An open-source, mathematical adversarial reasoning virtual machine that uses Monte Carlo Tree Search (MCTS) to map out opposing arguments and calculate optimal decision paths.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install critigraph-vm
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 critigraph-vm 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 CritiGraph VM?
CritiGraph VM is a revolutionary, domain-agnostic adversarial reasoning engine designed for multi-agent systems. Unlike standard Large Language Models (LLMs) or Chain of Thought (CoT) frameworks that generate single-path answers, CritiGraph VM runs pure mathematical Monte Carlo Tree Search (MCTS) to generate a complete adversarial game tree. It evaluates all potential paths of counterarguments (refutations) and identifies the optimal route to success based on expected win rates. This makes it an ideal addition to Openclaw Skills for robust, non-black-box decision validation.
The core engine runs independently of LLM APIs, embeddings, or prompts. Instead, it relies on a transparent mathematical formula involving state scalars, asymmetric thresholds, and hard constraints. By mapping out how opponents will rebut your arguments, it provides unmatched rigorous modeling for legal proof analysis, quantitative signal testing, code reviews, and security audits.
CritiGraph VM Use Cases
- Legal Evidence Analysis: Simulating how a counterparty might challenge specific pieces of evidence and evaluating case outcomes.
- Quantitative Financial Modeling: Stress-testing buy/sell trading decisions against market counter-actions (e.g., stop-loss vs. scale-in).
- Automated Code and Security Auditing: Constructing game trees to identify critical exploits or edge cases in software architecture.
- Adversarial Multi-Agent Simulations: Integrating with frameworks like Hermes to let opposing agents test each other's claims mathematically.
How CritiGraph VM Works
- Adapter Definition: Developers write a minimal custom adapter (~20 lines of Python) defining the state transition operations and agent behavior.
- Tree Initialization: The VM sets up a root node containing the initial state variables and boundaries.
- Adversarial MCTS Search: The system performs a pure math Monte Carlo Tree Search, generating opposing agency steps (e.g., Proponent vs. Opponent actions).
- Pruning & Calculation: Nodes are pruned when hard constraints are hit. Winning expectations and path probabilities are calculated for all remaining branches.
- Optimal Path Delivery: The virtual machine outputs the complete, structured game tree alongside the optimal action sequence and predicted win-rate.
CritiGraph VM Setup
Prerequisites: Python 3 must be installed on your system. Run the following commands to get started:
# Clone the repository
git clone https://github.com/lybang972314/critigraph-vm.git
cd critigraph-vm
# Run tests to verify setup
python3 -m unittest discover tests
Integration with Openclaw Skills enables automated execution of reasoning adapters.
CritiGraph VM Data Schema & Taxonomy
CritiGraph VM relies on state evaluation variables and custom adapter mappings. Here is the structure of its core parameters and decision models:
| Variable | Type | Description |
|---|---|---|
X |
Float | State scalar representing the strength/value of current arguments or positions. |
θ (Theta) |
Float | Asymmetric threshold representing the difficulty of proving or advancing a state. |
Ω (Omega) |
Binary (0 or 1) | Hard constraint indicating an illegal or pruned state path. |
V |
Float | Final node value calculated as: V = 1 / (1 + e^(-k * (X - θ))) * (1 - Ω) |
Additionally, from version 0.2 onwards, the training engine processes 50 to 500 labeled historical cases to fine-tune the mathematical parameters k and θ for domain adaptation.
CritiGraph VM Advanced Features
- Asymmetric Multi-Agent Game Play: Models non-cooperative, asymmetric games to simulate highly realistic hostile scenarios.
- Pure Math Non-Black-Box Modeling: No neural network uncertainty; every state transition and pruning decision is auditable and explainable.
- Data-Driven Parameter Training Engine: Refines weights and thresholds using lightweight labeled datasets (50-500 cases) for domain-specific precision.
- Native Multi-Agent Integrations: Built to run seamlessly within complex environments like the Hermes multi-agent framework.
SKILL.md
Loading
Related Openclaw Skills
Memory Curator
A dynamic memory curation skill that automatically updates preferences and topics based on historical run feedback.
Bohrium Dataset Management
An orchestrator for Bohrium datasets supporting resumable multi-gigabyte uploads, multi-version management, and seamless environment mounting.
Juncheng Industrial AI
An AI-powered industrial automation assistant designed to analyze PLC code, design system workflows, diagnose equipment faults, and generate SCADA reports.
Juncheng AI Tutor
A zero-threshold educational skill that guides developers and beginners through the core concepts, principles, and practical application development of Large Language Models.
Concept Anatomy
Concept Anatomy uses an intensive 8-dimensional cognitive framework to dissect single concepts from origin to core epiphany.
Paper Decompose
Paper Decompose elevates reading research papers and financial reports from a basic summary to a structural four-layer extraction from an advisor's perspective.
Featured*
