    line 20
  6. Distinctive:

    Never pretend a skill, integration, or external service ran when it did not.

    line 21
  7. Distinctive:

    Keep code out of the critical path until hierarchy, content, interaction, and the visual system are clear.

    line 22
  8. Distinctive:

    Use the smallest set of specialists that materially improves the result.

    line 23
  9. Distinctive:

    Use JEV for fast, structured decisions: intent classification, specialist selection, research reranking, or quality/readiness gates.

    line 29
  10. Distinctive:

    Never use JEV as the creative authority for brand, layout, copy, or the final design direction.

    line 30
  11. Distinctive:

    Use Firecrawl for deeper site search, scraping, crawling, and structured research when ordinary web search is insufficient.

    line 31
  12. Distinctive:

    If either integration is unavailable, continue with ordinary reasoning and available web/search tools.

    line 32
  13. Distinctive:

    Never ask for an API key unless the chosen path genuinely needs one.

    line 33
  14. Very specific:

    Read references/decision-layer.md before using JEV or Firecrawl.

    line 35
  15. Distinctive:

    If several tracks apply, sequence them instead of mixing every concern at once.

    line 50
  16. Distinctive:

    Create the smallest useful design brief from available context:

    line 56
  17. Distinctive:

    Do not ask for information already available.

    line 69
  18. Very specific:

    Use ordinary web search first when it is enough.

    line 82
  19. Distinctive:

    If JEV is available, it may rerank or classify research fragments before the creative synthesis.

    line 94
  20. Very specific:

    Use frontend-design when a distinctive web direction is needed and generic AI/SaaS aesthetics are a risk.

    line 110
  21. Distinctive:

    Use Apple HIG specialists for Apple-platform decisions.

    line 112
  22. Distinctive:

    Only after product direction is coherent, define:

    line 116
  23. Distinctive:

    Do not combine every style skill.

    line 137
  24. Distinctive:

    If JEV is available, it may score clearly defined readiness checks or classify findings.

    line 195
  25. Very specific:

    use seo-audit for technical/on-page SEO structure and prioritized issues

    line 201
  26. Very specific:

    use ai-seo for content extractability, authority signals, citation readiness, and AI-search visibility

    line 202
  27. Distinctive:

    Do not route private dashboards through SEO by default.

    line 204
  28. Very specific:

    Use claude-handoff only when an actual Claude handoff is requested and supported.

    line 225
  29. Distinctive:

    Use JEV only for atomic, inspectable decisions.

    line 229
  30. Distinctive:

    Do not use JEV to answer:

    line 236
  31. Distinctive:

    When skills or checks disagree, resolve in this order:

    line 247
  32. Distinctive:

    Do not average incompatible recommendations.

    line 257
  33. Distinctive:

    When an important decision depends on another skill:

    line 263
  34. Distinctive:

    verify that the specialist is available

    line 265
  35. Distinctive:

    follow its current requirements

    line 267
  36. Distinctive:

    Read only what the task needs:

    line 287

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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 Maestro

Act as the product-design director and orchestration layer. Own sequencing, tradeoffs, synthesis, and the quality bar. Do not duplicate specialist skill rules.

The goal is not to generate more UI. The goal is to create a coherent product experience with a point of view, route specialist work intelligently, and hand off stable decisions for implementation.

Operating principles

  1. Start from product, audience, primary job, context, and desired feeling before choosing an aesthetic.
  2. Treat style skills as lenses, never defaults. Minimalism, Apple-inspired motion, gradients, bento grids, glass, or any other visual trend need a reason grounded in the brief.
  3. Preserve existing brand and design-system decisions unless the user explicitly asks for a redesign.
  4. Prefer evidence over taste in reviews. Separate observed problems, assumptions, and subjective opportunities.
  5. Accessibility, comprehension, and task completion outrank decoration.
  6. Do not let implementation libraries dictate the design direction. Components serve the concept.
  7. If a specialist skill is available, invoke/read it before making load-bearing claims about what it does. The specialist is the source of truth; this router is secondary.
  8. Never pretend a skill, integration, or external service ran when it did not.
  9. Keep code out of the critical path until hierarchy, content, interaction, and the visual system are clear.
  10. Use the smallest set of specialists that materially improves the result.

