An autonomous background reasoning agent that explores open questions and hypotheses using local models without consuming your primary token budget.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install mind-wander
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 mind-wander 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 Mind-Wander?
Mind-Wander is a sophisticated background reasoning agent designed to function as a digital hippocampus for your workspace. It autonomously explores unresolved questions and complex tangents listed in your project files without requiring manual intervention or consuming expensive API tokens from primary providers. By utilizing local LLMs and a private knowledge graph, it provides a powerful way to expand the capabilities of Openclaw Skills through deep, asynchronous research.
This skill acts as a specialized researcher that works in the background, performing web searches, running sandbox experiments, and tracking its own progress. It ensures that your primary AI context remains focused on immediate tasks while still benefiting from long-term knowledge consolidation. Only findings that pass a strict novelty gate are elevated to your main project files, preventing noise while maximizing intellectual discovery.
Mind-Wander Use Cases
- Exploring complex architectural trade-offs or research questions while you sleep.
- Tracking research dead ends to prevent the agent from repeating failed logic or searches.
- Running autonomous code experiments in a secure sandbox to verify technical hypotheses.
- Generating high-quality local reasoning data to improve future model training sessions.
- Consolidating project knowledge using Openclaw Skills without polluting the primary agent context.
How Mind-Wander Works
- The agent monitors the ON_YOUR_MIND.md file in your workspace for new questions, tangents, or hypotheses.
- On a set schedule, a local Qwen3.5-9B model is triggered to select one unresolved item for exploration.
- The agent utilizes a suite of tools, including Perplexity web search, local file reading, and sandbox execution, to investigate the topic.
- Every finding is checked against a wander graph in FalkorDB to ensure it hasn't been explored previously.
- Results pass through a novelty gate; if the finding is genuinely new and valuable, the agent calls the elevate tool.
- Elevated findings are written to MENTAL_EXPLORATION.md, where they can be detected and indexed by other Openclaw Skills for use in your primary context.
Mind-Wander Setup
To get started with this skill, ensure you have Ollama and FalkorDB installed, then follow these steps:
# Install the skill and dependencies
bash mind-wander/scripts/install.sh
# Pull the optimized local reasoning model
ollama pull qwen3.5-wander-q8
# Initialize your research anchor file
echo "## Research Question: How does X impact Y?" >> ON_YOUR_MIND.md
# Run a manual session to verify setup
python3 mind-wander/run.py --verbose
Mind-Wander Data Schema & Taxonomy
Mind-Wander organizes its research data into a clear hierarchy to maintain workspace cleanliness:
| Component | File/Storage | Purpose |
|---|---|---|
| Input Anchor | ON_YOUR_MIND.md |
User-defined questions and research triggers. |
| Elevated Findings | MENTAL_EXPLORATION.md |
High-value, novel insights that have passed the novelty gate. |
| Dead End Log | DEAD_ENDS.md |
A record of closed threads to prevent redundant research. |
| Session Logs | completions/wander/ |
Full JSON traces of every reasoning session for audit and training. |
| State Tracking | FalkorDB (wander graph) | Persistent memory of exploration history separate from primary context. |
Mind-Wander Advanced Features
- Local LLM reasoning powered by Qwen3.5-9B to minimize dependency on cloud APIs and reduce costs.
- A dedicated sandbox environment for running Python snippets to empirically test hypotheses.
- Integration with Perplexity AI for real-time web research and fact-checking.
- Configurable novelty gates that prevent the restatement of known facts from cluttering your workspace.
- Seamless compatibility with other Openclaw Skills like graph-rag-memory for unified knowledge management.
SKILL.md
Loading
Related Openclaw Skills
Hyperliquid Names (HLN) API
An AI agent skill designed to interact with the Hyperliquid Names API for domain resolution, profile lookups, and HyperEVM integration.
Awesome Closet Stylist
A high-autonomy AI agent skill that transforms your digital wardrobe into a personal stylist through natural language interaction and intelligent outfit planning.
Capability Evolver
A recursive self-evolution engine that allows AI agents to autonomously analyze their runtime history, fix bugs, and optimize their own source code.
CNOOC Brand & Industry Intelligence
A specialized knowledge skill providing structured information and professional brand intelligence on CNOOC for AI agents.
PDF Translator
A sophisticated translation engine that preserves complex document layouts by leveraging LaTeX reconstruction and direct arXiv source integration.
Graph-RAG Memory Skill
A temporal Graph-RAG system providing AI agents with persistent, queryable long-term memory through a Mixture-of-Experts knowledge graph architecture.
Featured*
