English | 한국어 | 中文 | 日本語 | Español

insane-research

insane-research — AI-powered multi-agent deep research with source verification and structured outputs.

Turn a single question into a comprehensive, citation-backed research report — automatically.

Quick Start • Why insane-research? • How it works • Commands • Output • Requirements

---

Quick Start

1. Add the marketplace (once)

/plugin marketplace add https://github.com/fivetaku/gptaku_plugins.git

2. Install the plugin

/plugin install insane-research

3. Restart Claude Code

Cache loads on startup — a restart is required after install.

4. Start researching

/insane-research AI coding assistants productivity impact

Claude will ask a few scoping questions, then deploy parallel research agents and deliver a structured report.

---

Why insane-research?

• Parallel agents, not sequential searches — 3-5 agents run simultaneously across web, academic, and technical sources, cutting research time significantly
• Source quality ratings (A–E) — Every source is graded from peer-reviewed papers (A) to speculative posts (E), so you always know what you're reading
• Hallucination-resistant — Every factual claim requires an inline citation; key claims are cross-verified against at least 2 independent sources
• Resumable sessions — Research state is saved to state.json; pick up where you left off if a session is interrupted
• Complete deliverables — Executive summary, full sectioned report, bibliography, and an optional interactive website — all generated automatically

---

How it works

User query
    │
    ▼
Phase 1: Question Scoping
  └─ AskUserQuestion → focus, depth, audience, sources
    │
    ▼
Phase 2: Retrieval Planning
  └─ Break into 3-5 subtopics → search query generation → plan approval
    │
    ▼
Phase 3: Iterative Querying  ←──────────────────┐
  ├─ Web Research Agent (x2-3)                  │
  ├─ Academic/Technical Agent (x1-2)            │ refine if gaps
  └─ Cross-Reference Agent (x1)                 │
    │                                            │
    ▼                                            │
Phase 4: Source Triangulation ─────────────────-┘
  └─ Cross-verify key claims (≥2 sources) → A–E quality rating
    │
    ▼
Phase 5: Knowledge Synthesis
  └─ Structure → write sections → inline citations
    │
    ▼
Phase 6: Quality Assurance
  └─ Hallucination check → citation verification → completeness
    │
    ▼
Phase 7: Output & Packaging
  └─ Executive summary + full report + bibliography + website (optional)

---

Commands

Natural language triggers

deep research on [topic]
research [topic]
[topic] 리서치해줘
딥리서치 [주제]
심층 연구 [주제]

---

Agents

Three agent types run in parallel during Phase 3:

---

Source Quality Ratings

---

Output Structure

RESEARCH/{topic}_{timestamp}/
├── state.json                    # Session state (for resume)
├── README.md                     # Navigation guide
├── outputs/
│   ├── 00_executive_summary.md   # 3–5 page summary
│   ├── 01_full_report/           # Full sectioned report
│   ├── 02_appendices/            # Supporting material
│   └── comparison_data.json      # Structured comparison data
├── sources/
│   ├── sources.jsonl             # Collected sources
│   ├── bibliography.md           # Formatted bibliography
│   └── quality_report.md         # Source quality ratings
└── website/                      # (optional) Interactive presentation
    ├── index.html
    ├── styles.css
    └── script.js

---

Requirements

• Claude Code CLI
• WebSearch (built-in) or a web search MCP server

Optional MCP servers (enhance search coverage)

• Firecrawl
• Google Search MCP
• Exa Search

---

License

MIT

---