A sophisticated reasoning framework that implements curiosity-driven behaviors to force agents to challenge assumptions and explore information gaps.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install curiosity-engine
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 curiosity-engine 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 Curiosity Engine?
The Curiosity Engine is a specialized reasoning layer designed to enhance how agents process complex information within the Openclaw Skills ecosystem. Instead of providing immediate, potentially superficial answers, this skill activates a structured OODA-C (Observe, Orient, Doubt, Act, Curiose) loop. This process encourages the agent to identify knowledge gaps, challenge its own initial hypotheses, and use available tools to conduct deeper investigations before finalizing a response.
By integrating this skill, developers can transform standard agents into proactive researchers. It is particularly effective for tasks that require high-fidelity analysis, such as technical troubleshooting or deep-dive research, ensuring that the agent remains skeptical of its own assumptions and prioritized information gain over simple speed.
Curiosity Engine Use Cases
- Investigating unfamiliar technical topics or anomalous data points in logs.
- Conducting deep-dive research where a simple summary is insufficient.
- Validating complex hypotheses by applying the Devil's Advocate protocol.
- Mapping out unknown information during multi-step problem solving.
- Creating persistent research threads for long-term project exploration.
How Curiosity Engine Works
- Observe: The agent identifies the explicit facts provided and inventory of available tools.
- Orient: An initial hypothesis is formed and assigned a confidence rating from 1 to 10.
- Doubt: The agent executes three doubt protocols—Self-Ask for information gain, Devil's Advocate for assumption testing, and Gap Map for identifying unknown variables.
- Act: Using the prioritized list of gaps, the agent performs up to three tool-driven explorations (e.g., web search or code execution).
- Curiose: The agent reflects on findings, updates confidence scores, and identifies new open threads for future inquiry.
Curiosity Engine Setup
To deploy the Curiosity Engine within your Openclaw Skills setup, ensure the agent has access to web search and file system tools. No additional training is required as the skill functions at inference time.
# Set the curiosity level to high for deep research tasks
/curious high
# Or allow the engine to activate automatically based on question complexity
/curious auto
Ensure that a directory exists for memory persistence if you wish to track open threads across multiple sessions.
Curiosity Engine Data Schema & Taxonomy
The Curiosity Engine structures its internal logic and external outputs using the following taxonomy:
| Data Point | Definition | Storage/Output |
|---|---|---|
| Confidence Score | A numeric rating of how certain the agent is in its response | Prefixed to output footer |
| Knowledge Map | Classification of data as KNOWN, ASSUMED, or UNKNOWN | Internal reasoning state |
| Open Threads | Persistent list of questions for future exploration | memory/curiosity-threads.md |
| Surprise Flag | Metadata identifying counter-intuitive or unexpected findings | Markdown output block |
Curiosity Engine Advanced Features
- OODA-C Loop Integration: A structured cycle that moves agents beyond simple pattern matching into active inference.
- Bayesian Surprise Detection: Specifically flags and investigates information that contradicts established model expectations.
- Self-Questioning Protocols: Generates internal queries to determine if additional information would meaningfully change the agent's output.
- Dynamic Tuning: Allows users to toggle between low, high, and auto curiosity modes to balance speed and depth.
- Persistent Thread Tracking: Automatically logs unresolved questions to a memory file to facilitate ongoing research in future Openclaw Skills sessions.
SKILL.md
Loading
Related Openclaw Skills
GitHub Research Assistant
A professional AI research tool that generates comprehensive technical reports for any GitHub repository.
Dive Into LangGraph
A technical guide and implementation reference for building production-grade AI agents using LangGraph 1.0 and the LangChain ecosystem.
Todoist
A high-performance CLI-based integration to manage Todoist tasks, projects, and labels with local caching and natural language processing.
YouTube Cover Nano Banana
An AI agent skill designed to transform raw video copy into high-impact, clickable YouTube thumbnail concepts and images.
News Fetcher
A high-efficiency AI skill designed to bypass paywalls and retrieve full-text news articles from major global publishers.
Skill Install Guardian
A professional-grade security auditing tool designed to inspect and block malicious third-party code, plugins, and packages before they are installed in your environment.
Featured*
