Design Orchestrator
The single entry skill for design work. It locates the task on the intent × deliverable
map and delegates to one of 9 specialist spokes. The cross-cutting model every design task
shares — research and constraints before generation, the decision ledger, and the
anti-AI-slop quality gate — lives in design-core; read it before generating any pixels.
Routing map (intent → spoke)
Design a real product UI (default path)
- Any interface, landing page, dashboard, redesign, or visual polish →
refero-design (research is mandatory — reference before implementation)
- Need styles / palettes / font pairings / chart types per stack →
ui-ux-pro-max (searchable design-intelligence DB)
- Want a specific, opinionated look applied →
swiss-design (Swiss/modernist system in Tailwind)
Engineer the interface to a senior bar
- Metric-based rules, component architecture, motion, hardware-accelerated performance →
taste-skill
- Emit a reusable, agent-enforceable spec file (anti-generic standards) →
stitch-design-taste (produces DESIGN.md)
Run a design-agent workflow
- Iterate drafts / branch variations / extend to more pages from a component →
superdesign
Theme an existing artifact
- Apply or generate a theme for slides, docs, reports, or HTML →
theme-factory
Produce static visual art
- Poster, art piece, print-ready
.png / .pdf → canvas-design
Pull brand identity from the web
- Extract logos, colors, backdrops, brand name from a URL →
openbrand
Standard Operating Flow
- Locate the task on intent × deliverable: is the output a product UI, a spec, a themed artifact, static art, or brand data?
- If the output is a product UI or anything user-facing, pull the gate from
design-core first — research → lock references → ledger decisions → generate → check against the slop gate. Skipping research is the failure mode.
- Delegate to the spoke(s). Multi-step asks fan out in deliverable order (e.g. "brand this landing page" →
openbrand for tokens → refero-design for the layout → taste-skill for the build).
- Return: chosen spoke(s), the references/constraints that anchored the work, the deliverable type, and the next action.
Guardrails
See design-core. In short: research before generation — never average from the model's
defaults; ground every choice in a concrete reference and record it in the decision ledger.
Keep type scale, spacing, and color on a system, not ad-hoc. Treat "make it look good" as a
request to find references and commit to a direction, not to improvise. The cluster's value
is craft that resists generic AI output — don't quietly fall back to the default look.
Loading spokes on demand
To keep CLI startup context lean, this cluster's spokes are not separately registered as
skills — only this orchestrator and its *-core are enumerated. When you route to a spoke
named above, load it on demand by reading its file:
~/.agents/skill-clusters/skills/<spoke-name>/SKILL.md (or skills/<spoke-name>/SKILL.md inside the skill-clusters repo).
1---2name: design-orchestrator3description: Route a UI/visual-design task to the right skill among 9 design specialists — research-first methodology, the design-intelligence database, concrete style systems, senior engineering rules, spec emitters, design-agent workflows, theming, static art, and brand extraction. USE WHEN a user is designing, styling, redesigning, or polishing an interface, artifact, or brand surface but hasn't named the specific skill or approach.4---56# Design Orchestrator78The single entry skill for design work. It locates the task on the **intent × deliverable**9map and delegates to one of 9 specialist spokes. The cross-cutting model every design task10shares — **research and constraints before generation**, the decision ledger, and the11anti-AI-slop quality gate — lives in `design-core`; read it before generating any pixels.1213## Routing map (intent → spoke)1415**Design a real product UI (default path)**16- Any interface, landing page, dashboard, redesign, or visual polish → `refero-design` *(research is mandatory — reference before implementation)*17- Need styles / palettes / font pairings / chart types per stack → `ui-ux-pro-max` *(searchable design-intelligence DB)*18- Want a specific, opinionated look applied → `swiss-design` *(Swiss/modernist system in Tailwind)*1920**Engineer the interface to a senior bar**21- Metric-based rules, component architecture, motion, hardware-accelerated performance → `taste-skill`22- Emit a reusable, agent-enforceable spec file (anti-generic standards) → `stitch-design-taste` *(produces DESIGN.md)*2324**Run a design-agent workflow**25- Iterate drafts / branch variations / extend to more pages from a component → `superdesign`2627**Theme an existing artifact**28- Apply or generate a theme for slides, docs, reports, or HTML → `theme-factory`2930**Produce static visual art**31- Poster, art piece, print-ready `.png` / `.pdf` → `canvas-design`3233**Pull brand identity from the web**34- Extract logos, colors, backdrops, brand name from a URL → `openbrand`3536## Standard Operating Flow37381. Locate the task on **intent × deliverable**: is the output a *product UI*, a *spec*, a *themed artifact*, *static art*, or *brand data*?392. If the output is a product UI or anything user-facing, pull the gate from `design-core` first — **research → lock references → ledger decisions → generate → check against the slop gate.** Skipping research is the failure mode.403. Delegate to the spoke(s). Multi-step asks fan out in deliverable order (e.g. "brand this landing page" → `openbrand` for tokens → `refero-design` for the layout → `taste-skill` for the build).414. Return: chosen spoke(s), the references/constraints that anchored the work, the deliverable type, and the next action.4243## Guardrails4445See `design-core`. In short: **research before generation** — never average from the model's46defaults; ground every choice in a concrete reference and record it in the decision ledger.47Keep type scale, spacing, and color on a system, not ad-hoc. Treat "make it look good" as a48request to *find references and commit to a direction*, not to improvise. The cluster's value49is craft that resists generic AI output — don't quietly fall back to the default look.5051## Loading spokes on demand5253To keep CLI startup context lean, this cluster's spokes are **not** separately registered as54skills — only this orchestrator and its `*-core` are enumerated. When you route to a spoke55named above, **load it on demand** by reading its file:5657`~/.agents/skill-clusters/skills/<spoke-name>/SKILL.md` (or `skills/<spoke-name>/SKILL.md` inside the skill-clusters repo).
Sheshiyer