Deep Research playbook (translated)
Skills used in this guide (3)
构建证据可追溯的市场研究报告，以及基于假设的细分或预测情景。用于市场定义、行业与客户证据、竞争格局、TAM/SAM/SOM 对账、预测敏感性，以及可审计的报告骨架。
npx skills add k-dense-ai/claude-scientific-skills -g读研究大纲，为每个条目派独立 agent 深度研究。禁用 task output。
npx skills add Weizhena/Deep-Research-skills -g上游市场研究方法论，自上而下与自下而上双算 TAM/SAM/SOM、调研样本量与细分评分。
npx skills add alirezarezvani/claude-skills -g
Deep Research compresses hours-to-days analysis into tens of minutes: you give the question, the agent searches in steps, cross-checks, and returns a sourced structured report. Community method stresses: break the question fine first, cross-verify from multiple sources, every claim traceable — fast but trustworthy.
Who this is for / not for
For: market / academic / competitive deep research needing citations. Not for: those asking AI to invent data, or quoting without checking sources.
Step-by-step
1. Break the question fine
Don't drop one big word; split into searchable sub-questions:
Split "XX market opportunity" into: size, players, pain, moat, risk — 5 sub-questions, each with search leads.2. Multi-agent search
research-deep runs multi-step search, market-research orchestrates, covers sub-questions in parallel.
Search the 5 sub-questions in parallel, each with 3 sources and a one-line takeaway.3. Cross-verify
Every claim backed by ≥2 independent sources; conflicts listed separately, not forced to agree.
For each key claim mark source count and agreement; list conflicts separately, don't force.4. Structured report
market-research-reports outputs sections, data tables, citations.
5. Traceability check
Spot-check 3 citations actually open and are read correctly.
Pitfalls
| Pit | Symptom | Fix |
|---|---|---|
| Too broad | Vague | Break fine |
| Single source | Biased | Cross-verify |
| No cite | Untrust | Force trace |
| Invented | False | You check |
FAQ
How fast?
Tens of minutes.
Search yourself?
Spot-check cites.
Academic use?
Yes, with cites.
Conflict?
List, don't force.
相关源码
- How AI redefines research (jimo.studio survey)
- Deep Research Agents: A Systematic Examination (arXiv 2506.18096)
- Universal Agent Research Methodology (CHINGBOH, GitHub)
一句话结论
结论：Deep Research 先把问题拆细、多源交叉验证、每条结论可溯源，快且可信；你抽查引用是否真能打开、是否被误读。