self-improvement

What This Skill Does

The self-improvement skill is the core feedback loop for OpenClaw agents. It enables the AI to learn from its operational history by logging mistakes, user corrections, and successful procedural optimizations. By utilizing a structured directory format (.learnings/), this skill turns ephemeral conversation into persistent knowledge. When an agent experiences a failure or is corrected by a user, it leverages this skill to capture the context, which can then be synthesized into permanent improvements for future sessions. This capability shifts the agent from being stateless to becoming progressively more refined and effective over time.

Installation

Installation can be performed via the automated CLI or manually. For most users, the recommended approach is using the ClawdHub installer:

clawdhub install self-improving-agent

If performing a manual setup, ensure the directory structure exists within your OpenClaw workspace:

1. Create the hidden directory: mkdir -p ~/.openclaw/workspace/.learnings
2. Initialize the three primary log files: LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md.
3. Verify that your agent has write permissions for these files to ensure logs are captured reliably.

Use Cases

This skill is indispensable for long-running projects or complex development tasks. Key use cases include:

• Debugging: When an API integration fails or a terminal command returns a non-zero exit code, log the failure to ERRORS.md to prevent repeated diagnostic loops.
• Refinement: When a user clarifies a misunderstanding or corrects a generated output, log it as a correction in LEARNINGS.md.
• Knowledge Management: If you discover a superior library or a cleaner coding pattern, log it as a best_practice to guide the agent in future tasks.
• Capability Expansion: When a task requires a tool or feature the agent lacks, use FEATURE_REQUESTS.md to track requirements for upcoming developmental sprints.

Example Prompts

1. "The script failed with a permission error. Log this to ERRORS.md and analyze what went wrong."
2. "Actually, you should use the AWS SDK v3 instead of v2 for this task. Please update our best practices in LEARNINGS.md."
3. "Review the current state of our .learnings folder and suggest three improvements based on our recent successes."

Tips & Limitations

To maximize the utility of this skill, prioritize 'Promotion' over passive logging. Periodically review the .learnings directory and migrate frequently referenced entries to SOUL.md (for behavior) or TOOLS.md (for technical workflows). The skill is limited by the agent's ability to self-reflect; therefore, it is best practice to initiate a review of logs before beginning any high-stakes or major development work.