mcpField
B
blueprint-discovery

blueprint-discovery

github.com/majesticlabs-devAgent 工作流5065 次安装7 周前更新
代码仓库

关于 blueprint-discovery

用于引导功能发现流程的技能,从初始描述推进到需求澄清、研究信号、访谈决策、验收标准和分类,也涵盖仓库背景分析。

主要功能

  1. 01澄清含糊的功能构想
  2. 02判断研究与访谈需求
  3. 03收集验收标准
  4. 04按类型分类功能
  5. 05总结仓库结构与规范

适用场景

  1. 01澄清描述不充分的功能需求
  2. 02决定是否访谈相关人员
  3. 03定义可验证的功能行为
  4. 04为陌生代码库制定计划

安装

复制下面的命令,在终端中运行。

$npx skills add https://github.com/majesticlabs-dev/majestic-marketplace --skill blueprint-discovery

安装后在 Claude Code 中输入/blueprint-discovery确认已加载。

详细介绍

蓝图发现

处理蓝图工作流的第 1 至 5 步:完善想法、研究决策信号、访谈决策、收集验收标准以及功能分类。

输入

feature_description: string  # Raw feature description from user

1. 完善想法

在深入发现前快速澄清。通过 1-3 个有针对性的问题,尽早发现误解。

触发条件:

  • feature_description 少于 100 个字符
  • 包含不确定表述:“maybe”“probably”“something like”“I think”“not sure”
  • 缺少核心要素:没有明确的动作动词,或没有明确的主体

跳过条件:

  • 描述详细(超过 200 个字符且意图明确)
  • 用户说“proceed”或“skip refinement”
  • 有复现步骤的 Bug 修复

触发时:

AskUserQuestion:
  question: "Quick check - what's the primary goal?"
  header: "Goal"
  options:
    - label: "Add new capability"
      description: "Feature that doesn't exist yet"
    - label: "Fix broken behavior"
      description: "Something that should work but doesn't"
    - label: "Improve existing feature"
      description: "Enhancement to current functionality"
    - label: "Refactor/cleanup"
      description: "Better code without behavior change"

后续提问(如果回答仍有遗漏):

If goal == "Add new capability":
  AskUserQuestion:
    question: "Who will use this and when?"
    header: "Context"
    options:
      - label: "End users in the app"
      - label: "Admins/internal team"
      - label: "Developers/API consumers"
      - label: "Automated systems"

If goal == "Fix broken behavior":
  AskUserQuestion:
    question: "How does it fail?"
    header: "Symptom"
    options:
      - label: "Error/crash"
      - label: "Wrong output"
      - label: "Missing data"
      - label: "Performance issue"

最多提问 3 次。 完善后:

refined_description = original + goal + context/symptom (if asked)

提供跳过选项:

“明白了:{refined_description}。准备继续,还是需要进一步澄清?”

2. 研究决策信号

完善想法时,收集信号,以便在 blueprint-research 中做出研究决策。

从对话中推断:

信号检测方式
user_familiarity能指出现有代码示例?知道文件所在位置?→ high
user_intent“Quick fix”“ship today”→ speed;“want it right”“research first”→ thoroughness
topic_risk关键词:auth、payment、stripe、security、encrypt、API key、webhook → high
uncertainty_level“Not sure how”“what's the best way”,或正在探索选项 → high

如果信号不明确,可快速询问:

AskUserQuestion:
  question: "What matters more for this task?"
  header: "Priority"
  options:
    - label: "Get it done fast"
      description: "Good enough solution, ship quickly"
    - label: "Get it done right"
      description: "Research best practices first"

保存信号,供研究阶段使用。

3. 访谈决策

建议进行访谈的情况:

  • 功能描述少于 2 句话
  • 包含不确定词语:“maybe”“probably”“something like”“not sure”
  • 涉及多个利益相关方或系统
  • 用户似乎不确定

跳过访谈的情况:

