AgentSkillsCN

Deep Research

深度研究

SKILL.md

Deep Research Skill

Type: Composite (Manager + Workers)
Source: agentfile/examples/research-pipeline
Status: ⚠️ Requires exa MCP server

Quick Use

code
Load deep-research skill.
Research: "your topic here"

This will:

  1. •Decompose topic into aspects
  2. •Research each aspect in parallel
  3. •Synthesize findings
  4. •Quality check
  5. •Generate final report

Output

All artifacts written to ~/clawd/artifacts/{session_id}/:

  • •plan.yaml — research plan
  • •aspects/*.yaml — individual findings
  • •synthesis.yaml — aggregated insights
  • •quality.yaml — quality metrics
  • •FINAL_REPORT.md — final deliverable

Setup Required

FIRST TIME: Follow ~/clawd/skills/deep-research/SETUP.md

Critical: Install exa MCP server with API key.

Options

yaml
depth: quick|medium|deep  # 3/5/7 aspects
max_sources: 15           # per aspect
quality_threshold: 50     # saturation %

Files

  • •manager-research.md — orchestration logic
  • •agents/aspect-researcher.md — worker agent
  • •agents/report-generator.md — report writer
  • •skills/* — supporting skills

How It Works

  1. •ROOT loads manager-research.md
  2. •Manager spawns workers for each aspect
  3. •Workers use exa MCP for search + crawl
  4. •Results aggregate → synthesis → quality gate
  5. •Final report generated

Architecture

code
ROOT (manager-research)
  │
  ├─→ research-planner (skill)
  │    └─→ plan.yaml
  │
  ├─→ aspect-researcher × N (workers)
  │    └─→ aspects/*.yaml
  │
  ├─→ synthesis (skill)
  │    └─→ synthesis.yaml
  │
  ├─→ quality-gate (skill)
  │    └─→ quality.yaml
  │
  └─→ report-generator (worker)
       └─→ FINAL_REPORT.md

See Also

  • •SETUP.md — installation guide
  • •manager-research.md — full orchestration spec
  • •~/clawd/custom/references/agentfile.md — framework overview