deep-research

name	deep-research
description	Investigate a topic through web search and multi-agent synthesis, archive raw sources, and deliver a structured research report. USE FOR: "research", "investigate", "deep dive", DO NOT USE FOR: brief factual answers, single-document summaries, processing local materials.
allowed-tools	Task, Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch

Deep Research

Responsible for external information research, source archiving, and synthesis reporting. Does not handle bulk digestion of local materials.

USE FOR

• User explicitly requests "research / investigate / deep dive"
• Requires multi-round search, source archiving, and structured synthesis
• Results need to form a traceable research directory and report

DO NOT USE FOR

• Only needs a brief factual answer
• Summarizing a single document
• Primary task is processing locally available materials, not expanding external sources

Execution Skeleton

1. Decompose the research question into dimensions; define the research directory and subtopics.
2. Multi-round search, fetch, and archive raw sources; then write subtopic findings.
3. Synthesize into a final report.md, marking sources and gaps.
4. Directory logic, error handling, and supervisor heuristics are in execution.

Reference Map

• decomposition: Topic decomposition templates and dimension checklist
• formats: Sub-agent prompt, findings structure, and archival format
• tools: Tool selection for each research phase
• examples: End-to-end examples
• execution: Workflow, directory logic, error handling, and best practices
• trigger-cases: Minimal trigger examples
• execution-cases: Key execution scenarios
• report-en / report-zh: Final report templates

Output Contract

• Default output includes research directory, archived sources, per-subtopic findings, and a final report
• Key conclusions must be traceable to specific sources

Collaboration and Handoff

• If long-form writing or presentation design is needed after research, hand off to the appropriate skill