A three-tier evolution system for AI agents that automatically captures errors, distills experience, and promotes recurring knowledge into permanent skills.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install lorin-self-evolving-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 lorin-self-evolving-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-Improving Agent?
The Self-Improving Agent is a comprehensive framework designed to transform every mistake and successful workflow into a reusable asset. By combining automated hooks, active distillation, and a structured promotion system, this skill ensures that an AI agent never repeats the same error twice and continuously refines its operational efficiency.
Inspired by advanced agentic management protocols, it implements a three-layer architecture: automatic error detection (Hooks), proactive experience extraction (Skills), and systematic knowledge promotion (Soul/Tools). This allows the agent to build a persistent memory that scales across different tasks and environments, making it a cornerstone of the Openclaw Skills ecosystem.
Self-Improving Agent Use Cases
- Automating the recording of CLI command failures and debugging steps for future reference.
- Distilling complex multi-step workflows into standardized procedural skills.
- Capturing specific environment configurations and project-specific quirks to prevent regression.
- Managing a lifecycle of continuous learning through recurring error analysis and knowledge promotion.
How Self-Improving Agent Works
- Automatic Capture: Shell hooks detect non-zero exit codes or error patterns in tool outputs and log them to specialized markdown files.
- Active Distillation: The agent evaluates successful complex tasks or user corrections to determine if a new skill should be created.
- Skill Creation: If the criteria are met, the agent generates a structured SKILL.md file containing procedures, triggers, and pitfalls.
- Systematic Promotion: Frequently recurring learnings (3+ times) are promoted from local skill files to system-level documents like SOUL.md or TOOLS.md.
- Periodic Review: The system triggers regular reviews of pending learnings to clean up old data or upgrade validated solutions.
Self-Improving Agent Setup
To enable the automatic error detection and self-improvement hooks, run the following commands in your terminal:
# Copy hooks to the Openclaw directory
cp -r hooks/openclaw ~/.openclaw/hooks/self-improvement
# Enable the self-improvement hooks
openclaw hooks enable self-improvement
# (Optional) Run the extraction script to test skill distillation
./scripts/extract-skill.sh <skill-name> --dry-run
Self-Improving Agent Data Schema & Taxonomy
The skill manages information through a structured directory in .learnings/ and the Openclaw workspace:
| File Type | Purpose | Taxonomy/ID |
|---|---|---|
ERRORS.md |
Tracks command and tool failures | ERR-YYYYMMDD-XXX |
LEARNINGS.md |
Stores general distilled experiences | LRN-YYYYMMDD-XXX |
FEATURE_REQUESTS.md |
Records desired agent capabilities | FEAT-YYYYMMDD-XXX |
SKILL.md |
Standardized reusable skill definition | Semantic Naming (e.g., mysql_connect_prod) |
Data is organized by area (frontend, backend, infra) and priority (low to critical).
Self-Improving Agent Advanced Features
- Automated Error Hooking: Real-time monitoring of bash outputs via error-detector.sh to catch failures as they happen.
- Decision Tree Logic: Intelligent filtering to ensure only non-obvious, recurring, and verified workflows are turned into skills.
- Three-Tier Promotion System: Moves knowledge from temporary logs to project-level memory and finally to global agent personality files.
- Skill Extraction Scripts: CLI tools to automate the migration of verified learnings into the Openclaw Skills directory structure.
- Forgetfulness Prevention: Implements a review cycle to ensure high-priority knowledge is promoted before it is buried by newer logs.
SKILL.md
Loading
Related Openclaw Skills
fb-expense-skill
An AI-driven expense assistant that automates invoice recognition, travel reimbursement, and policy compliance workflows.
Douyin Video Downloader
A powerful utility for downloading Douyin videos without watermarks by bypassing anti-hotlinking protections using custom curl commands.
Nginx Reverse Proxy
An automated utility for configuring Nginx reverse proxies, binding domains to specific ports, and securing traffic with SSL.
GitHub Bounty Hunter Skill
An automated system for discovering GitHub issue bounties and solving them using AI-driven coding agents.
K-Pop Idol Generator
A specialized AI image generation skill for creating polished K-pop debut-style portraits and photocard aesthetics via the Neta AI API.
Shi Kongwu Home
A structured multi-agent collaboration framework that organizes specialized AI entities into a cohesive team hierarchy for complex project management.
Featured*
