A Python-based epistemic reasoning engine that evaluates claims against retrieved evidence to output structured belief states and calibrated confidence metrics.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install belief-state-runtime
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 belief-state-runtime 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 belief-state-runtime?
belief-state-runtime is an advanced, LLM-driven epistemic reasoning engine designed to assess the trustworthiness and reliability of claims against verifiable evidence. Acting as a critical validation layer for AI-driven workflows, this tool combines automated Python rule layers with deep LLM evaluation to eliminate hallucinations and build rigorous evidence-backed claims.
As a versatile addition to the ecosystem of Openclaw Skills, it helps AI coding agents classify facts into distinct structured states (VERIFIED, CONTESTED, or UNCERTAIN) with high mathematical precision. By evaluating source reliability, evidence density, and temporal freshness locally, it guarantees secure, privacy-first reasoning without external network leaks.
belief-state-runtime Use Cases
- Verifying factual claims when asked "is this true?" or "can I trust this?" during coding or research workflows.
- Aggregating and cross-referencing information from multiple search sources to detect inconsistencies.
- Detecting structural contradictions or logical mismatches across separate evidence inputs.
- Quantifying the exact statistical confidence of an AI-generated conclusion to guarantee reliable outputs.
How belief-state-runtime Works
- Search for Evidence: The host agent utilizes search tools to aggregate relevant, verifiable facts.
- Initialize Rule Layer: Python computation in
assess.pyanalyzes source reliability, evidence density, and temporal freshness. - LLM Evaluation: The AI agent answers 6 distinct boolean questions mapping logical alignment, limitations, and contradictions.
- Statistical Aggregation: The engine aggregates the rule signals and LLM judgments to calculate a calibrated confidence score.
- Belief State Output: Returns a structured JSON containing the calibrated state (VERIFIED, CONTESTED, or UNCERTAIN) along with a confidence range and executive summary.
belief-state-runtime Setup
Make sure Python 3 is installed on your local macOS or Linux environment. Systems using Openclaw Skills work natively with this setup.
# Navigate to your skill directory
cd path/to/your/skills
# Clone or download belief-state-runtime
git clone https://github.com/hqzzdsda/belief-state-runtime.git
Configure your custom domain rules, thresholds, and keyword weights by adjusting config.json or by using the online belief-state-runtime configurator.
belief-state-runtime Data Schema & Taxonomy
The belief-state-runtime utilizes a compact structured workspace containing configuration and reasoning scripts. It processes inputs and generates structured evaluation objects.
File Structure
| File | Description |
|---|---|
assess.py |
Core self-contained Python validation script containing domain rules, thresholds, and weights. |
config.json |
System configurations specifying threshold limits and belief state criteria. |
Output Schema
When evaluation completes, the engine produces a structured JSON output with the following properties:
- state: The evaluated status. It can be
VERIFIED(confidence >= 0.65),CONTESTED(0.25 < confidence < 0.65), orUNCERTAIN(confidence <= 0.25). - confidence: A floating-point number from
0.00to1.00specifying computed reliability. - confidence_range: A two-element list showing lower and upper confidence bounds.
- features: An object mapping boolean evaluations (e.g.,
direct_support,new_info,logical_consistent,direct_refute,limitation,error_outdated). - summary: A human-readable synthesis statement summarizing the finding.
belief-state-runtime Advanced Features
- Dual-Layer Reasoning: Marries rigid Python computation rules (freshness, source ranking) with cognitive LLM evaluations.
- Six-Point Semantic Judgment: Evaluates logical consistency, direct refutations, hidden limitations, and outdated information to avoid blind-spot hallucinations.
- No-Key Privacy Security: All calculations, parsing, and epistemic evaluation execute purely locally, keeping search logs and local data confidential.
- Seamless Integration: Integrates flawlessly into developer environments utilizing Openclaw Skills along with Claude Code, Cursor, Codex, and GitHub Copilot without requiring external API endpoints.
SKILL.md
Loading
Related Openclaw Skills
AI Compute (ai-compute-x402)
A 27-endpoint AI compute agent providing seamless LLM inference, image/video generation, speech synthesis, on-chain intelligence, and Bittensor-powered decentralized AI powered by x402 micropayments.
Energy and Civilization Analyzer
An executable historical and thermodynamic analysis framework that models energy transitions, energy density limits, and EROEI dynamics.
Quantum Computing for Everyone
An executable AI-guided toolkit helping developers and researchers master the core principles of quantum mechanics, qubits, and advanced algorithms.
Arabic UI Localization Reviewer
An AI-powered localization reviewer to audit, refine, and optimize Arabic user interfaces, ensuring flawless RTL layouts and culturally authentic copy.
Douyin Scraper
An automated data scraper and keyword search utility for Douyin leveraging Playwright and public APIs for seamless social trend retrieval.
Born to Run Skill
An executable AI agent toolkit based on Christopher McDougall's 'Born to Run' to help runners unlock natural endurance, fix running form, and safely transition to minimalist running.
Featured*
