Agent trace 诊断器:当用户提供 trace、日志 JSON、agent 执行记录、工具调用序列,或说“看下这个日志/trace,定位根因”“只说明问题不要改文件”时使用;用中文输出核心根因、可视化执行流程图、trace 证据链、可能文件/行和修复建议,默认只读。不用于没有日志证据的新功能设计、普通代码解释或直接修复代码。
日本語の概要は準備中です。原文の説明を表示しています。
PRD 到高保真 UI 交付对齐 / 桌面工作台 UI mockup:当用户已有 PRD、UI 规范、真实前端项目路径、截图或页面状态清单, 需要先从 PRD 提炼 UI 结构、状态模型和 ASCII 布局,再进入高保真 mockup、项目原生 preview 或开发交付 handoff 时使用。
优先输出项目原生 preview route/component;只有早期概念确认或无法接入真实项目时才输出 standalone HTML。必须包含结构阶段产物、screen contract、component map、implementation notes、截图/验证结果和 HTML/preview 的迁移边界。
インストールする前に、エージェントに与えられる指示の中身を確認できます。
ui-mockup-desktop-workbench基于 PRD 和 UI 规范出高保真 mockup、先从 PRD 出 UI 结构和 ASCII 布局再做高保真、生成桌面工作台真实页面、把这个 Agent Client PRD 做成可交付开发的 UI、让前端按这个 UI 复刻Use this Skill when the goal is not just a pretty standalone page, but a staged UI handoff that product, design, and engineering can inspect and implement from.
The default output is a UI implementation contract, not disposable HTML. The contract starts with UI structure and states, then moves to visual fidelity only after the screen model is coherent. Standalone HTML is allowed only as one output mode. When the user provides a real frontend project path and wants development to reproduce the design, prefer a project-native preview route/component using the real app's components, tokens, icons, and layout conventions.
The workflow combines:
structure-only discipline: structure and state before polish.structure-only is the lightweight mode for users who only want UI structure, state model, ASCII layout, or a quick low-fidelity HTML wireframe. It is part of this Skill, so there is only one UI trigger surface.
This Skill owns the full chain:
PRD -> screen inventory -> state model -> ASCII layout -> structure review gate -> high-fidelity visual handoff or project-native preview -> implementation handoff
If the user asks for high fidelity, real project alignment, visual handoff, frontend replication, or component mapping, stay in this Skill and include the structure stage internally.
If the user explicitly wants only low-fidelity structure and does not need visual polish or implementation mapping, select structure-only and stop after the structure review gate.
Choose one mode before implementing visuals. State the chosen mode and why.
For detailed selection rules, read references/output-mode-selection.md.
| Mode | Use When | Primary Outputs | Do Not Use When |
|---|---|---|---|
project-native-preview | The user provides a real frontend project path or asks development to "replicate" the UI. | ascii-layout.md, preview route/component in the project, screen-contract.md, component-map.md, implementation-notes.md, screenshots. | The user only wants early concept exploration or forbids repo edits. |
visual-handoff | The PRD is stable and development needs exact visual guidance, but direct project integration is risky or not requested. | ascii-layout.md, high-fidelity HTML/React artifact, screenshots, component map, implementation notes, migration boundary. | The user expects production-ready code. |
concept-html | Early product/design discussion needs a quick visual reference after structure is checked. | ascii-layout.md, standalone mockup.html, screen contract, assumptions. | The user wants frontend to implement it accurately; then use visual-handoff or project-native-preview. |
Default to project-native-preview when all are true:
Default to visual-handoff when the real project exists but editing it would exceed the requested scope.
Use concept-html only when the goal is fast concept confirmation, not production implementation alignment.
This Skill is downstream of PRD maturity:
prd-architect.prd-review before producing the formal mockup.prd-architect may include lightweight UI references, screenshots, or mockup notes inside a PRD. Formal multi-state desktop HTML mockups belong here.
If the UI is meant to drive implementation, update the PRD or prepare a PRD UI appendix after the final visual decision. The PRD should link the final mockup/preview, screenshots, and screen contract.
Before generating UI, collect or discover:
Minimum input maturity:
If these are missing, route back to prd-architect or prd-review depending on whether the PRD must be created or reviewed.
Ask at most three questions only when missing information changes the page set, design constraints, or output format. If the user supplied PRD and UI spec paths, read them and proceed with explicit assumptions.
