A systematic framework for AI agents to log errors, user corrections, and new insights to enable continuous self-optimization.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install skilldevelop
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 skilldevelop 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 mistakes and user interactions in real-time. By maintaining a local .learnings/ directory, agents document unexpected command failures, knowledge gaps, and specific user corrections. This persistent memory ensures that the agent doesn't repeat the same errors and evolves its understanding of project-specific conventions over time.
Built to work seamlessly within the ecosystem of Openclaw Skills, this tool promotes high-value insights into core project files like AGENTS.md or SOUL.md. This process transforms one-off fixes into durable guidance, effectively turning your AI assistant into a more capable and autonomous developer with every interaction.
Self-Improvement Skill Use Cases
- When a terminal command or operation fails unexpectedly during a development task.
- When a user corrects the agent output with phrases like "No, that's wrong" or "Actually, do it this way."
- To document a knowledge gap where an API or library version is outdated compared to the agent's training data.
- To record a better approach discovered for a recurring development task to ensure it is used in the future.
- When a user requests a capability that doesn't exist, creating a roadmap for future feature implementation.
How Self-Improvement Skill Works
- The agent detects a trigger such as a non-zero exit code, a user correction, or a newly discovered best practice.
- It logs the event into a categorized Markdown file (LEARNINGS.md, ERRORS.md, or FEATURE_REQUESTS.md) within the
.learnings/directory. - Related entries are linked using a specific metadata schema and unique IDs to track recurring patterns.
- High-priority or recurring learnings are analyzed to see if they apply to the broader project context.
- The agent promotes these insights to permanent project memory files or extracts them into entirely new Openclaw Skills for use in other projects.
Self-Improvement Skill Setup
Install the skill via the ClawdHub registry for the best experience:
clawdhub install self-improving-agent
Alternatively, you can perform a manual installation by cloning the repository into your local directory:
git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent
Ensure your workspace is prepared by creating the necessary directory structure:
mkdir -p ~/.openclaw/workspace/.learnings
Self-Improvement Skill Data Schema & Taxonomy
The skill organizes data into three primary Markdown files, each serving a specific role in the agent's growth:
| File | Purpose | Key Metadata |
|---|---|---|
LEARNINGS.md |
Stores corrections, knowledge gaps, and best practices. | Priority, Area, Pattern-Key, Recurrence-Count |
ERRORS.md |
Captures command failures, exceptions, and stack traces. | Error Message, Context, Suggested Fix |
FEATURE_REQUESTS.md |
Tracks user-requested capabilities and implementation ideas. | Complexity, User Context, Frequency |
Every entry is assigned a unique ID in the format TYPE-YYYYMMDD-XXX to facilitate cross-referencing and resolution tracking.
Self-Improvement Skill Advanced Features
- Automatic Hook Integration: Enable activator and error-detector scripts to automatically prompt the agent to evaluate learnings after tasks.
- Skill Extraction: Automatically generate new, reusable Openclaw Skills from verified, high-value learnings using the extraction helper.
- Inter-Session Communication: Share critical learnings between different active sessions using built-in session tools like
sessions_send. - Recurring Pattern Detection: Ingest candidates from the simplify-and-harden workflow to identify systemic issues across different tasks.
- Multi-Agent Support: Extensive configuration options for Claude Code, Codex, and GitHub Copilot to ensure a consistent learning experience.
SKILL.md
Loading
Related Openclaw Skills
Text Detection Skills
A comprehensive suite of forensic tools for identifying AI-generated text through advanced NLP, pattern analysis, and machine learning classification.
Smart Cache
A high-performance caching utility providing intelligent data retention through LRU and LFU eviction strategies.
Rate Limit Pro
A robust rate-limiting utility for managing API quotas and request throttling through customizable user tiers.
Parent Portal Parental Controls
A comprehensive guardian interface for monitoring activity, managing screen time, and configuring safety filters for AI agent interactions.
Telegram Inline Buttons
A specialized skill for implementing interactive inline keyboards in Telegram to streamline agent-to-human decision-making and confirmations.
Talking Circle AI Video Generator
A specialized tool for Openclaw Skills that converts text or audio into animated circular avatar videos with automated lip-sync and blink cycles.
Featured*
