product-design

Product Design

Decide what the interface should do, then route who builds and verifies it: pick the right interaction, make scope and consequence clear, cover reality beyond the happy path. This skill owns the decision; it routes the build, verification, and copy out (ownership map in Related skills).

• IS: the decision layer. From a brief, spec, mockup, intent, or existing UI: choose the right interaction and control, name the object, scope, and consequence of actions, enumerate every reachable state, set resilience expectations, require accessibility as task completion. It decides, then routes build and verification out.
• IS NOT:
  • building or styling UI, visual direction, palettes, type: use ui-design.
  • auditing the built result (rendered quality, a11y markup, keyboard, layout, performance, type surface, React/Next code-level UX with a ship verdict): use ui-design Audit mode.
  • copy wording, persuasion, or AI-ism removal: use copywriting.
  • deep typography or motion: use typography-audit or ui-animation.

product-design, ui-design, or ui-animation?

An interface is a set of states and the passages between them. That decomposition assigns the work.

• Subject beats artifact. When motion is what the request is about, it is ui-animation whether or not code exists yet.
• Artifact is the opening presumption, not the verdict. A brief, spec, mockup, or intent with no code is this skill. Code, a diff, or a running UI in hand presumes ui-design, and the next test can overturn that: this skill reads existing UI whenever the question is what it should do.
• Capability beats presentation. With code in hand, ask whether the change alters what a user can do, which objects an action affects, whether it is reversible, or whether a state exists at all. That is a capability, so this skill decides and ui-design implements. If it only changes how the same capability looks, reads, or behaves, ui-design owns it end to end.
• A gesture that replaces a control is a capability decision. Swipe-to-delete, hold-to-confirm, and drag-to-reorder change what the user can do and how recoverable it is, so this skill settles the interaction and ui-animation builds its physics.
• Motion incidental to a build stays in ui-design. A hover transition or a fade added while building a component is a property of that component. It becomes ui-animation's when motion is the subject or its craft is in question.

Two edges the tiebreak does not settle on its own:

• Choosing between control patterns with different reachability is a capability, so this skill. Modal against inline, drawer against full page, and dialog against toast each change what stays visible, how the task is dismissed, and where focus lands (rule/inline-before-modal). Styling whichever is chosen is ui-design's.
• A missing state nobody would debate is ui-design's to detect and build. An empty list, a failed fetch, and a pending submit all obviously need a state, and its states- audit rules find and fix them. This skill decides which states must exist only where that is genuinely open, such as whether a partial or an expired state should exist at all.

Worked: "Delete should be undoable" is this skill. "The undo toast is ugly" is ui-design. "The undo toast should slide, not pop" is ui-animation.

One artifact often needs both in sequence: this skill decides the states that must exist, ui-design Audit mode verifies the built code and rendered result implement them. This skill reviews the decision and stops at decision altitude; it never writes line-level code fixes.

Operating contract

• Cite a stable rule ID for every finding or non-mechanical decision. Never invent an ID; if none fits, record a coverage gap.
• The project's design system and AGENTS.md outrank this skill's defaults. Defer to them.
• Never restyle or rebuild. Decide, then route the build to ui-design.
• One mode per request, resolved from the user's verb before acting.

Request modes

Resolve the mode from the user's verb and artifact, then load that mode's references. references/rules.md loads in every mode: the citation contract binds all of them, and you cannot conclude that no existing rule governs a decision without the registry in front of you.

review mode is about a flow, not an artifact. "Audit this component", "check my UI", or "design QA this page" point at built markup and belong to ui-design Audit mode. This skill's review asks whether the decisions behind a flow are right, and stops at decision altitude.

Modes chain: shape leads into spec; review leads into harden. When intent is ambiguous, use the narrowest mode the verb supports. A URL, screenshot, route, or component identifies scope; it does not authorize edits.

A material decision: see references/product-judgment.md.

Decision authority

Conflict order, highest first:

1. The user's explicit goal and constraints.
2. Verified user and product evidence, and what the system actually does.
3. Project-canonical guidance: AGENTS.md or CLAUDE.md, the project's design system, routed sibling skills.
4. Sibling-skill ownership: route, do not duplicate (ownership map in Related skills).
5. This skill's product design standards (below).
6. General interface and platform conventions.

When a request spans authorities, name the owning skill and hand off.

