A sophisticated logging and promotion framework that enables AI coding agents to learn from errors, user corrections, and discovered best practices.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install self-improving-agent-6
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 self-improving-agent-6 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 specialized framework designed to create a feedback loop for AI coding agents. By capturing learnings, errors, and feature requests in structured markdown files, it ensures that an agent's knowledge evolves alongside the project. It moves beyond simple task execution by allowing agents to recognize their own knowledge gaps and outdated information, effectively turning every interaction into a growth opportunity.
At its core, this Openclaw Skills contribution provides a standardized way to document 'gotchas' and workflow improvements. Instead of repeating the same mistakes across different sessions, the skill allows for the promotion of high-value insights into permanent project memory files like CLAUDE.md or AGENTS.md, ensuring that both human developers and AI agents stay aligned on project-specific conventions.
Self-Improvement Skill Use Cases
- When a terminal command or operation fails unexpectedly during a coding session.
- When a user provides a direct correction to an agent's logic or knowledge.
- Tracking requests for capabilities that do not currently exist in the agent's toolkit.
- Discovering and documenting a superior architectural approach for a recurring task.
- Synchronizing learnings across multiple sub-agents in a complex Openclaw Skills workspace.
How Self-Improvement Skill Works
- Detection: The agent identifies a trigger event, such as a command error, a user correction, or a non-obvious solution discovery.
- Logging: The agent appends a structured entry to the appropriate file in the .learnings/ directory, assigning a unique ID (e.g., LRN-YYYYMMDD-XXX).
- Pattern Analysis: The system checks for recurring patterns using keys to link similar issues, increasing the priority of repeating problems.
- Review: At natural breakpoints, such as the start of a new task, the agent reviews existing logs to avoid past mistakes.
- Promotion: Broadly applicable learnings are distilled and moved to permanent workspace files like SOUL.md or TOOLS.md to influence future behavior.
- Extraction: High-value, reusable patterns are identified and can be exported as entirely new Openclaw Skills.
Self-Improvement Skill Setup
To get started with the Self-Improvement Skill, the recommended path is through the Openclaw Skills ecosystem:
# Install via ClawdHub
clawdhub install self-improving-agent
# Or manual installation
git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent
# Initialize the learning directory
mkdir -p ~/.openclaw/workspace/.learnings
For optional automatic reminders, enable the OpenClaw hook:
cp -r hooks/openclaw ~/.openclaw/hooks/self-improvement
openclaw hooks enable self-improvement
Self-Improvement Skill Data Schema & Taxonomy
The skill organizes data within a .learnings/ directory using three primary log files:
| File | Purpose | Key Metadata |
|---|---|---|
LEARNINGS.md |
Corrections and best practices | Priority, Area, Pattern-Key |
ERRORS.md |
Failures and exceptions | Error message, Reproducible status |
FEATURE_REQUESTS.md |
Capability gaps | Complexity estimate, Frequency |
All entries follow a strict header format [TYPE-YYYYMMDD-XXX] to facilitate easy searching and cross-referencing between sessions.
Self-Improvement Skill Advanced Features
- Inter-Session Communication: Utilize tools like sessions_send and sessions_history to share learnings across active agent threads.
- Simplify & Harden Feed: Automate the ingestion of recurring patterns to turn transient fixes into durable prompt guidance.
- Automatic Skill Extraction: Built-in scripts to transform verified, non-obvious solutions into standalone Openclaw Skills.
- Multi-Agent Activation: Support for Claude Code hooks, GitHub Copilot instructions, and Codex CLI integration.
- Promotion Gates: Logic-based rules that trigger the promotion of learnings to project memory only after they meet specific recurrence and frequency thresholds.
SKILL.md
Loading
Related Openclaw Skills
Meeting Prep
A professional AI skill that generates structured meeting preparation materials including research, agendas, and action items in HTML and Tencent Docs formats.
AI News Weekly Agent
An automated reporting agent that synthesizes weekly AI industry news, OpenClaw skill rankings, and research paper metrics into a clean Markdown digest.
Ozon-Wildberries Analyzer
A specialized market intelligence agent for Russian e-commerce platforms providing deep-dive analysis into Ozon and Wildberries performance metrics.
Amazon FBA Product Finder
A comprehensive Amazon FBA research agent that validates product ideas by analyzing rank, sales volume, and competition scores using a Jungle Scout-inspired methodology.
NFT Reference Tool
A comprehensive CLI-driven knowledge base for developers working with NFT standards, regulations, and strategic blockchain workflows.
Social Growth Hacker
A comprehensive strategy engine that generates personalized 90-day growth plans for major social media platforms.
Featured*
