A recursive learning framework for AI agents to log errors, user corrections, and best practices into a structured memory system.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install frank-self-improving-agent
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 frank-self-improving-agent 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-cognition layer designed for AI coding agents. It enables agents to move beyond transient chat sessions by capturing persistent learnings, technical errors, and user-requested features into a structured .learnings directory. This data serves as a long-term memory bank that allows agents to avoid repeating past mistakes and refine their workflows over time.
By integrating this skill into Openclaw Skills, developers create a feedback loop where an agent's performance actually improves with every interaction. It transforms one-off troubleshooting into reusable project intelligence, ensuring that non-obvious solutions and project-specific conventions are never lost. Whether you are using OpenClaw, Claude Code, or Copilot, this skill provides the necessary scaffolding for building truly autonomous and evolving AI assistants.
Self-Improvement Skill Use Cases
- Automatically logging non-zero exit codes and command failures for later analysis.
- Capturing specific user corrections when an agent's initial logic is flawed.
- Tracking recurring knowledge gaps where documentation is outdated or missing.
- Identifying high-value coding patterns to be promoted into permanent project instructions.
- Generating feature request backlogs based on conversational triggers.
How Self-Improvement Skill Works
- Monitoring: The agent monitors session activity for triggers like command errors, manual user corrections, or discovery of new best practices.
- Logging: Relevant data is captured into specialized markdown files (.learnings/LEARNINGS.md, ERRORS.md, or FEATURE_REQUESTS.md) using standardized ID formats like LRN-YYYYMMDD-XXX.
- Analysis: The agent uses cross-referencing tools (like grep or session history) to identify recurring patterns and link related entries.
- Promotion: High-value, recurring learnings are distilled and 'promoted' to core project files like CLAUDE.md or AGENTS.md.
- Skill Extraction: Validated, non-obvious solutions are extracted into standalone, reusable skills for the broader ecosystem.
Self-Improvement Skill Setup
To get started with Openclaw Skills, install the self-improving agent via ClawdHub:
# Recommended installation via ClawdHub
clawdhub install self-improving-agent
# Manual installation for OpenClaw
git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent
# Initialize the directory structure
mkdir -p ~/.openclaw/workspace/.learnings
To enable automatic session reminders, activate the hook:
openclaw hooks enable self-improvement
Self-Improvement Skill Data Schema & Taxonomy
The skill organizes data within a .learnings/ directory using a standardized taxonomy for high scannability:
| File | Purpose | Key Metadata Fields |
|---|---|---|
| LEARNINGS.md | Corrections & Best Practices | Priority, Area, Pattern-Key, Status |
| ERRORS.md | Tool & Command Failures | Exit Code, Context, Suggested Fix |
| FEATURE_REQUESTS.md | Missing Capabilities | Complexity, Frequency, Implementation |
Each entry uses a unique ID (e.g., LRN-20260509-001) to facilitate inter-session communication and pattern tracking.
Self-Improvement Skill Advanced Features
- Pattern-Key Deduplication: Automatically tracks the frequency of specific issues using stable keys to trigger priority bumps.
- Automated Skill Extraction: Provides helper scripts to transform a resolved learning into a standalone Openclaw Skills package.
- Inter-Session Messaging: Uses session_send and sessions_history to share critical learnings between active sub-agents.
- Hook-Based Reminders: Optional activator scripts that prompt the agent to evaluate its performance after every task execution.
SKILL.md
Loading
Related Openclaw Skills
Todo Skill
A robust, SQLite-driven task management system supporting nested sub-tasks, multi-project workflows, and intelligent priority sorting.
Productlane Integration
A powerful integration skill that enables AI agents to manage Productlane feedback boards, cards, and roadmaps through the Membrane ecosystem.
Maliang Doubao Image Gen
A powerful image generation skill that brings the Doubao Seedream model to your workspace via Volcano Engine ARK.
OSTree Atomic System Updates
A professional toolset for managing atomic system updates, immutable infrastructure, and version rollbacks in modern Linux environments.
Markdown to PDF Converter
An automated workflow skill that transforms Markdown reports into professional PDF documents for seamless sharing on platforms like WeChat and Feishu.
Engage
A comprehensive integration for managing customer engagement, automated messaging, and user lifecycle data through the Membrane platform.
Featured*
