A systematic framework for AI coding agents to capture errors, user corrections, and best practices for continuous iterative improvement.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install jose-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 jose-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 provides a structured methodology for AI coding agents to learn from their experiences. By capturing failures, corrections, and new discoveries in real-time, this skill ensures that an agent's intelligence grows with every task. It is designed to work seamlessly with Openclaw Skills to provide a persistent memory layer that spans across different sessions and projects.
This skill transforms passive task execution into an active learning loop. When an agent encounters an error or receives a correction, it doesn't just fix the immediate issue; it documents the context and the resolution. This documentation is then processed and promoted into project-level guidance, ensuring that the same mistake is never repeated and that best practices are standardized across the codebase.
Self-Improvement Skill Use Cases
- When a shell command or build operation fails unexpectedly during a development task.
- Capturing specific user corrections like "Actually, we use pnpm here, not npm."
- Documenting knowledge gaps discovered when an AI realizes its internal training data for a library is outdated.
- Tracking recurring patterns that should be automated or simplified through new agent capabilities.
- Identifying missing features or tools that a user requests during a natural conversation.
How Self-Improvement Skill Works
- The agent identifies a learning trigger, such as a command error, a manual user correction, or the discovery of a non-obvious project convention.
- Using the Openclaw Skills framework, the agent appends a structured entry to the relevant markdown file in the .learnings/ directory.
- The entry is assigned a unique ID and categorized by priority and area (e.g., frontend, infra, or docs).
- During periodic reviews or session starts, the agent analyzes these logs to identify recurring patterns.
- High-value learnings are promoted to permanent project memory files like CLAUDE.md or workspace-level files like SOUL.md and TOOLS.md.
Self-Improvement Skill Setup
To get started with this skill on the OpenClaw platform, follow these installation steps:
Via ClawdHub (Recommended):
clawdhub install self-improving-agent
Manual Installation:
git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent
mkdir -p ~/.openclaw/workspace/.learnings
Enable Automatic Reminders:
openclaw hooks enable self-improvement
Self-Improvement Skill Data Schema & Taxonomy
The skill maintains a structured taxonomy within the .learnings/ directory to ensure data is easily searchable and actionable:
| File | Purpose | Key Metadata |
|---|---|---|
LEARNINGS.md |
Corrections and best practices | Priority, Area, Pattern-Key |
ERRORS.md |
Failures and stack traces | Exit Code, Context, Suggested Fix |
FEATURE_REQUESTS.md |
Requested capabilities | Complexity, Frequency, Status |
All entries follow a standardized ID format: [TYPE-YYYYMMDD-XXX] to maintain a clear chronological history of improvements.
Self-Improvement Skill Advanced Features
- Inter-session communication allows agents to send and receive learnings across different active sessions.
- Automatic skill extraction scripts turn frequently resolved errors into standalone Openclaw Skills.
- Simplify and Harden feed integration automatically ingests recurring patterns from specialized refactoring tasks.
- Multi-agent hooks provide native support for Claude Code and Codex CLI via specialized activator scripts.
- Promotion logic gates ensure only verified and recurring patterns reach the high-level project memory.
SKILL.md
Loading
Related Openclaw Skills
Agent Radio
An advanced internet radio controller for Openclaw Skills that enables seamless playback, station discovery, and volume management via AI agent commands.
Agent Profile Images
A comprehensive visual identity management system for AI agents, featuring custom uploads and AI-powered themed image generation.
CatLab Intelligent Data Assistant
A professional data analysis expert providing secure, accurate, and high-efficiency MongoDB querying and statistical reporting within the Openclaw Skills ecosystem.
acpx
A headless, scriptable CLI client for the Agent Client Protocol (ACP) designed for seamless agent-to-agent communication.
Jira API
A specialized tool for advanced Jira Cloud automation, enabling programmatic worklog management and direct REST API interactions.
Agent3 Hub
A universal registry and gateway that allows AI agents to search and invoke other agents, MCP servers, and APIs through a single endpoint.
Featured*