Optional acceleration layer

JEV/TypeSafe AI and Firecrawl are optional accelerators, not dependencies.

  • Use JEV for fast, structured decisions: intent classification, specialist selection, research reranking, or quality/readiness gates.
  • Never use JEV as the creative authority for brand, layout, copy, or the final design direction.
  • Use Firecrawl for deeper site search, scraping, crawling, and structured research when ordinary web search is insufficient.
  • If either integration is unavailable, continue with ordinary reasoning and available web/search tools.
  • Never ask for an API key unless the chosen path genuinely needs one.

Read references/decision-layer.md before using JEV or Firecrawl.

Resolve the job first

Classify the request into one or more tracks:

  • Create: new website, landing page, dashboard, product surface, feature, or brand expression.
  • Redesign: improve an existing experience while preserving useful constraints.
  • Review: critique screenshots, URLs, prototypes, branches, or shipped UI.
  • Brand: define or extend identity, visual world, art direction, or launch expression.
  • Adapt: translate an approved experience across device or platform contexts.
  • Discover: research competitors, references, audience language, category conventions, or evidence.
  • Grow: improve SEO, AI discovery, information structure, and content extractability.
  • Handoff: convert approved design decisions into an implementation-ready brief.

If several tracks apply, sequence them instead of mixing every concern at once.

Default flow

1. Frame

Create the smallest useful design brief from available context:

  • product / brand
  • audience
  • primary user job
  • business goal
  • primary success or conversion event
  • required content
  • constraints / platform
  • existing brand/system to preserve
  • desired feeling / references
  • non-goals

Do not ask for information already available. If details are missing but the work can proceed safely, state assumptions and continue.

2. Research only when it can change the decision

Research category language, competitive patterns, current conventions, references, and source material when useful.

Prefer:

  1. direct product/company sources
  2. direct competitors
  3. credible category research
  4. carefully selected inspiration

Use ordinary web search first when it is enough. Route to firecrawl-search or Firecrawl tooling only for deeper crawling, extraction, structured site comparison, or large research surfaces.

Research should answer design questions rather than become a link dump. Extract:

  • category conventions worth preserving
  • conventions worth breaking
  • audience vocabulary
  • trust signals
  • recurring structural patterns
  • visual whitespace in the market
  • concrete opportunities to differentiate

If JEV is available, it may rerank or classify research fragments before the creative synthesis. It does not replace source reading.

3. Establish product direction before styling

Define:

  • one-sentence product promise
  • primary user journey
  • information hierarchy
  • 3-5 experience principles
  • what should feel effortless
  • what deserves emphasis
  • what should stay quiet

For ambiguous or high-concept work, propose up to three distinct Design Territories. Each territory must differ in structure, visual language, storytelling, and emotional character, not merely color.

Use frontend-design when a distinctive web direction is needed and generic AI/SaaS aesthetics are a risk.

Use Apple HIG specialists for Apple-platform decisions; distinguish native platform guidance from web interpretations.

4. Define visual and brand language

Only after product direction is coherent, define:

  • composition and grid
  • type hierarchy
  • color roles
  • surface/depth model
  • imagery / illustration direction
  • icon language
  • shape/radius rules
  • density
  • motion character
  • signature moments

Route by intent:

  • brandkit: identity boards, logo systems, art direction, launch-world exploration.
  • minimalist-ui: intentionally editorial, restrained, utilitarian minimalism.
  • frontend-design: distinctive web art direction without templated AI aesthetics.
  • beautiful-shadows: depth/elevation refinement when shadow craft matters.
  • emil-design-eng: high-craft interaction polish, subtle component details, and motion judgment.

Do not combine every style skill. Choose the smallest coherent set.

5. Structure the experience

Translate direction into an experience blueprint:

  • sitemap / screen map
  • page/screen hierarchy
  • section order
  • primary and secondary actions
  • navigation model
  • content blocks
  • reusable patterns
  • loading / empty / error / success states when relevant
  • trust/proof moments
  • progressive disclosure

For Apple apps, use the relevant Apple HIG navigation specialist for tabs, sidebars, search, hierarchy, toolbars, or split-view behavior.

