Token导航 LogoToken导航TokenDH.com
研究检索external-servicegithub未标认证来源可访问clear审计提醒

brainstorming-ideas集思广益

Agent Skill

brainstorming-ideas 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

470

周安装

19

GitHub Stars

17

下载量

147
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

3

许可证

MIT

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:brainstorming-ideas(集思广益)
来源仓库:https://github.com/alexei-led/claude-code-config
仓库路径:skills/brainstorming-ideas
安装命令:
npx skills add https://github.com/alexei-led/claude-code-config --skill brainstorming-ideas
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。不同来源提供的安装方式可能略有差异;本站展示可直接复制的安装命令,安装前请核对来源页面。

skills.shnpx skills
npx skills add https://github.com/alexei-led/claude-code-config --skill brainstorming-ideas

简介

通过七步对话流程将模糊 idea 精炼为可执行设计文档,杜绝 premature implementation。

  • 每阶段只提一个问题,优先选择题型交互以提高效率与准确性。
  • 使用 TodoWrite 跟踪进度,确保每个交付物都经过用户显式确认。
  • 特别强调不要在没有设计稿的情况下直接写代码,降低返工概率。
  • brainstorming-ideas 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

Brainstorming Ideas Into Designs

Transform vague ideas into fully-formed designs through structured collaborative dialogue.

Use TodoWrite to track these 7 phases:

  1. Understand the idea (dialogue first, no agents)
  2. Explore requirements (Starbursting questions)
  3. Checkpoint - offer exploration/research options
  4. Research similar solutions (if requested)
  5. Present approaches with recommendation
  6. Validate design incrementally
  7. Document and next steps

Core Principles

  • Dialogue first - Ask the user before spawning any agents
  • One question at a time - Never batch multiple questions
  • Multiple choice preferred - Easier to answer than open-ended
  • "Other" always available - Free text input for custom responses
  • YAGNI ruthlessly - Challenge every feature's necessity
  • Incremental validation - Present design in 200-300 word sections
  • Agents on request - Only explore/research when user chooses it

Phase 1: Understand the Idea

Start with dialogue, not agents. Ask the user directly.

1a. Initial Question

Use AskUserQuestion:

HeaderQuestionOptions
Idea typeWhat would you like to brainstorm?1. New feature - Add new functionality 2. Modification - Change existing behavior 3. Integration - Connect with external system 4. Exploration - Not sure yet, let's discover

1b. Follow-up (based on response)

Ask clarifying questions using AskUserQuestion. Keep it conversational:

  • "Can you describe this in a sentence or two?" (free text via "Other")
  • "What triggered this idea?" with context-appropriate options
  • "Is there an existing feature this builds on?"

Phase 2: Explore Requirements (Starbursting)

Ask questions one at a time using AskUserQuestion. Adapt based on idea type.

Question Framework (5WH)

Question TypeWhen to AskExample AskUserQuestion
WHOAlways first"Who will use this?" → Options: Existing users, New segment, Internal, API consumers
WHYAfter WHO"What problem does this solve?" → Options based on detected pain points
WHATAfter WHY is clear"What's the core capability?" → Open or options based on research
WHEREFor integrations/modifications"Where should this live?" → Options based on codebase exploration
HOWAfter approach research"How should we implement?" → Present 2-3 technical approaches

Adaptive Questioning

  • Skip questions when answers are obvious from context
  • If user seems certain, move faster to approaches
  • If user seems uncertain, explore deeper with sub-questions
  • Use "Other" option to allow custom responses

2b. Surface Assumptions

Before moving on, explicitly list the assumptions embedded in the idea so far:

Based on our discussion, here are the assumptions I'm seeing:

1. [Assumption about users] — e.g., "Users want X in real-time"
2. [Assumption about tech] — e.g., "Our current DB can handle this load"
3. [Assumption about scope] — e.g., "This doesn't need to work offline"

Use AskUserQuestion:

HeaderQuestionOptions
AssumptionsAny of these assumptions wrong or risky?1. All correct - Proceed as-is 2. Some wrong - I'll clarify 3. Not sure - Let's validate the risky ones during research

If "Some wrong": ask which ones and adjust requirements. If "Not sure": flag risky assumptions for verification in Phase 4.


Phase 3: Checkpoint - Gather More Context?

After understanding requirements, ask before spawning any agents:

HeaderQuestionOptions
Next stepHow should we proceed?1. Explore codebase - Check existing patterns and tech stack 2. Research solutions - Look up how others solve this 3. Check project history - Query past decisions on this topic (claude-mem) 4. Both - Explore then research 5. Skip to approaches - I know what I want

If user chooses "Explore codebase":

Task(
  subagent_type="Explore",
  prompt="Quick scan: project structure, tech stack, patterns relevant to [user's idea]",
  run_in_background=false
)

Then summarize findings and ask: "Based on this, should we also research external solutions?"

