Communitygithub.com

k1e1n04/claude-skills

UI 以外のユーザー接点(API・設定ファイル/設定項目・拡張機能/プラグイン)を、UX の 6 原則に沿って設計・レビューする。API 設計、REST/SDK/クライアントライブラリの使いやすさ、config.yml や環境変数や設定項目の追加、拡張ポイント/プラグイン機構の設計、「設定を足すべきか」「デフォルト値をどうするか」「APIのバージョニング」「開発者体験(DX)」「オンボーディング」「驚き最小の原則」などの話題が出たら、明示的に UX と言われなくても必ず使う。システム設計の初期段階で UX を先に決めたいときにも使う。

claude-skills 是什么?

claude-skills is a Claude Code agent skill that uI 以外のユーザー接点(API・設定ファイル/設定項目・拡張機能/プラグイン)を、UX の 6 原則に沿って設計・レビューする。API 設計、REST/SDK/クライアントライブラリの使いやすさ、config.yml や環境変数や設定項目の追加、拡張ポイント/プラグイン機構の設計、「設定を足すべきか」「デフォルト値をどうするか」「APIのバージョニング」「開発者体験(DX)」「オンボーディング」「驚き最小の原則」などの話題が出たら、明示的に UX と言われなくても必ず使う。システム設計の初期段階で UX を先に決めたいときにも使う。.

兼容平台✓Claude Code~Codex CLI~Cursor
npx skills add https://github.com/k1e1n04/claude-skills/tree/HEAD/skills/ux-design-for-architects

在你喜欢的 AI 中提问

打开一个已预加载此 Agent Skill 的新对话。

文档

ux-design-for-architects

A guide for architects to design and review the UX of non-UI user touchpoints (API / Configuration / Extensions). It does not cover UI screen design (colors, layout, etc.).

UX extends beyond the UI to every point of contact between the system and the user. Poor UX at these touchpoints is hard to change later (because users depend on it) and lowers the overall probability of success. Therefore, decide on UX before implementation.

How to Proceed

  1. Identify the touchpoint: The target is one (or more) of: API, Configuration, or Extensions.
  2. Identify the user (Principle 1): Who uses it, for what purpose, and at what technical level?
  3. Read the detailed guide for the relevant touchpoint:
    • Configuration → references/configuration.md
    • API → references/api.md
    • Extensions → references/extensions.md
  4. Check against the 6 principles (see below). If proposing a design, include a brief rationale referencing the principles.
  5. For a review, format findings according to references/review-checklist.md.
  6. If opinions differ, confirm with the user using a hypothesis (AskUserQuestion).

The 6 UX Principles

#PrincipleKey PointWhy
1Understand the userGrasp goals, technical level, mental model, and typical workflows. Make the system behave as expected (principle of least surprise)Users act differently than we predict. Observing and measuring is more effective than asking
2Do the minimum necessaryOffer only necessary features. Adopt widely used standards (OAuth, OpenID, etc.) if availableEverything you provide becomes debt. Once released, users get used to it and it becomes hard to change
3Make manuals unnecessaryMake usage self-evident; guide with recommendations and cuesUsers don't read manuals. If they can't solve the problem, they blame the product
4Think in terms of information exchangeAvoid asking > ask less > ask in an easy-to-answer format. Never ask the same thing twiceInput is a burden. Reduce it through inference, memory, and derivation
5Keep simple things simpleRemove barriers for the first few operations. Provide pre-configured clients or samplesBeginners are most likely to drop off. The first experience determines retention
6Design UX before implementationDesign UX before architecture; iterate with feedbackChanging UX after implementation is costly. Delay implementation; deepen design

Deciding When to Ask for Information (Elaboration of Principle 4)

Think in terms of priority:

  1. Don't ask: Use reasonable defaults, auto-detection, remembering past inputs, and deriving from known information.
  2. Ask less: Only ask for what is truly necessary.
  3. Ask in an easy-to-answer way: Frame questions in terms the user can answer without mental calculation, not in internal implementation terms (e.g., instead of "maximum cache entry count," use "memory limit for cache" or derive details from a mode choice like "server or client").

Returning Output / Responses (Principle 4)

Unless there is a performance issue, return information generously in a form the user can understand, making the next action predictable. When passing the result of one call as input to another, allow the user to pass it on with minimal processing.

Summary by Touchpoint

  • Configuration: The best configuration is "no configuration." Next best is automatic determination, and last is with reasonable defaults. The question is: "Does the user have better context than we do?" — Only add a setting if the answer is Yes. Do not silently ignore unknown keys or failed applications; report errors. Validate settings at startup or when they are applied.
  • API: Users are developers. Make it easy to learn, hard to misuse, and always make the next step clear. Do one thing well. Don't make users write boilerplate code. Follow standards. Version from the start. Delay public release until necessary; first use it internally.
  • Extensions: Wait until you deeply understand the use case before adding them (they can be added through refactoring later). Do one thing well. Define inputs and outputs; do not expose internal structure. Once published, they are hard to change.

Common Mistakes (Suspect During Design or Review)

  • Lack of UX expertise / Delaying involvement of experts
  • Focusing only on UI while neglecting UX of APIs, configuration, and extensions
  • Technical leaders making UX decisions and only asking experts for UI tweaks
  • Ignoring expert proposals
  • Getting stuck between two options and using a configuration setting to escape (pushing the problem onto the user)

If no expert is available, state this explicitly and recommend involving a UX expert early. However, do not stop designing due to a lack of advice; produce a provisional plan following this guide.

Connection to Iterative Approaches

  • Derive design from user journeys → Apply this guide to each touchpoint in the journey.
  • Slice iteratively → Only build the touchpoints needed for the first slice (public APIs and extensions can often wait until after an MVP).
  • Design deeply and implement slowly for things that are hard to change → UX is the prime example. Learn by observing real usage.

Output Format

  • Design proposal: State concisely in the order: "Target touchpoint → User profile → Proposal → Relevant principles and reasons → Open issues."
  • Review: Use the format from references/review-checklist.md.

相关技能