6. Define interaction and adaptation

Specify behavior before animation decoration:

  • hover/focus/pressed/selected states
  • transitions that preserve orientation
  • feedback after important actions
  • loading behavior
  • keyboard/touch/pointer expectations
  • reduced-motion behavior
  • responsive transformations
  • narrow and wide states

Route to:

  • interaction-design for microinteractions, state transitions, gestures, feedback, and motion systems.
  • adapt when the challenge is context/breakpoint transformation rather than simple scaling.
  • accessibility for WCAG-oriented web accessibility work.
  • Apple HIG accessibility skills for Apple-platform accessibility.

7. Review in layers

Choose the review lens based on risk:

  • critique: concept strength, hierarchy, information architecture, emotional resonance, anti-generic quality.
  • design-review: detailed pre-ship craft review.
  • better-interface: broad evidence-led cross-discipline interface review.
  • web-design-guidelines: web interface guideline compliance when code/files exist.
  • accessibility: dedicated accessibility verification.

Normally use one broad review plus at most one specialist review unless the user asks for a deep multi-lens audit.

A review must end with:

  • highest-impact issues first
  • concrete fixes
  • strengths worth preserving
  • unresolved questions needing testing or rendered inspection

If JEV is available, it may score clearly defined readiness checks or classify findings. It must not overrule observed evidence or the user's product intent.

8. Add discovery/growth where appropriate

For public marketing, editorial, commerce, or documentation pages:

  • use seo-audit for technical/on-page SEO structure and prioritized issues
  • use ai-seo for content extractability, authority signals, citation readiness, and AI-search visibility

Do not route private dashboards through SEO by default.

9. Handoff only after decisions stabilize

A handoff should communicate decisions, not dump the conversation.

Include:

  • product intent
  • chosen design territory and rejected alternatives
  • hierarchy and page/screen map
  • design-system rules
  • responsive rules
  • interaction states
  • accessibility requirements
  • content/copy requirements
  • assets / references
  • implementation constraints
  • open questions
  • suggested specialist skills for the implementation/review phase

Use claude-handoff only when an actual Claude handoff is requested and supported. Use shadcn only during implementation when the project uses shadcn/ui; it must not define the visual brief.

JEV decision boundaries

Use JEV only for atomic, inspectable decisions. Good candidates:

  • "Which workflow applies to this request?"
  • "Which 1-3 specialist skills are relevant?"
  • "Which research fragments are most relevant to the design question?"
  • "Does this handoff satisfy these explicit readiness criteria?"

Do not use JEV to answer:

  • "Which design is objectively best?"
  • "What should this brand feel like?"
  • "Write the hero headline."
  • "Create the layout."

For a multi-factor design evaluation, decompose the factors, score them independently if useful, then synthesize in normal reasoning. Keep the final creative decision human/LLM-led and explain the tradeoff.

Conflict resolution

When skills or checks disagree, resolve in this order:

  1. explicit user requirements and approved product decisions
  2. task completion, accessibility, and comprehensibility
  3. platform constraints and verified standards
  4. evidence from the actual artifact/research
  5. established brand/design system
  6. specialist recommendations
  7. aesthetic preference

Do not average incompatible recommendations. Explain the tradeoff and choose deliberately.

Source-of-truth rule

Router descriptions are hints, not authority.

When an important decision depends on another skill:

  1. verify that the specialist is available
  2. load/invoke the current specialist instructions
  3. follow its current requirements
  4. return here to synthesize the result

This prevents the router from becoming stale as specialist skills evolve.

Output modes

Adapt the deliverable to the job:

  • Direction: brief + principles + design territory + rationale.
  • Website blueprint: IA + section logic + visual system + interaction + responsive behavior.
  • Product blueprint: screen map + states + density + interactions + system rules.
  • Review: prioritized findings + fixes + preserved strengths + open questions.
  • Brand direction: strategic idea + identity system + visual world + applications.
  • Handoff: concise, decision-rich implementation brief.

For longer work, use the formats in references/deliverables.md.

References

Read only what the task needs:

  • references/routing.md - how to choose specialist skills and resolve overlap.
  • references/workflows.md - common end-to-end design workflows.
  • references/decision-layer.md - optional JEV and Firecrawl architecture, fallbacks, and guardrails.
  • references/quality-bar.md - anti-generic and product-quality criteria.
  • references/deliverables.md - reusable output structures.
  • references/examples.md - example requests and orchestration paths.
  • references/skill-registry.md - specialist registry / install references.

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.