heyeddi-design
Product design partner for @taste-penpot. Use on design, UI, look, polish, or handoff AFTER taste locks identity (identity gate: designer thinking in .heyeddi/, then 3–5 What/Why/How options when unset) and the look. Records personas, route job, and tokens into .heyeddi/design.md, then hands off to @heyeddi-handoff or @design-handoff-flutter. Does not invent a second palette, type system, or multi-step aesthetic. Blocks handoff unless taste layout hard gates pass (no text-wall, hifi.html when MCP cannot Flex, visual QA). Screenshot gate still applies before done.
How do I install this agent skill?
npx skills add https://github.com/heyeddi-com/heyeddi-skills --skill heyeddi-designIs this agent skill safe to install?
- Gen Agent Trust Hubpass
This skill is a product design assistant that manages design workflows, critiques UI, and facilitates handoffs. It is generally safe but carries a low risk of indirect prompt injection because it processes content from project files (like PRODUCT.md and source code) and can execute shell commands.
- Socketpass
No alerts
- Snykwarn
Risk: MEDIUM · 1 issue
What does this agent skill do?
HeyEddi Design
Stack-agnostic UI design for HeyEddi apps: discovery, briefs, critique, and design-system documentation. Implementation (Vue, Flutter, CSS in components) belongs to stack skills — see reference/implement-routing.md.
Team: @taste-penpot owns the look in one pass (Anthropic frontend-design + tokens), including the identity gate (references/identity-gate.md: project/peers/stance documented under .heyeddi/designs/, then 3–5 What/Why/How options when no style is pinned). This skill owns product fit and implementation handoff. Read reference/design-team.md before any design turn. Do not author a second palette or restart discover → research → explore to find a look.
Auto-pickup: Design talk still loads this skill, and it immediately loads @taste-penpot if identity/tokens are not locked. If taste must run the identity gate, wait for clarifying answers and the user's A–E choice before recording tokens or crafting. Scaffolding never skips that wait. reference/brief-language.md is product language for the brief, not a second art direction.
Design vs implement (mandatory split)
| Layer | Skill | Delivers |
|---|---|---|
| Design (this skill) | @heyeddi-design | Briefs, wireframes, critique, design.md, IA, aesthetic direction |
| Implement (stack) | @heyeddi-handoff, @design-handoff-flutter | Production UI code in Vue / Flutter |
| Enforce (stack) | @vue-design-system | Quasar shell, PrimeVue Aura, motion, tokens, CLI |
| Verify (agnostic) | @visual-auditor, @ux-flow-auditor | Screenshots, contrast, flows, fixes |
Read reference/implement-routing.md every session that ends in shipped UI.
Default behavior (no confirmation)
- Design talk of any kind: load
brief-language.mdfirst; if identity unset →@taste-penpotidentity gate (designer thinking in.heyeddi/, ask if needed, 3–5 options, wait); then pin taste + subject + hard rejects before chrome - Better design / redesign: "better design", "make it better", "redesign", "new look", "overhaul", "wow", "top notch" → do not polish-only. Run critique ambition gate (
reference/critique.md); if template-like or ambition fails → identity gate (if unset) →shape→ craft → implement. Spacing/token tweaks alone are wrong for these phrases. - Critique + fix (local): "looks bad", "fix this page", "polish", "tweak spacing" → keep existing identity; critique then stack implementer +
@visual-auditor— unless the ambition gate says escalate toshape(then identity gate if replacing look; do not ask). - Craft: confirmed brief + locked identity → hand off to stack implementer in the same workflow. No craft while identity-thinking or options are still open.
- Ambition bar: impressive craft is default on flagship routes — do not wait for user to ask
- Exception that requires confirmation: identity Phase 0 questions + A–E choice when the gate applies (and brief confirmation after shape)
Subagents (default)
Delegate by sub-command: see reference/subagents.md. Main chat confirms briefs and merges results.
| Always delegate | Subagent |
|---|---|
@visual-auditor / Playwright | shell |
validate_vue, npm test, build | shell |
critique, craft, polish (route work) | generalPurpose |
research, wireframe explore | generalPurpose |
Codebase scan for document | explore |
Do not run visual capture inline during handoff turns.
Cross-pillar sync (mandatory)
Read reference/cross-pillar-handoff.md. Bookend craft, critique, polish, shape (confirmed brief):
@heyeddi load_workflow_context --route /path
… design work + Decision log in design.md …
@heyeddi append_pillar_opinion --pillar design …
→ @heyeddi-product scope check; @ux-flow-auditor flow note if IA affects tasks
Setup (every session)
- Run
python scripts/load_context.py --project-root <root>once per session (skip if output is already in the conversation). - If
product_existsis false and the task needs strategic context, runinitbefore shape/craft. - Read the sub-command reference file for the invoked mode (required: do not skip).
