An autonomous multi-model research system that orchestrates four parallel AI agents to generate framework-driven, cross-validated technical reports.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install opusflame-deep-research
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install opusflame-deep-research using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Deep Research?
The Deep Research skill is a sophisticated orchestration layer designed for Openclaw Skills to perform exhaustive, multi-perspective investigations. By spawning four specialized AI models in parallel—Gemini 2.5 Pro for broad search, OpenAI o3 for deep logic, Anthropic Opus for nuanced synthesis, and MiniMax for alternative viewpoints—it eliminates single-model bias and ensures high data integrity through rigorous cross-validation.
Unlike standard search tools, this skill integrates industry-standard analytical frameworks like Porter's Five Forces, TAM/SAM/SOM, and Wardley Mapping into the core reasoning process. It doesn't just aggregate data; it synthesizes findings into a professional-grade report, complete with an agreement matrix and confidence scoring, providing developers and analysts with a definitive source of truth for complex decision-making.
Deep Research Use Cases
- Executing comprehensive competitive strategy analysis using Porter's Five Forces or SWOT frameworks.
- Performing market entry and sizing research via TAM/SAM/SOM and Blue Ocean Strategy models.
- Conducting technical assessments and build vs. buy matrices for new software architecture.
- Developing investment theses and valuation reports using DCF or comparable analysis.
- Evaluating business models through unit economics and platform-vs-point-solution testing.
How Deep Research Works
- The system identifies the optimal analytical framework (e.g., SWOT, OKR, or JTBD) based on the user's research query.
- The primary agent decomposes the topic into 5-8 investigative sub-questions to ensure granular coverage.
- Openclaw Skills spawn four parallel model agents (Gemini, o3, Opus, MiniMax), each tasked with a specific research persona and minimum search requirements.
- Agents perform extensive web searches and fetches, citing a minimum of 15 unique sources each.
- The primary agent collects the individual reports and performs a cross-validation process, creating an agreement matrix to flag discrepancies.
- A final report is synthesized, merged, and converted into a PDF for delivery, ensuring all framework components are addressed with high confidence levels.
Deep Research Setup
To implement this skill within your Openclaw Skills environment, ensure you have API keys configured for Google (Gemini), OpenAI (o3), Anthropic (Opus), and MiniMax.
Install the necessary dependencies for PDF generation:
pip install pymupdf
Configure the skill by placing the SKILL.md in your agent's skills directory and ensuring the sessions_spawn tool is enabled in your runtime configuration.
Deep Research Data Schema & Taxonomy
The skill maintains a structured memory hierarchy to ensure research persistence and traceability. Findings are organized into Markdown files and final outbound PDFs.
| File Path | Description |
|---|---|
memory/research/[topic]-[model]-[date].md |
Raw research output from an individual model agent. |
memory/research/[topic]-终极版-[date].md |
The final merged and cross-validated Markdown report. |
~/.openclaw/media/outbound/[topic].pdf |
The final delivered PDF document generated via PyMuPDF. |
Metadata includes total search counts, source URLs (targeting 60+ across all models), and model-specific confidence scores.
Deep Research Advanced Features
- Parallel agent spawning via
sessions_spawnfor massive efficiency gains over sequential processing. - Automated Agreement Matrix generation to identify model-specific hallucinations or logic gaps.
- Multi-perspective synthesis including non-Western or contrarian viewpoints through MiniMax integration.
- Dynamic Framework Lookup Table that automatically maps question types to industry-standard methodologies.
- Confidence level scoring (High/Medium/Low) based on model consensus and source strength.
- Automated PDF generation and delivery workflows for professional-grade output.
SKILL.md
Loading
Related Openclaw Skills
Initial Traefik
A professional-grade skill to initialize and configure Traefik v3 as a high-performance reverse proxy and load balancer within a Docker environment.
Vietnam Traffic Fine Lookup
An automated utility for checking Vietnamese traffic violations using vehicle license plates and official government cross-referencing.
VBrokers Trader
A professional-grade automation skill for executing trades, fetching real-time market data, and managing portfolios on VBrokers via the OpenAPI Gateway.
OpenRouter Usage Tracker
A utility for tracking OpenRouter API spending, credit balances, and per-model cost breakdowns directly from session logs.
Spoticlaw
A lightweight Python-based Spotify Web API client designed for AI agents to manage music playback, search, and playlists.
add-wish
A universal product-saving skill that converts any retailer URL into a one-click wishlist link for AI agents.
METADATA
Featured*