If user chooses "Check project history":

If claude-mem tools are available:

search({ query: "[topic keywords]", limit: 10 })
timeline({ query: "[topic keywords]", depth_before: 5, depth_after: 5 })

Summarize past decisions, known issues, and relevant context. Then ask if they want to also explore or research.

If user chooses "Research solutions":

Proceed to Phase 4.

If user chooses "Skip to approaches":

Jump directly to Phase 5 (Present Approaches).


Phase 4: Research Similar Solutions (If Requested)

Only run when user explicitly chose research in Phase 3.

4a. Perplexity Query

mcp__perplexity-ask__perplexity_ask({
  messages: [{
    role: "user",
    content: "How do leading [industry] products implement [feature type]? Include architectural patterns, UX approaches, and trade-offs. Focus on [tech stack] implementations."
  }]
})

4b. Follow Citations

After Perplexity response, WebFetch top 2-3 relevant sources:

WebFetch(url="<citation-url>", prompt="Extract implementation details, code patterns, and lessons learned for [feature]")

4c. Synthesize Findings

Present research summary before asking approach preference:

## Research Findings

**Common patterns:**

- [Pattern 1]: Used by X, Y. Trade-off: ...
- [Pattern 2]: Used by Z. Trade-off: ...

**Recommended for our context:** [Pattern] because [reasons]

Phase 5: Present Approaches

Use AskUserQuestion with 2-4 options:

HeaderQuestionOptions
ApproachWhich approach fits best?1. [Recommended] - Description + key trade-off 2. [Alternative] - Description + key trade-off 3. Minimal - YAGNI version

Approach Template

For each option, briefly cover:

  • What: Core implementation
  • Trade-offs: Complexity vs flexibility, Now vs later
  • Fits when: Scenario where this shines

Phase 6: Validate Design Incrementally

Present design in sections (~200-300 words each). After each section, use AskUserQuestion:

HeaderQuestionOptions
ValidateDoes this [section] look right?1. Yes, continue - Move to next section 2. Needs changes - I'll explain 3. Go back - Revisit earlier decisions

Design Sections

  1. Architecture Overview - Components, responsibilities, relationships
  2. Data Flow - How information moves through the system
  3. API/Interface - External contracts and user interactions
  4. Error Handling - Failure modes and recovery strategies
  5. Testing Strategy - How to verify it works

YAGNI Checkpoints

At each section, actively challenge:

  • "Do we need this now, or is it speculative?"
  • "What's the simplest version that solves the problem?"
  • "Can we defer this complexity?"

Phase 7: Document and Next Steps

7a. Write Design Document

Write(
  file_path="docs/plans/YYYY-MM-DD-<topic>-design.md",
  content="# [Feature] Design\n\n## Problem\n...\n## Solution\n...\n## Architecture\n..."
)

7b. Commit Design

git add docs/plans/*.md && git commit -m "docs: add [feature] design document"

7c. Implementation Handoff

Use AskUserQuestion:

HeaderQuestionOptions
Next stepsReady to proceed with implementation?1. Create worktree - Isolated workspace via using-git-worktrees 2. Create plan - Detailed implementation steps 3. Done for now - Just save the design

Methodology Reference

This skill incorporates proven brainstorming techniques:

TechniqueHow It's Used
Starbursting (5WH)Structured questions in Phase 2
Design ThinkingEmpathize (context) → Define (WHY) → Ideate → Prototype (design sections)
SCAMPERFor modifications: Substitute, Combine, Adapt, Modify, Put to other use, Eliminate, Reverse
Reverse Brainstorming"How could this fail?" during validation
Mind MappingArchitecture section visualizes relationships

Examples

/brainstorming-ideas                    # Start open-ended brainstorm
/brainstorming-ideas user notifications # Brainstorm notification feature
/brainstorming-ideas auth flow          # Brainstorm authentication changes

Execute this collaborative brainstorming workflow now.

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

04

需要参考平台分布和安装热度时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

28.71%
按下载量换算42

Antigravity

23.23%
按下载量换算34

Claude Code

17.32%
按下载量换算25

Gemini CLI

13.01%
按下载量换算19

windsurf

7.74%
按下载量换算11

OpenCode

2.95%
按下载量换算4

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

权限和风险

external-service

该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在 warning/failed 结果,不能写成本站确认安全。

来源信息

继续浏览同类 Skills