A systematic decision-making framework that applies Bayes' theorem to prevent cognitive biases, base-rate neglect, and prosecutor's fallacies during complex evidence interpretation.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install deciqai-bayesian-reasoning
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 deciqai-bayesian-reasoning 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 Bayesian Reasoning?
The Bayesian Reasoning skill is a powerful framework for AI agents to apply rigorous probability discipline to complex decision-making. In high-stakes environments, reasoning often falls prey to systemic errors—such as ignoring base rates, over-updating on vivid confirming evidence, or conflating conditional probabilities. This skill equips your agent with a mathematical engine to systematically update its strength of belief as new evidence arrives, adhering strictly to Bayes' theorem: Posterior odds = Prior odds × Likelihood ratio.
Integrating this skill within Openclaw Skills ensures your autonomous systems evaluate evidence systematically, prevent common logical fallacies, and provide transparent, mathematically sound justifications for high-stakes decisions like threat detection, medical triage, or product optimization.
Bayesian Reasoning Use Cases
- High-stakes clinical decision systems or medical screening pipelines interpreting test results with low base-rate prevalence.
- Security alert triage, log monitoring, and fraud-flagging where false positive rates are high.
- Strategic hiring pipelines needing to balance candidate interview impressions against historical success rates.
- AI-powered A/B testing and performance optimization requiring rigorous updates of experimental priors.
- Spotting logical fallacies in text processing where "evidence consistent with X" is treated as absolute proof of X.
How Bayesian Reasoning Works
- Define Exhaustive Hypotheses: Formulate the primary hypothesis (H) and its alternative (not-H) to establish a clear, exhaustive decision boundary.
- Anchor the Prior: Determine the historical base rate P(H) before any new evidence is evaluated, grounding the starting probability in empirical data.
- Estimate Likelihood Ratio: Compute P(E|H) and P(E|not-H). Analyze the diagnostics of the evidence to establish the Likelihood Ratio (LR).
- Compute the Posterior: Multiply prior odds by the Likelihood Ratio and convert the result back to posterior probability P(H|E).
- Assess Evidence Independence: Validate that multiple pieces of evidence are mathematically independent rather than being correlated updates from a single source.
- Formulate and Execute Action: State the final posterior probability, cross-reference against decision thresholds, and define the next critical evidence needed.
Bayesian Reasoning Setup
# Clone your agent workbench or Openclaw directory
git clone https://github.com/example/openclaw-skills.git
cd openclaw-skills
# Install dependency thinking packages if applicable
npm install @openclaw/skills-core
# Copy or register the bayesian-reasoning skill into your active agent profiles
cp skills/bayesian-reasoning/SKILL.md ~/.config/openclaw/skills/
Bayesian Reasoning Data Schema & Taxonomy
The skill organizes data and outputs structured Bayesian updates to ensure absolute transparency. The primary model produces a standard markdown output as detailed below:
Bayesian Update Format
| Field Name | Type | Description |
|---|---|---|
| H / not-H | String | The exhaustive hypotheses being compared. |
| Prior P(H) | Float/Percentage | The starting probability based on historical data. |
| Source | String | Reference or evidence source verifying the baseline prior. |
| Evidence E | String | The new data, signal, or test result observed. |
| P(E|H) | Float | Probability of the evidence if the hypothesis is true. |
| P(E|not-H) | Float | Probability of the evidence if the hypothesis is false (False positive rate). |
| LR | Float | Likelihood Ratio (P(E|H) / P(E|not-H)). |
| Posterior P(H|E) | Float/Percentage | The calculated posterior belief after accounting for evidence. |
| Interpretation | String | Narrative analysis of the mathematical shift. |
| Independence check | String/Boolean | Assessment of whether multiple evidence sources are correlated. |
| Decision threshold | Float/Percentage | Required probability limit to trigger an operation. |
| Action | String | Recommended next course of action based on the threshold. |
| Next evidence | String | The next critical data point that would shift posterior odds. |
Bayesian Reasoning Advanced Features
- Adaptive Front Door Coach Mode: Restricts steps to a strict one-question-at-a-time loop with hard stops to prevent agent hallucinations and guide novice users systematically.
- Domain Pack Bundles: Out-of-the-box configuration settings containing domain-specific priors and common likelihood ratio patterns (e.g., Medical Screening, Security Triage, Hiring, A/B testing).
- Rigorous Anti-Rationalization Guardrails: Built-in verification lists to catch logical traps like hindsight contamination, intuitive non-numerical updates, or treating correlated evidence as independent.
- Multi-Skill Composability: Integrates seamlessly with other Openclaw Skills such as critical-thinking, logical-fallacies, and probabilistic-thinking to formulate high-performance reasoning chains.
SKILL.md
Loading
Related Openclaw Skills
OPC Zhinao
OPC Zhinao is a full-stack startup diagnosis agent designed for solo entrepreneurs, guiding them through a quantified five-stage validation and growth framework.
Lamb (The Gospel According to Biff)
An AI coding agent skill that maps Christopher Moore's Lamb to user queries about deep friendship, coping with humor, accepting tragedy, and finding one's purpose.
Lamb: The Gospel According to Biff - AI Companion Skill
An AI agent skill based on Christopher Moore's Lamb that helps users navigate complex themes of friendship, faith, humor, and purpose through the lens of Joshua and Biff.
The Agony and the Ecstasy - Creative Perseverance & Discipline AI Skill
A specialized AI agent skill designed to mentor creators, developers, and artists through the psychological and professional struggles of building great work.
First Principles
An advanced analytical skill that strips claims down to irreducible truths and rebuilds them without unverified assumptions or flawed analogies.
Wintergirls - Eating Disorder Support & Recovery Guide
An AI companion skill that leverages the deep narrative of Laurie Halse Anderson's Wintergirls to guide users through understanding eating disorders, managing grief, and navigating the difficult path to mental health recovery.
Featured*
