design-systems（设计系统）

Design Systems

Scope

Covers

• Creating or upgrading a design system (tokens + components + guidelines)
• Using blockframes (lo-fi, system-aware wireframes) to lock logic before hi-fi execution
• Designing a future-ready visual foundation (depth/elevation, motion, texture) without breaking consistency
• Making the system easy for non-experts to use (guardrails, examples, starter templates)
• Driving adoption + governance (contribution model, champions, release cadence)

When to use

• “We need a design system / component library and a plan to build it.”
• “Our UI is inconsistent—define tokens + components + documentation to standardize.”
• “We want to refresh our UI style (more depth/texture/motion) without chaos.”
• “We need to scale design across teams or support enterprise customers with customization.”
• “We want faster hi-fi output by locking flows in lo-fi first.”

When NOT to use

• You’re defining a brand identity or logo system (different process).
• You need user research/discovery to decide *what* to build.
• You only need to ship one isolated UI change (just implement it).
• You’re doing pure front-end architecture unrelated to UI consistency.
• You want to stand up a design engineering function or define the hybrid role (use design-engineering).
• You need to run a one-off design review or critique session (use running-design-reviews).
• You want to improve engineering team culture, rituals, or hiring (use engineering-culture).
• You need a detailed feature spec or UX flow document (use writing-specs-designs).

Inputs

Minimum required

• Product + surfaces: web/iOS/Android; key flows
• Current state: existing UI kit/design system (if any), design tool (e.g., Figma), code stack (if relevant)
• Goals: speed, consistency, accessibility, scalability, customization, enterprise adoption
• Constraints: timeline, team ownership, level of engineering support, compliance/a11y needs

Missing-info strategy

• Ask up to 5 questions from references/INTAKE.md, then proceed with explicit assumptions.
• If platform/stack is unknown, assume a modern web product with a component library and design tokens.
• Do not request secrets or credentials.

Outputs (deliverables)

Produce a Design System Operating Pack in Markdown (in-chat by default; write to files if requested):

1. Context snapshot (goals, constraints, success signals)
2. Design system charter (mission, scope, principles, audiences, in/out)
3. UI audit + operational blockers (what’s slowing teams down; what must standardize first)
4. Blockframe-to-component map (lo-fi flows + mapping to components/tokens)
5. Token model (taxonomy, naming rules, and initial token backlog—include elevation/depth)
6. Component inventory + roadmap (tiers, prioritization, milestones)
7. Documentation + enablement plan (non-designer-friendly, “teaches by structure”)
8. Governance + adoption plan (contribution workflow, decision rights, champions, release cadence)
9. Quality gate (checklists + rubric score) + Risks / Open questions / Next steps

Templates: references/TEMPLATES.md

Workflow (7 steps)

1) Intake + success definition (who is this for?)

• Inputs: User context; references/INTAKE.md.
• Actions: Confirm primary users of the system (designers, engineers, PMs, “non-designers”). Define success signals (cycle time, consistency, adoption, fewer UI bugs, faster onboarding).
• Outputs: Context snapshot (draft).
• Checks: Success is measurable or at least falsifiable (e.g., “80% of new screens use system components”).

2) Audit the current UI and find the operational “hook”

• Inputs: Screens/flows, existing components, pain points, enterprise needs (if any).
• Actions: Inventory inconsistencies (spacing/type/color/components), identify the operational blocker the system will remove (e.g., slow production, inconsistent UI, customization needs). Choose the first high-leverage slice.
• Outputs: UI audit + operational blockers list; initial scope slice.
• Checks: The first slice is narrow enough to ship but broad enough to set patterns.

3) Lock logic with blockframes (separate thinking from styling)

• Inputs: Key flows; current IA; constraints.
• Actions: Create or specify “blockframes” (lo-fi, system-aware wireframes). Map each block to intended components and token usage so hi-fi execution becomes faster and more consistent.
• Outputs: Blockframe-to-component map (v1).
• Checks: A reviewer can validate flow/IA without debating visual details.

4) Define the token model (make the future style changeable)

• Inputs: Brand constraints, accessibility targets, desired direction (e.g., depth/texture/motion).
• Actions: Define token taxonomy + naming; include elevation/depth and state tokens. If doing a visual refresh, design the token model so style can evolve without rewriting components.
• Outputs: Token model + token backlog (v1).
• Checks: Tokens support theming and states; accessibility constraints are addressed (contrast, focus, motion).

5) Define the component model + delivery plan

• Inputs: Audit + blockframes + token model; engineering constraints.
• Actions: Tier components (primitives → composites → patterns). Prioritize by reuse and user impact. Define milestones, owners, and acceptance criteria.
• Outputs: Component inventory + roadmap (milestones).
• Checks: Milestone 1 ships within 1–2 weeks and establishes “golden path” patterns.

6) Make it easy to use (guardrails for non-experts)

• Inputs: Target user types; common mistakes; documentation needs.
• Actions: Design documentation and component guidelines so they “teach by structure”: sensible defaults, constrained options, examples, do/don’t. Provide starter templates for common layouts.
• Outputs: Documentation + enablement plan (v1).
• Checks: A non-expert can assemble a consistent screen using templates with minimal training.

7) Governance + adoption + quality gate

• Inputs: Draft pack; stakeholder map; toolchain (Figma/Storybook/etc.).
• Actions: Define decision rights, contribution workflow, review gates, and release cadence. Create a champion/office-hours plan to drive adoption. Run references/CHECKLISTS.md and score with references/RUBRIC.md. Finalize Risks / Open questions / Next steps.
• Outputs: Final Design System Operating Pack.
• Checks: Ownership is unambiguous; adoption plan exists; quality bar is explicit and repeatable.

Quality gate (required)

• Use references/CHECKLISTS.md and references/RUBRIC.md.
• Always include: Risks, Open questions, Next steps.

Examples

See references/EXAMPLES.md.

Boundary example 1: "We want to create a design engineering team that owns UI craft and ships production code." Response: this is about standing up a hybrid practice/role, not building a system of reusable tokens and components; redirect to design-engineering. Use design-systems when the goal is the system itself (tokens, components, governance), not the organizational function.

Boundary example 2: "Can you run a design critique on our new checkout flow mockups?" Response: a one-off design review is not a design system engagement; redirect to running-design-reviews. Use design-systems when the goal is building or evolving the reusable component/token layer, not reviewing a specific flow.

Anti-patterns

Avoid these common failure modes when producing a Design System Operating Pack:

1. Token taxonomy without consumers — Defining hundreds of tokens (color, spacing, typography, elevation) before any component uses them. Start with the tokens needed by the first milestone of components; expand the taxonomy as real consumers demand it.
2. Component library as a museum — Building components that look great in Storybook but are never adopted by product teams. Every component in the roadmap must have at least one committed consumer (a real product flow), and the governance plan must track adoption rates.
3. Governance without teeth — Defining contribution guidelines and decision rights on paper but having no enforcement mechanism (no review gate in CI, no required design-system review on PRs). Governance that relies solely on goodwill will erode within one quarter.
4. Blockframe skipping — Jumping directly to hi-fi mockups without locking information architecture and flow logic in lo-fi blockframes first. This leads to expensive rework when IA changes invalidate hi-fi work.
5. One-size-fits-all documentation — Writing component docs aimed only at designers or only at engineers. The documentation plan must explicitly address both audiences and include recipes/starter templates for non-experts.