Multi Agent Deep Research — Development skill for Claude Code

Multi-Agent Deep Research

An open-source alternative to Google Gemini Deep Research. Uses a **4-role Agent collaborative architecture** that simulates how a real consulting team works, automatically generating in-depth research reports with academic citation footnotes.

Highlights

• 4-Role Architecture: Agent-manager (planning + QA + delivery) + Agent-researcher (search + data) + Agent-report consultant (writing + visualization self-check) + Agent-data reviewer (independent data verification)
• 5 Decision Gates: User and Agent-manager can send work back upstream at key checkpoints to ensure quality. Plus mandatory visualization compliance check before output
• Visualization Gate: Prevents text-only reports — forces trend data into charts, hierarchies into diagrams, timelines into structured tables. Pandoc mechanical conversion explicitly forbidden
• Source Priority Pyramid: Annual reports/broker research > government data > authoritative media > corporate websites, with strict blacklist filtering
• Anti-Hallucination: research_data.json persistence + Agent-data reviewer independent review + scripts forbidden from hardcoding numbers
• Data Freshness Gates: Every chart/table caption carries a data version anchor (period + metric definition); forecasts auto-expire when actuals land; listed companies require the annual report / Q1 / interim report / IR-minutes checklist with a "today vs latest reporting period" check
• Supplement-Round Diff Gate: Re-issuing a report on new data is a version refresh, not a chapter append — every new data point triggers an impact list, and every old chart must be explicitly keep / upgrade / declare-gap rescanned before publication
• Dual Format Output: .docx (with citation superscripts + bibliography) + .pptx (thesis-statement-driven consulting-grade slides)
• Consulting Framework Library: IBM Five-Forces by default, supports PEST/Porter's Five Forces/SWOT/3C/BCG Matrix
• Completely Free: Use