By name: _deep-research description: > Use when users need in-depth research on a complex topic or a comprehensive report requiring multiple sources and synthesis. Avoid for simple fact-finding, single-source lookups, or quick exploratory searches.
Deep Research Skill
Core Purpose
Transforms research questions into detailed reports using a three-agent system: a lead orchestrator, multiple researcher agents executing parallel searches, and a report-writer synthesizing findings.
Agent Architecture
Lead Agent (Orchestrator): Interviews users, plans research threads, coordinates subagents using Agent and AskUserQuestion tools.
Researcher Agents: Execute focused investigations on assigned subtopics, employing WebSearch, WebFetch, and Write tools to save structured notes.
Report-Writer Agent: Synthesizes research notes into final reports using Read, Glob, and Write tools exclusively.
Five-Phase Research Process
Phase 1: Interview 2-3 rounds of user interviews covering objectives, depth, audience, key questions, time constraints, and scope.
Phase 2: Landscape Mapping
3-5 broad searches to map the topic landscape, identifying 10+ research threads, documented in research_plan.md.
Phase 3: Parallel Research
Launch 10+ researcher agents in parallel, each saving findings to research_notes/[subtopic].md with:
- Summary
- Key findings
- Sources with URLs
- Notable quotes
- Identified gaps
Phase 4: Synthesis Spawn report-writer agent to synthesize notes into a comprehensive report with:
- Executive summary
- Critical analysis organized by theme
- Numbered citations
- Identified limitations and conflicting information
Phase 5: Delivery Deliver output files and summarize key findings for user follow-up.
Quality Standards
- Prioritize authoritative, recent sources
- Cross-reference claims across multiple researcher notes
- Distinguish facts from expert opinions
- Clearly identify conflicting information or research limitations
- Include direct URLs for all cited sources