Before any visual implementation, inspect project-local design evidence when available. Do not ask the user for information that can be read from the repo.
Read or inspect:
Before generating UI, output 5-8 bullets under Discovered UI Constraints:
If the project cannot be started or inspected fully, say so and mark the mockup as static inference. Never claim "matches production" without project evidence.
Always model structure before visual fidelity. This is the reusable core of structure-only mode.
Create or include:
screen-inventory: pages, modals, popovers, panels, and major regions.state-model: empty, draft, loading, queued, running, stopped, success, partial failure, error, permission, and settings states relevant to the PRD.ascii-layout.md or an equivalent section: desktop layout zones, fixed/flexible widths, scroll containers, primary actions, and state-specific changes.wireframe-handoff.md or an equivalent review note: confirmed structure, assumptions, open questions, and the gate before high fidelity.interaction-notes: user actions, agent actions, state transitions, and recovery paths.prd-traceability: the PRD section, UI spec rule, screenshot, or assumption that justifies each screen/state.Run a structure review gate before high fidelity:
prd-review or prd-architect before polishing.Read authoritative inputs
prd-review.Choose output mode
project-native-preview, visual-handoff, or concept-html.Run the Design Discovery Gate
Discovered UI Constraints before implementing.Run the Wireframe Stage
ascii-layout.md or an equivalent layout section before styling.Pass the structure review gate
Extract a screen contract
Create a component map
Refine structure before visual styling
Design realistic page states
Implement the chosen artifact
project-native-preview, create a non-production preview route, story, fixture state, or preview component following local conventions. Keep it isolated and reversible.visual-handoff, create a high-fidelity HTML/React artifact and mark it as visual reference, not production code.concept-html, create a standalone HTML/CSS mockup that opens locally without a build step.ascii-layout.md.component-map.md and implementation-notes.md when the UI is meant to guide frontend implementation.prd-review for conflict/readiness issues, prd-architect for missing PRD sections.Read references/desktop-workbench-ui-principles.md before designing a desktop Agent workbench shell, multi-panel workspace, settings page, artifact page, or task execution UI.
Key defaults:
Read references/mockup-output-contract.md before finalizing deliverables.
Default deliverables:
README.md or handoff note with purpose, assumptions, and open/run instructions.screen-inventory.md and state-model.md covering the structure-stage contract.ascii-layout.md or equivalent wireframe-stage section covering UI structure, layout zones, and state model.wireframe-handoff.md or equivalent structure review note before visual polish.screen-contract.md or equivalent section describing screens, states, and PRD traceability.component-map.md when the result is used for implementation alignment.implementation-notes.md when a frontend team will build or convert the UI.mockup.html, high-fidelity React artifact, or project-native page/component depending on the chosen mode.When a product-delivery-manifest.yaml is present, read the canonical contract from prd-architect/references/product-delivery-manifest.md before packaging the result.