  • 有明确复现步骤的 Bug 修复
  • 小型且定义明确的任务(可能涉及少于 3 个文件)
  • 引用了现有规格说明/PRD 的功能

如果建议访谈:

AskUserQuestion:
  question: "This feature could benefit from a requirements interview. Explore in depth first?"
  options:
    - "Yes, interview me first" → Invoke /interview with feature_description
    - "No, proceed to planning" → Continue

4. 验收标准

必须询问“完成”的定义。

验收标准仅描述功能行为。质量门槛(测试、lint、评审)由其他 Agent 处理。

AskUserQuestion:
  question: "What behavior must work for this feature to be done?"
  header: "Done when"
  multiSelect: true
  options:
    - label: "User can perform action"
      description: "Feature enables a specific user action"
    - label: "System responds correctly"
      description: "API/backend behaves as expected"
    - label: "UI displays properly"
      description: "Visual elements render correctly"
    - label: "Data is persisted"
      description: "Changes are saved to database"

良好的验收标准示例:

  • “已认证用户可以登录并跳转到仪表盘”
  • “表单会在提交前验证电子邮件格式”
  • “API 对不存在的资源返回 404”

不佳的验收标准示例(由其他 Agent 处理):

  • “测试通过”→ always-works-verifier
  • “代码已评审”→ quality-gate
  • “没有 lint 错误”→ slop-remover

记录每项标准的验证方法:

标准验证方式
用户可以登录curl -X POST /login 或手动验证
表单会验证rspec spec/features/signup_spec.rb
API 返回 404curl /api/nonexistent

5. 功能分类

类型检测关键词操作
UIpage、component、form、button、modal、design、view、template检查设计系统
DevOpsterraform、ansible、infrastructure、cloud、docker、deploy、server委派给 devops-plan
APIendpoint、route、controller、request、response、REST、GraphQL标准流程
Datamigration、model、schema、database、query标准流程

UI 功能流程:

  1. 读取配置:config_read("design_system_path")
  2. 如果为空,检查:docs/design/design-system.md
  3. 如果没有设计系统:先建议使用 Skill("ux-brief")

DevOps 功能流程:

Skill(skill: "majestic-devops:devops-plan")

6. 仓库分析(入门背景)

处理不熟悉的代码库时,在规划功能前收集结构背景信息。

研究领域
领域检查内容
架构ARCHITECTURE.md、README.md、CONTRIBUTING.md、CLAUDE.md、AGENTS.md
Issue/PR 模式.github/PULL_REQUEST_TEMPLATE*、.github/ISSUE_TEMPLATE/
贡献指南编码标准、测试要求、评审流程
代码库模式命名约定、模块边界、实现模式
结构分析工具
  • 当文本搜索无法满足需求时,通过 Bash 使用 ast-grep 进行语法感知的结构匹配:
    ast-grep --pattern 'class $NAME < ApplicationRecord' --lang ruby
    
  • 交叉比对不同来源中的发现
  • 优先参考官方文档,而非推断出的模式
  • 记录不一致之处或文档缺失
代码库分析输出
## Repository Research Summary

### Architecture & Structure
- Project organization and tech stack
- Key architectural decisions

### Conventions
- Issue/PR formatting and labels
- Coding standards and testing requirements

### Implementation Patterns
- Common code patterns and naming conventions
- Project-specific practices

### Recommendations
- How to align with project conventions
- Areas needing clarification

输出

discovery_result:
  refined_description: string  # Original + refinement context
  refinement_skipped: boolean
  interview_conducted: boolean
  interview_output: string | null  # If interview was run
  acceptance_criteria:
    - criterion: string
      verification: string
  feature_type: "ui" | "devops" | "api" | "data" | "general"
  design_system_path: string | null  # For UI features
  # Research decision signals (for blueprint-research)
  user_familiarity: high | medium | low
  user_intent: speed | thoroughness
  topic_risk: high | medium | low
  uncertainty_level: high | medium | low
  ready_for_research: boolean

评论

支持 Markdown