shield_lock
金大哥
← 返回 Skill 列表
extension
分类: 开发与工程无需 API Key

research-tools

用于研究主题、查找YouTube视频、使用NotebookLM分析来源或进行需求收集访谈。触发条件包括：研究、找视频、YouTube搜索、分析来源、NotebookLM、研究流程、深度访谈、采访我、需求收集、苏格拉底式提问、yt-dlp、文字记录、音频概览、信息图、思维导图、学习指南、常见问题解答、简报文档、查找并分析、为我做研究、分析这些视频、竞争性研究。

person作者: jakexiaohubgithub
open_in_new仓库download下载资源包

Research Tools

Route research, analysis, and requirements-gathering tasks to the correct workflow.

Task Router

| Task | Reference | Key Actions | |------|-----------|-------------| | Analyze sources with NotebookLM, generate deliverables | references/notebooklm-workflow.md | Create notebook, add sources, ask questions, generate audio/infographic/FAQ | | Search YouTube for videos, fetch transcripts | references/youtube-search.md | yt-dlp search, structured results table, VTT transcript extraction | | End-to-end research: YouTube -> NotebookLM -> Obsidian | references/research-pipeline.md | Chain all 3 tools, save structured notes to vault | | Socratic requirements gathering, spec crystallization | references/interview-technique.md | Ambiguity scoring, challenge agents, spec document generation |

Quick Decision

  • "Analyze these sources" -> NotebookLM workflow
  • "Find videos about X" -> YouTube search
  • "Research X for me" / "find and analyze" -> Full pipeline
  • "Interview me" / "I have a vague idea" -> Deep interview
  • URL provided directly -> Skip search, go to NotebookLM or transcript

MCP Tools Required

Use ToolSearch at task start to load:

  • mcp__notebooklm__* — NotebookLM notebook management and questions
  • mcp__obsidian__* — Obsidian vault note creation (for pipeline saves)

Shared Principles

  • Always return full YouTube URLs so they can be passed to NotebookLM as sources
  • NotebookLM handles its own transcription — do not pre-fetch transcripts when passing to NotebookLM
  • Visual deliverables (infographic, mind map) can take 5-15 minutes — set expectations
  • Deep interview: one question at a time, target weakest clarity dimension, never batch
  • Research pipeline default: 10 videos, analysis only (escalate to deliverable if requested)

Layer 3 Skills

  • notebooklm-research — NotebookLM MCP workflow for source analysis and deliverable generation
  • youtube-research — YouTube video search and transcript extraction via yt-dlp
  • research-pipeline — Full YouTube -> NotebookLM -> Obsidian orchestration
  • deep-interview — Socratic requirements gathering with mathematical ambiguity scoring