self-improvement

What This Skill Does

The self-improvement skill is a robust feedback-loop mechanism designed for OpenClaw agents to capture, track, and operationalize learnings. It functions by logging errors, user corrections, and discovered best practices into a structured .learnings/ directory. By creating an audit trail of mistakes and successful patterns, the agent transitions from a stateless responder to a system that evolves with the codebase and user preferences. It effectively bridges the gap between ad-hoc debugging and long-term behavioral refinement by promoting transient observations into permanent documentation files like SOUL.md, AGENTS.md, and TOOLS.md.

Installation

You can install this skill either via the package manager or manually. For most users, using ClawdHub is the preferred method: clawdhub install self-improving-agent. If you prefer a manual setup, clone the repository from the source at https://github.com/peterskoett/self-improving-agent.git into your ~/.openclaw/skills/ directory. After installation, ensure you initialize the logging structure by running mkdir -p ~/.openclaw/workspace/.learnings to allow the agent to start writing log files immediately.

Use Cases

This skill is ideal for complex, multi-day development tasks. Use it when an external API returns an unexpected error code, as it helps you debug recurring integration failures. It is also essential when a user identifies a recurring pattern in the agent's behavior they dislike (e.g., "Stop providing verbose disclaimers"), or when a better workflow is discovered during experimentation. Finally, use it to track missing features or gaps in knowledge that trigger frequent hallucinations, ensuring the agent adapts to your specific project requirements over time.

Example Prompts

1. "The last shell command failed with a connection timeout; please log this in the error tracking file and analyze the cause."
2. "Actually, when you refactor React components, please always use functional components instead of class-based ones. Log this as a best practice in our learnings folder."
3. "Review the current state of LEARNINGS.md and suggest which entries should be promoted to our SOUL.md behavioral guidelines."

Tips & Limitations

Always prioritize reviewing your .learnings/ files during the onboarding phase of a new session. While the agent handles the writing, you (the human) should act as the editor to ensure only high-quality, relevant data is promoted to core system files. Be careful not to clutter log files with noise; focus on actionable items that help the agent improve future performance. Note that this skill relies heavily on file-write permissions within the workspace; ensure your local environment configuration supports these operations without external blockages.