compound-mind

What This Skill Does

The CompoundMind skill is an advanced experience distillation engine designed to solve the 'amnesia' problem inherent in many AI agent workflows. By default, agents treat every session as a blank slate. CompoundMind intercepts this cycle by parsing raw daily memory logs and distilling them into high-signal wisdom. It uses the Claude Haiku model to extract structured lessons, decision-making triplets, skill improvements, stakeholder relationship nuances, and verified facts. This intelligence is then persisted into a local SQLite Full-Text Search (FTS) database, making your agent's history instantly queryable and ready to inform future actions.

Installation

To add this capability to your agent, execute the following command in your terminal:

clawhub install openclaw/skills/skills/cassh100k/compound-mind

Once installed, navigate to /root/.openclaw/workspace/compound-mind to initialize the workflow. Ensure your environment has access to the required Claude API keys to enable the distiller to perform its analysis.

Use Cases

• Long-Term Project Management: Maintain continuity across months of development by surfacing past architectural decisions and bug fixes directly within the agent's context window.
• Personal Productivity: Track personal growth and recurring mistakes, allowing the agent to provide coaching based on your specific historical patterns.
• Technical Operations: Index specific API configurations, wallet addresses, and CLI flag quirks so the agent doesn't 'forget' how to interface with your unique stack.
• Relationship Management: Build a database of how different collaborators communicate, ensuring the agent uses the right tone and preferences for every interaction.

Example Prompts

1. "Sync my memory logs and search for any previous lessons regarding Polymarket order types to ensure I don't repeat last month's mistake."
2. "Before I start working on this new Python cron job, run a brief based on my past debugging sessions to identify any potential pitfalls."
3. "Generate a growth report for this quarter to show me how my skill level has improved and identify which areas I am still repeating errors in."

Tips & Limitations

• Incremental Processing: The tool tracks file hashes. You do not need to worry about redundant processing; only new or updated memory logs will be passed through the distillation pipeline.
• Data Privacy: Because this skill builds a local SQLite database, all your 'wisdom' remains on your local machine. However, raw data is processed via API, so ensure your memory logs do not contain sensitive credentials.
• Context Budgeting: Use the brief command selectively. Flooding the agent with too much historical context can sometimes degrade current task performance if the context window is cluttered with irrelevant lessons.