reference-driven-design
Skill3 starsChanged 37 days agoMITView on GitHub
Build an evidence-backed visual reference brief before UI concept or polish work. Use when a product needs a modern visual direction, category-aware references, comparative design analysis, or protection from generic AI-generated styling without copying a specific brand.
In this repository21 other files it gives its agents.
- AGENTS.mdAGENTS.md
- starter-kit/AGENTS.mdAGENTS.md
- accessibility-checkSkill
- accessibility-reviewerSkill
- agent-contextSkill
- design-evidence-researcherSkill
- design-reviewerSkill
- design-system-architectSkill
- design-understandSkill
- information-architectSkill
- interaction-designerSkill
- knowledge-graph-architectSkill
- mobile-ux-expertSkill
- motion-designerSkill
- performance-reviewerSkill
- product-designerSkill
- prompt-architectSkill
- responsive-checkSkill
- ux-reviewerSkill
- visual-designerSkill
- visual-qaSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 2 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
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
- Distinctive: line 27
No source, screenshot, product behavior, or graph context is available.
- Distinctive: line 99
No specific brand composition or proprietary asset is copied.
- Very specific: line 115
Use product-designer if product direction or the primary action is unclear.
- Very specific: line 116
Use design-evidence-researcher when current evidence is missing.
- Very specific: line 117
Use visual-designer to convert the brief into hierarchy and token decisions.
- Very specific: line 118
Use motion-designer when continuity or interaction feedback needs movement.
- Very specific: line 119
Use prompt-architect to encode the direction into a graph-backed prompt.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.
Files
Reference-Driven Design Skill
Role
Translate current product and platform references into an abstract, traceable visual direction.
Mission
Improve visual judgment without turning taste, trends, or one admired product into unsupported design rules.
When To Use
- A user asks for a more modern, polished, premium, expressive, or product-specific interface.
- A new surface needs a visual direction before implementation.
- Existing UI feels generic but the workflow and object model are already clear.
- A prompt needs reference categories and explicit anti-copy constraints.
When Not To Use
- Product direction, user goal, or object model is unresolved.
- The task is a narrow visual bug with an existing token and component answer.
- No source, screenshot, product behavior, or graph context is available; record a research gap first.
Inputs
- Product goal, primary action, audience, and platform.
- Current UI or implementation.
- Existing components, tokens, and product design direction.
- Focused graph context from
npm run context. - Supplied references or traceable current product and platform evidence.
Process
- Resolve graph context for the target task.
- Classify the surface by product type, workflow density, platform, and emotional tone.
- Select two or three relevant reference families, not a single brand to imitate.
- For each source, separate observed behavior from interpretation.
- Extract reusable abstractions:
- content and functional layers;
- hierarchy and density;
- typography and shape roles;
- material and color meaning;
- state and motion behavior;
- responsive and input adaptation.
- Map recommendations to existing research, rules, patterns, and tokens.
- Mark unsupported ideas as observations or research needs instead of instructions.
- Define what must not be copied: branding, proprietary assets, distinctive composition, or platform-inappropriate effects.
- Hand the brief to
visual-designer,motion-designer, orprompt-architect.
Output Format
## Reference Brief
- Product category:
- User goal:
- Platform and density:
- Evidence date:
## Reference Matrix
| Source | Observed behavior | Reusable abstraction | Limit |
| --- | --- | --- | --- |
## Visual Direction
- Content layer:
- Functional layer:
- Focal emphasis:
- Typography and shape:
- State and motion:
- Responsive adaptation:
## Graph Traceability
- Research:
- Rules:
- Patterns:
- Missing evidence:
## Anti-Copy Boundary
- Do not copy:
- Platform constraints:
## Handoff
- Next skill:
- Reason:Quality Checklist
- Recommendations come from observable source behavior.
- At least two reference families are compared when the task is exploratory.
- Product purpose determines expression level and density.
- Content and primary action remain dominant.
- Platform-specific materials have fallbacks.
- The brief names unsupported ideas as gaps.
- No specific brand composition or proprietary asset is copied.
Related Research
RESEARCH-00009Expressive System UI In 2026RESEARCH-00010Motion As State ContinuityRESEARCH-00011Contextual And Observable Agentic Interfaces
Related Rules
RULE-00008/VIS-001Semantic Tokens OnlyRULE-00014/VIS-002Keep Expression Subordinate To ContentRULE-00015/UX-004Use Motion To Explain State Continuity
Handoff To Other Skills
- Use
product-designerif product direction or the primary action is unclear. - Use
design-evidence-researcherwhen current evidence is missing. - Use
visual-designerto convert the brief into hierarchy and token decisions. - Use
motion-designerwhen continuity or interaction feedback needs movement. - Use
prompt-architectto encode the direction into a graph-backed prompt.
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
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.

