Fe Design

Frontend Design Skill

Ad-hoc mode — outside the engine, YOU run this loop now

One question decides which mode you are in: is a /implement-ticket run dispatching this?

The third row is an axis, not an owner. Renderer selection resolves through machinery that already exists: search_stack in corpus-grounding (scripts/decision_engine.ts), whose stack corpus carries threejs.csv among its stacks, read beside the register in design-intelligence § Register. No second frontend executor is declared, and none is needed: a renderer changes what you ground against, never who writes the UI.

Outside the engine, nothing else owns the design quality of a UI write. Reading this skill and then writing the UI from priors is the failure it exists to prevent — the loop below is not optional context, it is the work.

Skip only when the change is ui-trivial, decidable off the diff: ≤ 1 file, ≤ 5 changed lines, no new component, no new state, no new dependency. Anything else runs the loop.

The loop

1. Audit first — run existing-ui-audit. What already exists (components, tokens, layout conventions) outranks every heuristic below. Reinventing an existing component is the #1 failure mode.

2. Brief — seven keys before any code. Five are the ones the engine gates (REQUIRED_BRIEF_KEYS): layout, components, states, microcopy, a11y. states means all five of empty, loading, error, success, disabled — a brief missing one is unfinished, not concise.

   Two more are declared, not gated: frequency and initiation. The engine does not halt on them, and that is stated rather than left to be discovered — adding a required key would halt every brief already written against the five.

   Key	Values	Why it is declared and never inferred
   frequency	one-time · low · medium · high (100+ per day)	How often a surface is used decides whether motion on it reads as polish or as a toll. Nothing in a diff reveals it.
   initiation	keyboard · pointer · system	A keyboard-initiated surface is reached by someone who already knows where they are going; an animation there is pure latency.

   Both are prefilled from the surface's role and overridden only on evidence. There is no resolver and no inference from handlers: a mousemove listener is not proof of pointer initiation, and a keyboard shortcut in the code is not proof anyone uses it.

   Surface role	frequency	initiation
   Command palette	high	keyboard
   Tooltip	high	pointer
   Modal / sheet	medium	pointer
   Onboarding flow	one-time	pointer
   Toast / notification	medium	system
   Table row action	high	pointer

   The default for a role not listed is medium / pointer, stated in the brief as a default rather than left blank.

3. Inventory — only when an artifact was provided, and before you build. List the artifact's interactions, keyframes, and script includes from its source (design-fidelity-mechanics § Data-basis ladder — read it, do not look at a picture of it). Then place every listed item in exactly one bucket, using the engine's own names (apply.ts, COVERAGE_BUCKETS) so the two surfaces read the same. Nothing enforces that they stay the same — this is a copied vocabulary, not a shared constant, and no test pins it; renaming a bucket in the engine leaves this list stale and silent:

   Bucket	Meaning
   honoured	carried over as-is
   translated	carried over in a different shape (a handler became a framework binding)
   flagged	not carried over — with the reason, in the output

   Dropping a handler stays allowed; hiding one does not. An item in no bucket is the failure this step exists to catch — "missing JavaScript" is what it looks like from the user's side. Inside the engine this is enforced (apply.ts, COVERAGE_BUCKETS, an unaccounted item is a halt); out here it is a duty you carry, and nothing checks it for you. (fixture: daf-adhoc-port-coverage.)

4. Build — against the audit's primitives, in the project's stack. Where the artifact's own markup/CSS/JS is stack-compatible, adapt that code; re-deriving it from scratch is a deviation needing confirmation (design-fidelity-mechanics § Adopt the code).

5. Review, then re-enter — run design-review before calling it done, and scope the verdict honestly: render-scoped when you can render it, otherwise explicitly static-scoped, naming which checks actually ran. "Looks good" with neither scope named is a verdict without evidence. Findings → fix them and re-enter step 5, at most 2 rounds, and stop early on a null: a round that produces no new finding ends the loop, and a round that produces only findings the provided artifact already covers produces no new finding. At the 2-round ceiling with findings still open, stop and hand the remaining list back — ship-as-is or abort is the user's call, never another silent pass. Judgement alone never buys a third round.