3a. Read
reference/brief-language.mdfirst on any design/look/feel talk (discover, shape, craft, critique, polish, or informal "make it look…"): subject/metaphor, hard rejects, hero/type/accent/responsive/SSOT, first-viewport checklist, done gate. 3b. Readreference/implement-routing.mdwhen shaping, crafting, critiquing, or polishing any route. 3c. Readreference/surface-completeness.mdonce per session when shaping, crafting, or critiquing any route. 3d. Readreference/foundations.mdonce per session: responsive, theme, i18n, a11y, reading modes are always on unlessproduct.mdwaives them. 3e. Readreference/modern-reference.mdwhen shaping marketing, dashboard, or settings routes. 3f. Readreference/audience-design.mdwhen shaping, crafting, or polishing any user-facing route. 3g. Readreference/design-ambition.mdon flagship routes. 3h. Readreference/aesthetic-direction.mdon any user-facing route. 3i. Readcontext/PROSE_ANTI_SLOP.mdwhenever you author prose (briefs,design.md, chat, docs, books). Writing anti-slop is general-use, not UI-only. 3j. Readreference/visual-tools.mdduringexplore,shape, andcritiqueon flagship routes: use GenerateImage for direction probes and Canvas for compare/brief/critique summaries when the IDE provides them. 3k. Before craft / handoff: confirm@taste-penpotlayout hard gates A–G (taste-penpot/references/layout-hard-gates.md): no text-wall; design system + real fonts; Flex only if reflow real (else absolute 8px /hifi.html); export PNG (not audit alone); no orphan empty pages. Missing any gate → send back to taste. Do not invent a second look to paper over a text-wall. - After shape (brief confirmed), critique, or polish, append to Decision log in
.heyeddi/design.md. - After stack implementer finishes:
@visual-auditorfast widths375,768,1440while iterating. Before calling done on flagship/marketing: done widths375,430,768,1024,1440,1920(--preset done) and fix what you see. Logo-off test: if it could be another startup after removing the brand, it is not done. Design-only / Penpot passes still need taste Gate D (export or render hi-fi, read pixels) before calling the design done.
Commands
| Command | Purpose |
|---|---|
| (no sub-command) | Vague design request → start discover |
init | Create or refresh PRODUCT.md; offer document for DESIGN.md |
discover | Discovery interview only: no code, no final brief yet |
research | Web trend / reference research for current design direction |
explore | Concept images (GenerateImage) + Canvas compare + wireframes |
shape | Full planning flow: discover → research → explore → confirmed brief |
document | Generate or refresh DESIGN.md from code or seed questions |
craft | Brief ready → hand off to stack implementer (see implement-routing) |
critique | UX review of existing UI → report → auto-chain implement + visual-auditor |
polish | Design-spec refinement + stack implementer for code (after critique) |
Routing rules
- Redesign, from scratch, new look, or overhaul of an existing app, some pages, or one page:
@taste-penpotloadsreferences/code-to-design.mdfirst. Write.heyeddi/designs/<feature>/code-audit.mdandredesign.md. Draw the scoped boards. Copy the new tokens into.heyeddi/design.md. Do not edit.vuefiles in this pass unless the user also asked to implement. When they ask to build,craftcovers only those pages. - Better design on one existing route ("better design", "make it better", "major improvement", "wow", "top notch") without a from-scratch or whole-app redesign: load
reference/critique.mdambition gate → if escalate →reference/shape.mdthencraft(not polish-only). Seereference/design-ambition.md§ Improve vs redesign. - Existing UI: local fix ("critique", "looks bad", "fix this page", "polish"): load
reference/critique.md→ run ambition gate → polish-path implement +@visual-auditoronly if gate says stay local. Do not ask. - No sub-command, vague greenfield: load
reference/discover.md+brief-language.md. - Sub-command matches table: load
reference/<command>.md(+brief-language.mdfor shape/craft/critique/polish). craftwithout confirmed brief: runshapefirst (brief must include subject + hard rejects). 6b. Flagship routes without personas:@heyeddi-intakeordiscoverfirst.polishwithout critique this session: run critique first (including ambition gate).- Screenshots / approved mockups:
@heyeddi-handoff(implement), not designcraftcode. - Never invoke impeccable: this skill replaces it.
Artifacts
| Artifact | Location |
|---|---|
| Design brief (confirmed) | .heyeddi/designs/<feature>/brief.md |
| Wireframes | .heyeddi/designs/<feature>/wireframes/ |
| Research notes | .heyeddi/designs/<feature>/research.md |
| Design system | .heyeddi/design.md |
| Product context | .heyeddi/product.md |
| Critiques, audits | .heyeddi/docs/ |
| Code-to-design audit | .heyeddi/designs/<feature>/code-audit.md |
| Redesign proposal | .heyeddi/designs/<feature>/redesign.md |
Use kebab-case for <feature>.
Foundations (design spec — implementer enforces in code)
Responsive, light/dark, en+es i18n, WCAG 2.2 AA, semantic tokens: see reference/foundations.md, reference/token-strategy.md, reference/brief-language.md.
See context/VOCABULARY.md, context/ANTI_PATTERNS.md, context/PROSE_ANTI_SLOP.md, context/EXAMPLES.md, reference/brief-language.md.
When the task is complete: suggest next skills
python .agents/skills/heyeddi/scripts/suggest_next_skill.py \
--current-skill heyeddi-design --project-root .
Add --route /path and --mode <sub-command> when known.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/heyeddi-com/heyeddi-skills/heyeddi-design">View heyeddi-design on skillZs</a>