Workflow

Product design pass:
- [ ] Step 1: Classify the request into one mode
- [ ] Step 2: Locate authority (user constraints, project design system, AGENTS.md)
- [ ] Step 3: Load only that mode's reference files
- [ ] Step 4: Name object, scope, and consequence for each action in scope (spec, action, review)
- [ ] Step 5: Enumerate reachable states and check coverage (shape, spec, harden)
- [ ] Step 6: Apply standards; cite a stable rule ID per finding or decision
- [ ] Step 7: Emit output (review and harden use P0-P3); route follow-on work to siblings
- [ ] Step 8: Run the pass self-check

Steps 4 and 5 are mode-scoped because their references are: a pure action pass has no state matrix to enumerate, and a shape pass has no built actions to name yet. Run the step that its mode's loaded files support.

For shape, spec, harden, or any material product or flow change, write the compact internal brief specified in references/product-judgment.md before proposing UI. If its job, desired outcome, and consequence fields cannot be filled in, stop and ask rather than guessing.

Output length follows the work, not the template. A single settled decision is a short answer; drop the sections a pass did not need rather than filling them.

Pass self-check

Close every pass with these, and label it INCOMPLETE if any fails:

• Every finding and non-mechanical decision carries a rule ID from references/rules.md.
• Every decision no existing rule governs is recorded inline as a coverage gap.
• No cited ID was invented: each one appears verbatim in references/rules.md.
• The internal brief is present with job, desired outcome, and consequence filled, for shape, spec, and harden.

Product design standards

Five pillars, each naming the rule IDs in references/rules.md that govern it and the reference that details it. Resilience shares rule/cover-reachable-states with state coverage rather than adding an ID of its own: it is the same requirement pointed at adverse inputs, which is where reachable states are most often left undesigned.

• Right interaction. Pick the control from the choice's shape; keep options visible and reversible; prefer inline disclosure over a modal; choose the smallest coherent intervention. rule/control-matches-cardinality, rule/navigation-vs-action, rule/inline-before-modal, rule/smallest-intervention. See references/product-judgment.md.
• Action naming. Name the object, scope, and consequence; destructive CTAs use Verb plus Noun, never "Confirm" or "OK"; make friction proportional to impact and offer undo when honest. rule/name-object-scope-consequence, rule/destructive-names-action, rule/destructive-proportional. See references/naming-and-copy.md.
• State coverage. Design every reachable state, not just the populated one; empty states name the object and a first action; errors explain and offer recovery; preserve user input. rule/cover-reachable-states, rule/empty-state-action, rule/error-states-recovery, rule/preserve-user-input. See references/surfaces.md.
• Resilience. Require that overflow, extreme data, localization and RTL, and network-failure states be designed; every fetch lands in a designed state. rule/cover-reachable-states. See references/surfaces.md. Whether the built UI renders them correctly is ui-design Audit mode's check.
• Accessibility as a product concern. Every control has an accessible name; the primary flow is completable by keyboard with visible focus; state and consequence are understandable, not just labeled. rule/accessible-name-required, rule/keyboard-complete-flow, rule/no-custom-focus-bypass. Route axe-style markup checks to ui-design Audit mode. See references/interface-quality.md.

Review output

In review and harden modes, lead with findings ordered by user impact (P0-P3), each with location, verification status, rule ID, user consequence, and the smallest concrete fix with the skill that owns it. Keep findings at decision altitude; a line-level code or framework fix is ui-design Audit mode's output. Full severity rubric and finding format in references/interface-quality.md > Severity rubric.

Linters vs agent guidance

Deterministic, structural, single-file checks (control selection by option count, nested modals, missing accessible names) belong in the consuming project's linter, wired to that project's components; judgment that needs product context (which object, what consequence) stays here. See references/lint-patterns.md for the decision tree and the rules worth encoding.

Gotchas

• Emitting a line-level fix (a prop, a hook, a className) instead of the decision. It arrives without the rendered check that would validate it, and the product decision it was supposed to carry goes unstated. Route it to ui-design Audit mode.
• Proposing UI when the internal brief's job, desired outcome, or consequence field cannot be filled. Every finding after that rests on a guessed job, so stop and ask (references/product-judgment.md).
• Citing a plausible-sounding rule ID that does not exist (rule/clear-labels). The citation resolves to nothing, so the finding cannot be deduped against a sibling audit or traced to a rule. Record a coverage gap instead.

Related skills

• ui-design: visual direction and building the decided interaction in code; its Audit mode covers the built result, both rendered quality and accessibility-markup audit and React or Next diff-level UX bug hunt with a ship verdict.
• copywriting: exact wording for names, errors, and empty and loading copy; defines shared copy rule IDs in its references/ui-states.md.
• ui-animation: the passage between two states (timing, easing, springs, gesture physics). This skill settles whether a gesture replaces a control and whether the action it triggers is reversible; that skill builds its physics.
• typography-audit: deep type.
• Taste Training (blode.co/taste-training): trains the eye these rules encode, across type, copy, craft, interaction, and motion.