self-improvement

What This Skill Does

The self-improvement skill is a structured logging framework designed to transform the OpenClaw AI agent into a learning engine. Rather than relying on ephemeral memory, this skill systematically captures operational failures, user-provided corrections, and knowledge gaps into dedicated markdown files within a .learnings/ directory. By categorizing entries into ERRORS.md, LEARNINGS.md, and FEATURE_REQUESTS.md, the agent creates a persistent knowledge base that can be indexed, searched, or processed by downstream automated scripts. This ensures that the agent becomes increasingly intelligent over the lifetime of a project, preventing the repetition of past mistakes and aligning its behavior with specific user preferences.

Installation

To integrate this skill, ensure you are in the root directory of your project. Run the following command in your terminal:

clawhub install openclaw/skills/skills/dc-acronym/self-improving-agent-1-0-0

Once installed, initialize the workspace directory by running mkdir -p .learnings. The agent will now automatically append entries to these files based on the standardized schemas provided in the skill documentation.

Use Cases

This skill is ideal for complex development environments or long-running research tasks. Use it when:

• An external API returns an unexpected error (log to ERRORS.md to track reliability).
• You find yourself correcting the agent's code style or logic (log to LEARNINGS.md to hardcode your preferences).
• You identify a missing tool or CLI feature you wish the agent had (log to FEATURE_REQUESTS.md to track backlog).
• You discover a superior approach to a recurring task (log as 'best_practice' to refine your workflow documentation).

Example Prompts

1. "The last command failed with a 403 error. Please document this in the error logs and suggest a way to fix the permissions for future runs."
2. "Actually, that isn't the correct way to handle the database connection. Log this as a correction in the learnings file so you remember the correct pattern next time."
3. "I wish I could generate test cases directly from this module. Please add a feature request for that capability to the logs."

Tips & Limitations

• Review Periodically: The logs are only useful if reviewed. Before starting a major new task, ask the agent: 'Review the .learnings/ directory and summarize any relevant constraints or best practices.'
• Promote Content: If a specific learning is frequently accessed, consider moving that documentation to a project-level CLAUDE.md or AGENTS.md file to ensure broad visibility.
• Keep It Concise: While detailed logs are great, focus on actionable data so that your future self—or a future autonomous agent—can easily parse the suggested fixes.