A specialized framework for AI agents to capture, analyze, and promote technical learnings and errors into permanent project memory.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install aditya
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 aditya 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 robust framework designed to turn every interaction into a growth opportunity for your AI coding agents. By systematically logging command failures, user corrections, and discovered best practices into structured markdown files, it ensures that your agent evolves over time. When integrated as part of your Openclaw Skills collection, it provides a persistent memory layer that prevents recurring mistakes and hardshells successful workflows.
This skill enables agents to maintain a dedicated .learnings directory where technical debt, knowledge gaps, and feature requests are tracked with high precision. By treating every error as a data point, Openclaw Skills empowered with this logic can eventually automate their own optimization, promoting local learnings to global project instructions or even extracting them into entirely new, reusable skills.
Self-Improvement Skill Use Cases
- Logging unexpected command or API failures to prevent future execution errors.
- Capturing user corrections to refine the agent's behavioral guidelines and style.
- Documenting project-specific 'gotchas' and non-obvious configurations for permanent memory.
- Tracking recurring patterns that indicate a need for a systemic architectural fix.
- Storing feature requests identified during development sessions to maintain a project roadmap.
How Self-Improvement Skill Works
- The agent identifies a trigger, such as a failed bash command, a direct user correction, or a discovered optimization.
- Information is captured in a standardized Markdown format within the .learnings/ directory, categorized as a Learning, Error, or Feature Request.
- The agent assigns a unique ID and metadata, including priority levels and technical area tags.
- Using Openclaw Skills tools, the agent can search past entries to detect recurring patterns and increment recurrence counts.
- High-value learnings are promoted to workspace-level files like SOUL.md or AGENTS.md to influence all future sessions.
- Verified solutions can be automatically extracted into standalone skills via provided scripts.
Self-Improvement Skill Setup
Openclaw is the primary platform for this skill. It uses workspace-based prompt injection with automatic skill loading.
# Installation 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
# Prepare the workspace
mkdir -p ~/.openclaw/workspace/.learnings
To maximize the utility of Openclaw Skills, ensure you enable the optional hooks for automatic error detection:
# Enable the self-improvement hook
openclaw hooks enable self-improvement
Self-Improvement Skill Data Schema & Taxonomy
The skill organizes data into specific log files within the .learnings/ directory to maintain clean Openclaw Skills architecture:
| File | Entry Type | Key Metadata Fields |
|---|---|---|
LEARNINGS.md |
Corrections & Best Practices | Priority, Area, Pattern-Key, Recurrence-Count |
ERRORS.md |
Command & Tool Failures | Error Message, Reproducible, Suggested Fix |
FEATURE_REQUESTS.md |
Missing Capabilities | Complexity, User Context, Frequency |
All entries follow a strict ID format: TYPE-YYYYMMDD-XXX (e.g., LRN-20250115-001).
Self-Improvement Skill Advanced Features
- Hook-Based Automation: Automatically trigger evaluation reminders or error detectors after tool use.
- Recurring Pattern Detection: Built-in logic to identify issues seen across multiple tasks and 30-day windows.
- Skill Extraction Helper: A script-based workflow to transform verified learnings into reusable Openclaw Skills.
- Inter-Session Communication: Tools like sessions_send allow the agent to pass critical learnings to other active sub-agents.
- Promotion Targets: Direct integration with project memory files such as CLAUDE.md and .github/copilot-instructions.md.
SKILL.md
Loading
Related Openclaw Skills
Swiggy Skill for Openclaw
Connect your AI agent to Swiggy to automate food delivery, Instamart grocery shopping, and Dineout restaurant bookings in India.
Markdown Anything
A high-performance conversion tool that turns PDFs, Office documents, images, and audio into LLM-ready Markdown.
Maps
A location-intelligence skill providing geocoding and routing capabilities without requiring external API keys.
Jina AI Reader & Search
A comprehensive toolset for converting web pages to LLM-friendly markdown and performing multi-step deep research.
MUKI Asset Fingerprinting
An advanced reconnaissance tool for red teamers to automate asset discovery and high-precision service fingerprinting.
Tor Browser Automation
Anonymously automate headless browser interactions and scrape .onion sites via the Tor network.
Featured*