artifacts.action_contract, artifacts.html, artifacts.screenshots, ui_baselines, and anchors.references/output-mode-selection.md:选择 project-native-preview、visual-handoff 或 concept-html。references/templates/screen-inventory.md、state-model.md、ascii-layout.md、wireframe-handoff.md:结构阶段模板;用户只要 structure-only 时优先读取。references/verification-checklist.md:最终视觉和交付包验证清单。references/desktop-workbench-ui-principles.md:桌面 Agent 工作台布局和状态规则。references/mockup-output-contract.md:交付物结构、迁移边界和 implementation handoff。../prd-architect/references/product-delivery-manifest.md:Package 中 HTML、Action Contract、截图、anchor、baseline 与写权限合同。scripts/check_mockup_package.py:检查 mockup package 的最小文件结构。The Skill run is complete when:
Discovered UI Constraints are documented when a project path or existing UI is provided.Smoke prompts:
基于 /path/to/prd.md、/path/to/ui-spec.md 和 /path/to/frontend 输出桌面端 Agent Client 高保真 preview,开发要按这个实现。从这份 PRD 开始,先输出 UI 结构、状态模型和 ASCII 布局,确认后再继续做高保真。把这个桌面工作台 PRD 做成视觉 handoff,包含 component map、screen contract 和 HTML mockup。只想早期讨论布局,按这份 UI 规范做一个可打开的 concept HTML。我已经确认 PRD 了,现在要让前端完全复刻这个 UI。Non-trigger prompts:
帮我写一份 PRD。只审一下这个 PRD 有没有问题。做一个手机端 landing page。直接实现生产功能并补测试。Capability checks:
ascii-layout.md, screen inventory, and state model before high fidelity.screen-contract.md; implementation-oriented runs include component-map.md and implementation-notes.md.product-prdactiveまだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Agent trace 诊断器:当用户提供 trace、日志 JSON、agent 执行记录、工具调用序列,或说“看下这个日志/trace,定位根因”“只说明问题不要改文件”时使用;用中文输出核心根因、可视化执行流程图、trace 证据链、可能文件/行和修复建议,默认只读。不用于没有日志证据的新功能设计、普通代码解释或直接修复代码。
日本語の概要は準備中です。原文の説明を表示しています。
协作校准 / 认知校准 / 问题脑暴:当用户有模糊感受说不清问题、方案越补越复杂、想挑战假设、 想重新定义问题、想知道成熟领域里怎么叫这个问题时使用。 可用中文唤起:"帮我想想""先聊一下""一起脑暴""我感觉有问题但不清楚是什么" "先别执行,帮我看清问题""挑战我的假设""这个方案是不是想错了""帮我做认知校准" "双向钢人""钢人论证一下""正反最强论证,给我明确判断"。 也适用于:用户直接说"帮我做 X 功能""接入 Y 平台"但背景复杂、或对话早期感觉方向可能跑偏时, AI 主动识别并建议进入脑暴模式。 这里的“挑战假设”主要指问题定义、目标、约束和领域定位层面的假设;如果问题已确认且用户已有具体方案、 计划、架构或决策要压力测试,应转用 grill-me。不用于翻译、摘要、格式整理、单文件小改等简单执行任务。
日本語の概要は準備中です。原文の説明を表示しています。
AI 工作资产化诊断器 / Assetization Router:当用户提供 AI 协作会话、重复任务、prompt、工作流描述、 团队 AI 使用场景或一次成功交付过程,并要求判断应该沉淀成 Prompt、Context Pack、Workflow、Skill、 Loop、System,或根本不值得沉淀时使用。它只做诊断、分层和下一步资产建议,不替代具体 Skill 创建、 PRD 起草、调研、实现或自动化执行。不用于普通事实查询、一次性文案、trace 根因分析或高责任专业判断。
日本語の概要は準備中です。原文の説明を表示しています。
设计脑暴 / 实现前方案校准:当用户想把已基本成立的想法、功能方向或产品问题,在写 PRD、画 mockup 或进入开发计划前,先比较方案、确认取舍、对齐 UI/视觉约束,并收敛成可执行设计 spec 时使用。 可用中文唤起:“先脑暴一下方案”“先不要写 PRD,帮我设计几种路径”“参考 brainstorming 把这个需求变成设计 spec” “实现前先讨论设计”。问题还没定义清楚时先用 ai-collaboration-calibration;已有方案要压力测试时用 grill-me; 直接写 PRD 时用 prd-architect;grill-me 返回精确设计 gap 时,本 Skill 只输出 Design Delta。
日本語の概要は準備中です。原文の説明を表示しています。
竞品决策分析 / Competitive Decision Brief:当用户要调研竞品、替代方案、新产品、SaaS/AI 产品或市场信号, 并希望把官网、定价、评论、文档、更新日志、登录态走查、截图或浏览器取证转成产品定位、路线图、定价、 功能优先级、差异化、Go/No-Go 或 PRD 输入时使用。核心行为是先锚定产品决策,再选择证据渠道, 最后输出可行动的 Product Decision Brief;不要把它降级成泛泛功能清单、纯 UX 走查或“点完所有按钮”报告。
日本語の概要は準備中です。原文の説明を表示しています。
复杂探索资产化 / Complex Exploration:当用户面对复杂、不确定、多轮迭代的产品策略、Roadmap、商业化定价、竞品定位、复杂 PRD 前置探索、项目复盘或方法论沉淀任务时使用。它先判断任务类型,暴露隐含假设,重构真正问题,规划探索路径和中间产物,并在结束后沉淀认知、结构、方法论、工具和影响力资产。不用于简单润色、翻译、摘要、明确执行任务;问题只需早期认知校准时优先用 ai-collaboration-calibration,已有方案要压力测试时用 grill-me,系统专题研究用 research-topic-compiler。
日本語の概要は準備中です。原文の説明を表示しています。