ClawKit Logo
ClawKitReliability Toolkit
Back to Registry
Official Verified

deep-research

Autonomous multi-model deep research with framework-driven reasoning. Spawns 4 parallel model agents (Gemini 2.5 Pro, o3, Opus, MiniMax), each applies best-practice frameworks to the question, then merges into a cross-validated final report. Use when: (1) user asks for in-depth research, (2) 'research X' or 'deep dive on X', (3) complex questions requiring multiple sources. NOT for: simple factual lookups.

skill-install — Terminal

Install via CLI (Recommended)

clawhub install openclaw/skills/skills/leadingot/opusflame-deep-research
Or

Deep Research (Multi-Model + Framework-Driven)

Autonomous research system that runs 4 AI models in parallel, each applying relevant analytical frameworks, then cross-validates and merges findings into a comprehensive cited report.

Architecture

User Question
     │
     ▼
┌─ Phase 0: Framework Selection ─┐
│  Identify best-practice         │
│  framework(s) for this question │
└────────────┬────────────────────┘
             │
     ┌───────┼───────┐───────┐
     ▼       ▼       ▼       ▼
  Gemini    o3     Opus   MiniMax
  2.5 Pro         4       M2.5
  (search  (deep  (nuance (China/
  heavy)   logic) +balance)alt view)
     │       │       │       │
     └───────┼───────┘───────┘
             ▼
      Phase 5: Merge & Cross-Validate
             │
             ▼
       Final Report (PDF)

Phase 0: Framework Selection (MANDATORY — before any research)

Before researching, ask: "Is there a best-practice framework for answering this type of question?"

Framework Lookup Table

Question TypeFrameworks to Apply
Competitive strategyPorter's Five Forces, 7 Powers (Helmer), Schwerpunkt/High Ground (Packy), SWOT
Market entry / sizingTAM/SAM/SOM, Blue Ocean Strategy, Jobs-to-be-Done
Business model evaluationBusiness Model Canvas, Unit Economics, Ramp vs Route test (point solution vs platform?)
Investment / valuationDCF, Comparable Analysis, Venture method, Power Law thesis
Product strategyJTBD, Kano Model, Value Prop Canvas, Hook Model
Growth / GTMAARRR Pirate Metrics, Bullseye Framework, STP (Segmentation-Targeting-Positioning)
Technology assessmentGartner Hype Cycle, Wardley Maps, Build vs Buy matrix
Risk analysisPre-Mortem, FMEA, Scenario Planning
Organizational / opsOKR analysis, RACI, Theory of Constraints
PricingVan Westendorp, Conjoint, Value-based pricing framework
Industry analysisValue Chain Analysis, Industry Lifecycle, Winner-Takes-More thesis
Person / hiringTrack Record Analysis, Reference Triangle, Founder-Market Fit

If a framework applies:

  • Include it in the prompt to each model
  • Structure the model's analysis around the framework's components
  • The final report should explicitly reference which framework(s) were used and why

If no standard framework applies:

  • State "No standard framework identified — using first-principles analysis"
  • Each model reasons from first principles with explicit assumptions stated

Phase 1: Decompose (30s)

Break the topic into 5-8 research sub-questions. Think like an investigative journalist:

  • What are the key facts?
  • What are different perspectives/sources?
  • What's the timeline/history?
  • What data/evidence exists?
  • What are the unknowns or controversies?

Phase 2: Spawn 4 Model Agents (Parallel)

Spawn 4 sub-agents using sessions_spawn, each with a different model:

Metadata

Author@leadingot
Stars1656
Views2
Updated2026-02-28
View Author Profile
AI Skill Finder

Not sure this is the right skill?

Describe what you want to build — we'll match you to the best skill from 16,000+ options.

Find the right skill
Add to Configuration

Paste this into your clawhub.json to enable this plugin.

{
  "plugins": {
    "official-leadingot-opusflame-deep-research": {
      "enabled": true,
      "auto_update": true
    }
  }
}
Safety NoteClawKit audits metadata but not runtime behavior. Use with caution.