product-design

这段内容是审核时保存的快照。外部来源才是完整且最新的版本。

---
name: product-design
description: 'Turn a vague product ask ("this page feels wrong", "we need a team view") into a grounded, shippable design by establishing business semantics and the user task before proposing a surface. Use when scoping a new surface, redesigning an existing one, or deciding what a feature should be. Not for auditing a built screen (ux-audit) or choosing visual details (ux).'
---

# Product Design

Decide **what to build and why** before arguing about spacing.

Two rules are non-negotiable:

> **Never design from the surface. Establish what the business actually models
> first.**

> **Never turn the business model into the information architecture. The domain
> constrains what the surface may claim; the user's retrieval and decision task
> determines how the surface is organized.** (`P-14`)

This skill handles product meaning first. It may hand a grounded frame to `ux` or
`design-prototype`, but discovery does not require implementation or a prototype.
Strip out every framework, table and component name: if no product insight remains,
the finding belongs in an engineering skill.

## Relationship to other skills

| Skill                                            | Question                                      | When                        |
| ------------------------------------------------ | --------------------------------------------- | --------------------------- |
| **product-design**                               | What should this surface be, and why?         | Before or during framing    |
| [ux](../ux/SKILL.md)                             | How should the interaction behave and feel?   | While designing or building |
| [ux-audit](../ux-audit/SKILL.md)                 | Does the built surface honor those rules?     | After it exists             |
| [design-prototype](../design-prototype/SKILL.md) | How can a materialized interaction be tested? | When a prototype is useful  |

## Select the smallest run mode

| Mode            | Finish line                                                                  |
| --------------- | ---------------------------------------------------------------------------- |
| **Discover**    | Evidence map, business model, user-model hypotheses and structural diagnosis |
| **Frame**       | Discover + view model, information architecture, scope and open decisions    |
| **Materialize** | Frame + a requested spec and/or prototype                                    |

Do not create or modify artifacts unless the user asked for a design artifact,
prototype or repository change.

For a substantial run, read [references/pattern-base.md](references/pattern-base.md)
and [references/layer-model.md](references/layer-model.md) in full. For a narrow
question, inspect only the relevant patterns and state what scope was checked.

## Step 1 — Ground the business model

Establish what is true before diagnosing the surface. For broad repositories,
parallelize independent evidence areas when delegation is available and useful;
the primary agent remains responsible for reconciling contradictions.

Produce:

- Concepts, roles and states, including variants the current surface hides (`P-04`).
- Who produces the terminal business fact and what closes the lifecycle (`P-12`).
- What each state obliges someone to do (`P-01`).
- What business event each action produces (`P-02`).
- Concepts the business does not have (`P-06`).
- Evidence sources, contradictions and confidence.

Treat domain types and state machines as strong evidence, not the sole source of
truth. Check relevant service behavior, permissions, policy, production facts and
user evidence when available. Code can be stale, accidental or incomplete.

Never proceed from the team's assumed model of the product. Treat it as a
hypothesis until checked against domain and user evidence.

## Step 2 — Frame the user's view model

Grounding constrains what the product may claim; it does not determine how a
person should see it. Separate o