deep-research

使用顺序思维和并行代理进行深入研究。

📖 中文导读:使用顺序思维和并行代理进行深入研究。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

name: deep-research description: Deep research with sequential thinking and parallel agents

Deep Research: $ARGUMENTS

Execute comprehensive multi-angle research using structured thinking and parallel investigation.

Phase 1: Decompose

Use mcp__MCP_DOCKER__sequentialthinking to break the topic into 3-5 independent research aspects:

  • Identify key questions per aspect
  • Define what "understanding" looks like for each
  • Create focused research prompts

ultrathink

Phase 2: Parallel Research

Launch one Task agent per aspect simultaneously (in a single message):

Task tool:
  subagent_type: "Explore"
  run_in_background: true
  prompt: "Research [aspect]: [questions]. Return key findings and insights."

Wait for all agents with TaskOutput (block: true).

Phase 3: Synthesize

Use mcp__MCP_DOCKER__sequentialthinking to:

  • Cross-reference findings from all agents
  • Identify patterns, contradictions, connections
  • Form conclusions

Phase 4: Present

Structure response as:

  • Key Findings (bullets)
  • Insights & Connections (cross-cutting themes)
  • Conclusions (synthesis)
  • Next Steps (if applicable)

Constraints

  • Minimum 3 agents, maximum 5
  • Always use sequential thinking for decomposition AND synthesis
  • Output in chat only