self-improvement

What This Skill Does

The self-improvement skill for OpenClaw is a systematic framework designed to bridge the gap between ephemeral AI interactions and long-term intelligence. It enables the agent to act as a reflective practitioner by capturing, documenting, and eventually promoting critical data points—such as failed command executions, user-led corrections, and discovered best practices—into persistent workspace files. By maintaining a structured directory in .learnings/, the agent creates a recursive feedback loop. This skill transforms the agent from a stateless assistant into an evolving entity that learns from its environment, reducing repetitive errors and streamlining workflows over time. It effectively forces the agent to acknowledge its own limitations and adapt its behavioral patterns based on actual performance metrics.

Installation

Installation can be handled automatically or manually. For automated setups, use the command clawdhub install self-improving-agent to pull the skill into your OpenClaw environment. For manual installation, clone the repository directly into your skills directory: git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent. Ensure that the .learnings/ directory exists in your workspace root, containing the mandatory LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md files to begin logging immediately.

Use Cases

This skill is essential for complex development environments where recurring tasks are common. Use it when an API endpoint returns unexpected results, allowing the agent to record the specific failure for future diagnostics. It is also critical when you correct the agent's logic; documenting these corrections prevents the same hallucination or reasoning error from recurring in subsequent sessions. Furthermore, use this when you identify a more efficient way to execute a task, allowing the agent to codify this new best practice into its operational knowledge base.

Example Prompts

1. "The last git push command failed because of an SSH configuration issue. Please log this error and the resolution in ERRORS.md so we remember the fix."
2. "Actually, you should use npx instead of npm for this package. Please document this as a best_practice in the learnings file."
3. "I realize you're having trouble with this API's rate limits. Create a feature request for a retry-logic wrapper and log the current limitation in the error file."

Tips & Limitations

To maximize the utility of this skill, perform regular reviews of your .learnings/ directory. Not every entry needs promotion; prioritize entries that represent frequent bottlenecks. A limitation of this skill is its reliance on the user to occasionally prompt the logging; while the agent is capable of autonomous logging, human verification ensures that the most relevant information is captured. Ensure that your SOUL.md and AGENTS.md are kept clean, as they serve as the ultimate 'brain' that the agent references to prevent past mistakes.