A technical framework for AI agents to capture durable lessons, log failures, and systematically improve future performance within a project workspace.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install actual-self-improvement
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 actual-self-improvement 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?
The Self-Improvement skill provides a structured methodology for AI coding agents to learn from their mistakes and user interactions. Instead of repeating the same errors, this Openclaw Skills extension allows agents to document non-obvious failures, project-specific conventions, and missing capabilities. It creates a feedback loop where investigations and corrections are transformed into reusable project memory, ensuring that every debugging session adds long-term value to the development environment.
By differentiating between a skill root and a workspace root, it ensures that project-specific knowledge stays with the code while the automation logic remains portable. This skill is essential for developers looking to build a self-correcting agent ecosystem that grows more efficient with every task it completes.
Self-Improvement Use Cases
- Capturing project-specific conventions after a user correction to prevent future friction.
- Logging non-obvious API or tool failures to build a searchable knowledge base of errors.
- Documenting requested features or missing agent capabilities as structured, prioritized requests.
- Promoting proven workflow patterns into shared project memory files like CLAUDE.md or AGENTS.md.
- Extracting recurring, validated solutions into standalone Openclaw Skills for broader deployment.
How Self-Improvement Works
- Identify the Workspace Root to ensure all captured data is stored locally within the project repository.
- Initialize the learning directory using the Python helper to create structured markdown logs.
- Search existing entries before logging new data to avoid duplication and identify recurring patterns.
- Log specific events as a Learning, Error, or Feature Request using standardized CLI commands with metadata.
- Review relevant learnings periodically or specifically before starting high-risk technical tasks.
- Promote validated lessons into durable memory files to update the agent's core operational rules.
- Extract mature, recurring patterns into independent Openclaw Skills to scale improvements across different projects.
Self-Improvement Setup
To begin using this skill within your project, follow these installation steps:
# Ensure Python 3.11+ is available in your environment
# Initialize the .learnings/ directory in your active workspace
python3 scripts/learnings.py init --root /absolute/path/to/workspace
# (Optional) Set up shell hooks for automated error detection
# See references/platform-setup.md for specific shell configurations
Self-Improvement Data Schema & Taxonomy
The skill organizes its data within a .learnings/ directory at the workspace root to ensure persistence across sessions. The schema is organized as follows:
| Component | Description |
|---|---|
.learnings/LEARNINGS.md |
A log of corrections, knowledge gaps, and durable project conventions. |
.learnings/ERRORS.md |
Tracks non-obvious failures, tool exceptions, and specific API gotchas. |
.learnings/FEATURE_REQUESTS.md |
Captures missing capabilities and recurring friction points requested by the user. |
Pattern-Key |
A stable identifier used to link related or recurring issues for deduplication. |
Status |
Tracks the lifecycle of a lesson (e.g., active, promoted, or promoted_to_skill). |
Self-Improvement Advanced Features
- Automated skill extraction utility to convert mature project lessons into portable Openclaw Skills.
- Lightweight bash hooks for proactive error detection and session reminders at prompt start.
- Deterministic ID generation for all entries to support precise cross-referencing and "See Also" linking.
- Multi-agent synchronization by promoting rules to shared memory files like AGENTS.md and SOUL.md.
- Built-in evaluation framework in the evals/ directory to test trigger accuracy and output quality.
SKILL.md
Loading
Related Openclaw Skills
Crypto Market Intel
A zero-config market data pipeline providing real-time cryptocurrency, stock, and macro intelligence for AI agents without requiring any API keys.
Novita Multimodal Execution
A comprehensive multimodal skill for executing text-to-image, video generation, and speech-to-text tasks using the Novita AI API.
HTML to PPTX Conversion Skill
Transform any HTML file or web design into a pixel-faithful PowerPoint presentation with native chart support and automated scaling.
Adversarial Robustness Toolbox Skill
A comprehensive local reference tool for securing machine learning models using the Adversarial Robustness Toolbox.
CORS Tester
A command-line utility for testing, debugging, and securing Cross-Origin Resource Sharing (CORS) configurations on live URLs.
Capistrano Reference Skill
A comprehensive, zero-config reference tool for mastering Capistrano deployment concepts and best practices.
Featured*
