重要前提
安装AI Skills的关键前提是:必须科学上网,且开启TUN模式,这一点至关重要,直接决定安装能否顺利完成,在此郑重提醒三遍:科学上网,科学上网,科学上网。查看完整安装教程 →
clarify%3Avague by team-attention/workshop-upstage
npx skills add https://github.com/team-attention/workshop-upstage --skill clarify:vague通过假设驱动式提问,将模糊或含糊的需求转化为精确、可操作的规范。始终使用 AskUserQuestion 工具——切勿以纯文本形式提出澄清性问题。
对于策略/规划中的盲点分析,请使用 unknown 技能。对于内容与形式的重新构建,请使用 metamedium 技能。
将可能的解释作为选项呈现,而不是提出开放式问题。每个选项都是关于用户实际意图的可测试假设。
错误示例:"你想要哪种登录方式?" ← 开放式问题,认知负荷高
正确示例:"OAuth / 邮箱+密码 / SSO / 魔法链接" ← 选择一个,认知负荷低
逐字记录原始需求。识别模糊之处:
使用 AskUserQuestion 来解决模糊之处。每次调用最多批量处理 4 个相关问题。 每个选项都是关于用户意图的一个假设。
上限:总共 5-8 个问题。 当所有关键的模糊之处都已解决,或者用户表示“足够好”,或者达到上限时,停止提问。
AskUserQuestion 调用示例:
questions:
- question: "登录应使用哪种身份验证方法?"
header: "身份验证方法"
options:
- label: "邮箱 + 密码"
description: "传统的注册方式,包含邮箱验证"
- label: "OAuth (Google/GitHub)"
description: "委托验证,无需管理密码"
- label: "魔法链接"
description: "基于邮箱的无密码登录"
multiSelect: false
- question: "注册后应该发生什么?"
header: "注册后"
options:
- label: "立即访问"
description: "用户可以立即使用应用"
- label: "先验证邮箱"
description: "必须在访问前确认邮箱"
multiSelect: false
广告位招租
在这里展示您的产品或服务
触达数万 AI 开发者,精准高效
呈现转化结果:
## 需求澄清总结
### 澄清前(原始需求)
"{原始请求逐字记录}"
### 澄清后(已明确)
**目标**:[精确描述]
**范围**:[包含与排除的内容]
**约束条件**:[限制、偏好]
**成功标准**:[如何知道何时完成]
**已做出的决策**:
| 问题 | 决策 |
|----------|----------|
| [模糊点 1] | [选择的选项] |
询问是否将澄清后的需求保存到文件。默认位置:requirements/ 或适合项目的目录。
| 类别 | 假设示例 |
|---|---|
| 范围 | 所有用户 / 仅管理员 / 特定角色 |
| 行为 | 静默失败 / 显示错误 / 自动重试 |
| 接口 | REST API / GraphQL / CLI |
| 数据 | JSON / CSV / 两者 |
| 约束条件 | <100ms / <1s / 无要求 |
| 优先级 | 必须拥有 / 最好拥有 / 未来 |
每周安装次数
57
代码仓库
GitHub 星标数
7
首次出现
2026年2月25日
安装于
claude-code57
opencode4
github-copilot4
codex4
kimi-cli4
amp4
Transform vague or ambiguous requirements into precise, actionable specifications through hypothesis-driven questioning. ALWAYS use the AskUserQuestion tool — never ask clarifying questions in plain text.
For strategy/planning blind spot analysis, use the unknown skill. For content-vs-form reframing, use the metamedium skill.
Present plausible interpretations as options instead of asking open questions. Each option is a testable hypothesis about what the user actually means.
BAD: "What kind of login do you want?" ← open question, high cognitive load
GOOD: "OAuth / Email+Password / SSO / Magic link" ← pick one, lower load
Record the original requirement verbatim. Identify ambiguities:
Use AskUserQuestion to resolve ambiguities. Batch up to 4 related questions per call. Each option is a hypothesis about what the user means.
Cap: 5-8 total questions. Stop when all critical ambiguities are resolved, OR user indicates "good enough", OR cap reached.
Example AskUserQuestion call:
questions:
- question: "Which authentication method should the login use?"
header: "Auth method"
options:
- label: "Email + Password"
description: "Traditional signup with email verification"
- label: "OAuth (Google/GitHub)"
description: "Delegated auth, no password management needed"
- label: "Magic link"
description: "Passwordless email-based login"
multiSelect: false
- question: "What should happen after registration?"
header: "Post-signup"
options:
- label: "Immediate access"
description: "User can use the app right away"
- label: "Email verification first"
description: "Must confirm email before access"
multiSelect: false
Present the transformation:
## Requirement Clarification Summary
### Before (Original)
"{original request verbatim}"
### After (Clarified)
**Goal**: [precise description]
**Scope**: [included and excluded]
**Constraints**: [limitations, preferences]
**Success Criteria**: [how to know when done]
**Decisions Made**:
| Question | Decision |
|----------|----------|
| [ambiguity 1] | [chosen option] |
Ask whether to save the clarified requirement to a file. Default location: requirements/ or project-appropriate directory.
| Category | Example Hypotheses |
|---|---|
| Scope | All users / Admins only / Specific roles |
| Behavior | Fail silently / Show error / Auto-retry |
| Interface | REST API / GraphQL / CLI |
| Data | JSON / CSV / Both |
| Constraints | <100ms / <1s / No requirement |
| Priority | Must-have / Nice-to-have / Future |
Weekly Installs
57
Repository
GitHub Stars
7
First Seen
Feb 25, 2026
Installed on
claude-code57
opencode4
github-copilot4
codex4
kimi-cli4
amp4
BMAD工作流编排与SSD结构化系统设计 - 多代理AI开发流程自动化
10,600 周安装
asciinema-analyzer:终端录制文件语义分析工具,支持关键词搜索与主题提取
67 周安装
Web界面规范检查工具 - 自动审查代码符合Web设计标准
67 周安装
NSFC标书质量控制工具nsfc-qc - LaTeX科研文档自动检查与引用验证
67 周安装
iOS MapKit API 参考指南:SwiftUI Map (iOS 17+) 与 MKMapView 完整对比与使用教程
72 周安装
Telos:个人与项目上下文管理系统 - 智能分析、目标追踪与自动化报告生成
57 周安装
Azure Developer CLI (azd) 容器应用部署指南:使用Bicep和远程构建
67 周安装