A recursive learning framework that enables AI agents to log errors, track user corrections, and evolve through persistent project memory.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install learning-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 learning-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 recursive feedback loop designed for AI coding agents. It provides a structured methodology for agents to acknowledge their mistakes, capture user feedback, and document new technical insights in real-time. By utilizing a dedicated directory for logs, the skill ensures that no valuable lesson is lost between sessions, transforming one-off corrections into permanent institutional knowledge.
Built to work seamlessly with Openclaw Skills, this tool allows agents to promote high-value learnings into core project files like CLAUDE.md or SOUL.md. This architecture prevents the repetition of errors and allows the agent's behavior to become increasingly aligned with developer preferences and project-specific conventions over time.
Self-Improvement Skill Use Cases
- A terminal command or build operation fails unexpectedly and requires root cause documentation.
- A developer corrects an agent's logic or code style, necessitating a persistent behavioral change.
- An agent identifies an outdated API or a knowledge gap that needs to be tracked for future updates.
- A user requests a recurring feature that currently doesn't exist in the agent's capability set.
- A complex debugging session yields a 'best practice' that should be standardized across the codebase.
How Self-Improvement Skill Works
- The agent detects a trigger such as a command failure, a user correction, or a discovered best practice.
- The agent generates a structured entry with a unique ID (e.g., LRN-YYYYMMDD-XXX) and appends it to the relevant markdown file in the .learnings folder.
- Logged entries are assigned metadata including priority, area (frontend, backend, etc.), and recurrence count.
- During periodic reviews or session starts, the agent scans existing logs to avoid repeating past mistakes.
- High-value, broadly applicable learnings are promoted to permanent workspace files to influence the agent's global behavior.
Self-Improvement Skill Setup
To get started with the recommended Openclaw Skills configuration, use the following commands:
# Install via ClawdHub
clawdhub install self-improving-agent
# Create the required directory structure
mkdir -p ~/.openclaw/workspace/.learnings
# Optional: Enable automatic hooks for reminders
openclaw hooks enable self-improvement
Self-Improvement Skill Data Schema & Taxonomy
The skill organizes data within a .learnings/ directory using three primary specialized markdown files:
| File | Purpose | Key Metadata |
|---|---|---|
LEARNINGS.md |
Corrections, knowledge gaps, and best practices. | Area, Pattern-Key, Source |
ERRORS.md |
Log of command failures and stack traces. | Reproducible, Related Files |
FEATURE_REQUESTS.md |
User-requested capabilities and suggestions. | Complexity, Frequency |
All entries follow a standardized header format including a unique ID, ISO-8601 timestamp, Priority, and Status (pending, resolved, or promoted).
Self-Improvement Skill Advanced Features
- Automatic Skill Extraction: Built-in scripts to convert recurring, resolved learnings into entirely new, reusable AI skills.
- Pattern Tracking: Uses a Pattern-Key system to detect and count recurring issues, triggering priority bumps for systemic problems.
- Inter-Session Communication: Tools to share learnings across different active agent sessions via session_send and session_history.
- Hook Integration: Supports UserPromptSubmit and PostToolUse hooks for automated error detection and learning prompts.
- Multi-Agent Compatibility: Specific configurations available for Claude Code, Codex, and GitHub Copilot integration.
SKILL.md
Loading
Related Openclaw Skills
Skill Publisher
A comprehensive management and distribution toolkit for publishing OpenClaw agent capabilities across global and regional platforms.
LLM Testing Skill
A comprehensive suite of security prompts and wordlists designed to evaluate the safety, robustness, and ethical alignment of Large Language Models.
Amazon Review Management
An intelligent AI agent dedicated to managing Amazon product reviews, professional responses, and compliant rating growth strategies.
Institutional Site Builder
An advanced AI skill that transforms brand assets, PDFs, and logos into high-performance React institutional websites using Vite and Tailwind CSS 4.
Flashloan Analysis Skill
A specialized technical skill for analyzing flashloan operations, evaluating DeFi protocol security, and managing on-chain reference data.
Session Recovery
An AI session management system that optimizes context usage and enables rapid workflow recovery through a structured 3-level documentation strategy.
Featured*