   The ceiling is 2, the same number the engine enforces (directives/ui/polish.ts, POLISH_CEILING), so the ad-hoc path and the ticketed path bound the loop identically rather than by two conventions that can drift. Fixtures: daf-adhoc-converges (round 2 is a null, loop ends) and daf-adhoc-ceiling (findings remain at the ceiling, loop hands back).

The floors — pulled immediately before the write

PULL references/craft-floor.md IMMEDIATELY BEFORE THE WRITE, NOT AT SKILL LOAD.

Twelve universal floors that do not vary by mode, register, intent or stack. They were inline here until the delivery point moved: a floor read at skill-load time can be a compaction away from the write it governs (ADR-227).

One precedence rule stays here because it is authority, not craft: the audit outranks every heuristic. Say [audit override] when it does.

Positioning — reference inside the engine, executor outside it

Inside a /implement-ticket run, fe-design is a universal reference skill: it carries stack-agnostic heuristics the UI directive set cites, and it does not own the flow. The ownership table below applies to that mode. Outside the engine the table has no dispatcher behind it — the ad-hoc loop above is the owner instead.

When to use

Cite this skill when:

• Planning a new page or feature UI before implementing
• Choosing between component patterns (modal vs. inline, table vs. cards)
• Designing forms with complex validation or multi-step flows
• Making responsive design decisions
• Reviewing UI for accessibility and usability

Inside a ticketed run, do NOT use this skill to:

• Implement components — that is the apply-step's stack-dispatched skill
• Audit an existing UI — that is existing-ui-audit
• Drive the full UI flow — that is the directives/ui/ orchestrator

Outside a ticketed run, the first and third lines invert: no dispatcher is running, so implementation quality and the flow are yours. The second holds in both modes — the audit is always existing-ui-audit; you invoke it, you do not replace it.

Resource-first, before taste. Any request to recreate / redesign / mock / prototype / improve an existing UI runs the resource-first context gate FIRST — search the project's tokens/design-system/assets, hard-stop and ask when a referenced source is inaccessible (never invent from memory), and prefer code over screenshots for exact values. Planning here starts from that inventory, not from generic aesthetic memory (design-artifact lifecycle, Inspect stage).

How the directive set cites this skill

directives/ui/design.ts produces the design brief (layout, components, states, microcopy, a11y). Selection decisions (style, semantic color tokens, typography pairing, layout pattern, anti-patterns) come grounded from design-intelligence — run its corpus query first; fall back to the heuristics in this reference only where the corpus reports an evidence gap or the audit already pins a project pattern. Stack-specific choices come from the dispatched implementation skill, not from here.

When the corpus is not installed. design-intelligence ships in the frontend-design pack; this skill ships in engineering-base. A consumer who installed only laravel or only react therefore has this skill and not the corpus. Fall back to the heuristics here and say so in the result — "selected from heuristics; frontend-design not installed, so no corpus grounding". The evidence-gap fallback above is about a corpus that answered "nothing here"; this is about a corpus that is absent. Never present an ungrounded pick as grounded, and never record a missing pack as an evidence gap.

