self-improvement

What This Skill Does

The self-improvement skill for OpenClaw is a systematic framework designed to bridge the gap between ephemeral AI sessions and persistent, long-term intelligence. It acts as an internal feedback loop for the OpenClaw agent, allowing it to capture learnings, errors, and user-driven corrections in a structured markdown-based knowledge base. By logging interactions into the .learnings/ directory, the agent essentially 'learns' from its failures and successes, ensuring that recurring problems are solved once and for all rather than repeated.

Installation

To begin using the self-improvement skill, you must ensure your workspace is prepared for log tracking. You can install it via ClawdHub using the command clawdhub install self-improving-agent. If you prefer a manual setup, clone the repository into your ~/.openclaw/skills/ directory. Once installed, initialize your workspace directory by running mkdir -p ~/.openclaw/workspace/.learnings and ensuring the LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md files exist. This creates the foundation for the agent's persistent memory.

Use Cases

This skill is indispensable in environments where precision and reliability are critical. Use it when:

1. A command or shell operation returns an unexpected error (log to ERRORS.md).
2. The user corrects the agent's logic or factual output, providing a 'teachable moment' (log to LEARNINGS.md with category correction).
3. You identify a 'best practice' that improves efficiency for a recurring task (log to LEARNINGS.md with category best_practice).
4. You encounter an outdated API or tool limitation (log to LEARNINGS.md with category knowledge_gap).
5. The user requests a feature that falls outside current capability (log to FEATURE_REQUESTS.md).

Example Prompts

• 'The previous command failed because the API syntax changed; please document this error in the logs so we avoid it next time.'
• 'Actually, that is not the preferred way to format the JSON response. Please log this as a best practice update to our standards.'
• 'I realize we've hit this roadblock three times this week. Can you consolidate our findings from the error logs into a strategy for the next attempt?'

Tips & Limitations

To get the most out of this skill, prioritize 'promoting' entries. If a specific learning in LEARNINGS.md proves to be a recurring pattern or a fundamental shift in behavior, move that insight into SOUL.md (for behavior), TOOLS.md (for integration tips), or AGENTS.md (for workflow logic). Do not let the logs become stale; review them before starting major tasks to ensure the agent is operating on the most current 'knowledge' available in its local environment. Note that while this builds a powerful personal memory, it remains local to your workspace; ensure you back up your .learnings folder frequently.