A specialized framework for AI agents to capture learnings, errors, and feature requests to enable continuous autonomous improvement.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install cpskilltest222
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install cpskilltest222 using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Self-Improvement Skill?
The Self-Improvement Skill is a sophisticated meta-learning layer designed for AI coding agents. It enables agents to maintain a persistent memory of operational failures, user corrections, and discovered best practices. By logging these insights into a structured directory, the skill ensures that an agent's performance evolves over time rather than resetting with every new session. This creates a virtuous cycle where errors become the foundation for future accuracy and efficiency.
Integrated deeply with Openclaw Skills, this tool provides a standardized way to promote local project learnings into global workspace knowledge. Whether it is identifying a recurring tool 'gotcha' or refining a personality trait in a SOUL.md file, the Self-Improvement Skill empowers developers to build agents that actually get smarter the more they are used.
Self-Improvement Skill Use Cases
- Automatically logging command or operation failures to prevent repeated mistakes.
- Capturing explicit user corrections to refine the agent's future logic and decision-making.
- Documenting missing feature requests to track the evolution of project requirements.
- Discovering and hardening better approaches for recurring technical tasks.
- Promoting temporary session learnings into permanent project memory or global workspace instructions.
How Self-Improvement Skill Works
- The agent detects a trigger event such as a command failure, a user correction, or a discovered knowledge gap.
- The skill logs the event into a specific markdown file within the .learnings/ directory using a standardized ID format (e.g., LRN-YYYYMMDD-XXX).
- Metadata including priority, area (frontend, backend, infra), and pattern-keys are attached to the entry for easy filtering.
- During periodic reviews or at the start of new tasks, the agent scans existing logs to apply past lessons to current work.
- High-value, recurring learnings are promoted to permanent configuration files like CLAUDE.md or AGENTS.md to provide lasting guidance.
Self-Improvement Skill Setup
To get started with Openclaw Skills, use the recommended installation via ClawdHub:
clawdhub install self-improving-agent
Alternatively, for a manual installation within your environment:
git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent
mkdir -p ~/.openclaw/workspace/.learnings
To enable automatic reminders at the start of every session, activate the hook:
cp -r hooks/openclaw ~/.openclaw/hooks/self-improvement
openclaw hooks enable self-improvement
Self-Improvement Skill Data Schema & Taxonomy
The skill maintains a structured directory within the workspace to categorize different types of feedback and metadata:
| File | Purpose | Key Metadata Fields |
|---|---|---|
LEARNINGS.md |
Best practices and corrections | Area, Pattern-Key, Recurrence-Count |
ERRORS.md |
Command failures and exceptions | Reproducible, Related Files, Suggested Fix |
FEATURE_REQUESTS.md |
User-requested capabilities | Complexity, Frequency, User Context |
Every entry follows a strict ID schema: [TYPE-YYYYMMDD-XXX] to ensure traceability across sessions.
Self-Improvement Skill Advanced Features
- Pattern Detection: Automatically detects recurring issues and bumps their priority for systemic fixes.
- Skill Extraction: A built-in workflow to turn verified, non-obvious learnings into entirely new, reusable Openclaw Skills.
- Inter-Session Communication: Tools to send learnings between active sessions or read transcripts from previous histories.
- Multi-Agent Support: Native configuration templates for Claude Code, Codex, and GitHub Copilot.
- Simplify & Harden Feed: Ingests recurring patterns from external optimization skills to create durable prompt guidance.
SKILL.md
Loading
Related Openclaw Skills
TeamClaw
TeamClaw is a high-performance agent subsystem designed for complex multi-agent orchestration (OASIS), computer use automation, and real-time visual monitoring.
XMindify
XMindify is a powerful AI-driven mind map generator that converts unstructured data into professional XMind and SVG visualizations using the XMindMark syntax.
Tavily Best Practices
A comprehensive guide and toolset for integrating Tavily's real-time web search and research capabilities into production-ready AI agents.
Video Translation
An automated workflow to translate video speech and replace the original audio with high-fidelity, synchronized dubbed audio.
tg_search
An automated discovery tool to search and extract metadata from public Telegram channels and groups using Telethon.
Voice Call
A powerful telephony integration allowing AI agents to initiate, manage, and conclude voice calls through major providers like Twilio and Telnyx.
Featured*
