self-improvement

What This Skill Does

The self-improvement skill for OpenClaw is a systematic framework designed to enhance an AI agent's performance over time. By centralizing logs for errors, corrections, and best practices, it transforms incidental interactions into structured long-term knowledge. It serves as an active feedback loop, ensuring that the agent does not repeat historical mistakes and adapts its behavior to meet user expectations more accurately. Through the maintenance of specific markdown logs within the .learnings/ directory, the agent builds a comprehensive record that informs future sessions.

Installation

To install this skill, you can use the ClawdHub manager by running clawdhub install self-improving-agent in your terminal. Alternatively, perform a manual installation by cloning the repository into your OpenClaw skills directory: git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent. Ensure that you have manually created the necessary storage structure by running mkdir -p ~/.openclaw/workspace/.learnings to allow the agent to read and write to the required log files.

Use Cases

This skill is essential for maintaining high-quality output in complex coding environments. Primary use cases include: 1) Capturing the specific details of a runtime error to prevent recurring debugging cycles. 2) Recording user corrections, such as when a user specifies a preferred coding style or terminology, ensuring consistency in future interactions. 3) Logging missing features when a user requests a tool capability that does not yet exist, creating a backlog for future development. 4) Updating the agent's knowledge base when internal information or external documentation becomes outdated, and 5) Codifying "best practices" discovered during active development to standardize successful workflows.

Example Prompts

1. "I corrected you earlier on the API authentication flow; please ensure that change is logged to the learning files so you remember it next time."
2. "We just resolved that persistent database connection error; please document the solution in the error logs so we have a reference for future troubleshooting."
3. "That was an excellent way to structure the microservices refactor. Please record this as a best practice in the learnings log."

Tips & Limitations

To maximize the utility of this skill, prioritize the promotion of frequently cited learnings to your project's root AGENTS.md or SOUL.md files. This "promotion" prevents the logs from becoming bloated and ensures core instructions are prioritized by the agent's context window. Limitations include a dependency on the user and agent to proactively trigger the log writes; if you do not explicitly ask the agent to log a correction, the improvement will not be captured. Additionally, ensure your logs remain well-structured to avoid token waste during the reading phase.