Section index — load on demand

Load the reference file whose sections the task needs — never all of them by default:

• references/design-patterns.md — Component Architecture · Form Design · Table Design · Responsive Strategy · Accessibility (a11y) · Motion — decision-tree and rationale · UX Principles · Craft details — typography & imagery · Presenting variants
• references/design-read-and-memory.md — Cross-task design memory — read DESIGN.md / PRODUCT.md first · Register — brand vs product · Design Read — articulate intent before generating · Aesthetic direction

Procedure

When directives/ui/design.ts (or any caller) cites this skill:

1. Inspect state.ui_audit first — review the audit produced by existing-ui-audit; it is mandatory. Stop and request the audit if missing.
2. Pick the smallest matching section — Component Architecture, Form Design, Table Design, Responsive Strategy, Accessibility, or UX Principles. Cite by H2/H3 heading, never paste the whole skill.
3. Defer to audit findings — when the audit pins a project pattern (token, primitive, layout convention), use it. The heuristics here are fallbacks for gaps, not overrides.
4. Defer to the stack apply skill — Blade vs. Livewire vs. Flux vs. React-shadcn choices come from the dispatched implementation skill, never from this reference.
5. Surface conflicts — if a heuristic here contradicts an audit finding or stack convention, name both and let the caller decide; do not silently pick.

Output format

When this skill's content is folded into a design brief or review:

1. Quote the cited heuristic verbatim, with the H2/H3 heading and a one-line "why this applies" tie-back to the request.
2. Map each heuristic to a concrete artifact in the brief (component, form section, table column, breakpoint rule, a11y check, UX state).
3. Keep stack-agnostic — never name Blade/Livewire/Flux/React primitives in the cited prose; the apply step adds those.
4. Mark anything overridden by audit findings as [audit override] and link to the audit entry.

Related

• Orchestrator: directives/ui/ — owns the UI flow
• Pre-step (mandatory): existing-ui-audit — inventory before design
• Stack apply skills (dispatched, not standalone):
  • blade-ui — Blade template implementation
  • livewire — Livewire component implementation
  • flux — Flux component library usage
  • react-shadcn-ui — React + shadcn primitives
• Adjacent reference: dashboard-design — monitoring dashboard design (different domain)

Gotcha

• Don't design components without running existing-ui-audit first — the audit's component/token inventory is the canonical source for "what already exists in this project". Reinventing is the #1 failure mode.
• Heuristics in this reference apply across stacks; do not promote them to project rules without checking the audit.
• Mobile-first is not optional — every layout must work on 320px width.

Read the authority object — never re-infer it

Read the resolved ui_authority (contract) before the loop: surface_mode, register, change_intent, reference_maturity, constraints. This skill is a declared consumer — a second decision table beside the object is a drift surface, so do not re-derive any field.

surface_mode sets density, hierarchy and expressiveness only, per design-modes § The second axis.

QUALITY FLOORS DO NOT VARY BY SURFACE MODE.
A FLOOR THAT MOVES WITH THE MODE IS A PREFERENCE WEARING A FLOOR'S NAME.

The source-led path

reference_maturity: runnable-artifact → the artifact's own markup, CSS and JS is the data basis; adapting it is the default and a from-scratch re-derivation is a deviation. Sort every mechanic into honoured / translated / flagged; one present, absent and unflagged is a silent drop, target zero. Procedure and the vanishing-mechanic classes: references/source-led-port.md.

Maturity is not decided here — it arrives on the authority object, and per-value provenance belongs to road-to-frontend-fidelity-calibration Phases 0 and 2.

Anti-slop discipline

Before proposing a direction, scan the Visual (V1–V8) and Layout (L1–L10) sections of design-antipatterns, plus Motion (M1–M8) for the interaction layer. On a match, choose differently or invoke the entry's own override condition in the brief — every entry has one. Two of them, T7 and T8, are additionally register-scoped: the brand and product registers admit different answers, and neither admits an undeclared pick. The Q* floors are in references/craft-floor.md instead of here, so they arrive at the write rather than at skill load.

Do NOT

• Do NOT skip mobile viewport testing.
• Do NOT use fixed pixel widths for responsive layouts.
• Do NOT ignore accessibility requirements.
• Do NOT use this skill as an executor INSIDE a ticketed run — there the executor is directives/ui/design.ts and this is the reference it cites. Outside one, the ad-hoc loop at the top of this file IS the executor; do NOT read the reference and then write the UI from priors.