An autonomous self-optimization loop that enables AI agents to learn from past errors and successes to update their own internal protocols.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install dsb-auto-improve
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 dsb-auto-improve 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 Active Self-Improvement?
The Active Self-Improvement skill is designed to turn AI agents into evolving systems that get smarter over time without requiring constant manual prompts. By implementing a continuous loop of scanning, proposing, and applying changes, this skill ensures that agents within the Openclaw Skills ecosystem are always refining their behavior based on real-world data. It moves beyond simple logging by actively modifying the agent's core playbooks and instruction sets.
This skill is essential for developers looking to build resilient AI systems that learn from their mistakes. It proactively identifies repeated user corrections and stale facts, ensuring that the agent's internal memory and operational logic stay current. By integrating this into your Openclaw Skills setup, you create a self-correcting environment where the agent takes responsibility for its own growth and efficiency.
Active Self-Improvement Use Cases
- Reducing repeated errors by identifying patterns in previous failures.
- Automatically updating SOUL.md to refine agent personality and behavior.
- Memorializing successful wins into permanent workflows to ensure they are repeatable.
- Refreshing outdated project facts within memory files to maintain context accuracy.
- Implementing long-term fixes for frequent user-requested corrections.
How Active Self-Improvement Works
- Scan: The agent systematically reads localized data sources including .learnings/ERRORS.md, .learnings/LEARNINGS.md, and project memory files.
- Pattern Detection: The system looks for repeated mistakes, unused wins, and stale information that could hinder performance.
- Risk Assessment: Improvements are categorized into three levels: low-risk (reversible), medium-risk (requires notification), and high-risk (requires a proposal).
- Application: The agent applies low-risk updates directly to the playbook, while higher-risk changes are staged for review or accompanied by alerts.
- Playbook Update: The final step ensures changes are written to target files like AGENTS.md or SOUL.md, making the improvement permanent.
Active Self-Improvement Setup
To enable Active Self-Improvement within your environment, ensure your workspace has the necessary learning directories initialized:
# Create the learnings directory structure
mkdir -p .learnings
touch .learnings/ERRORS.md .learnings/LEARNINGS.md
# Ensure your agent has write permissions to its own protocol files
# Add the active-self-improvement skill to your local Openclaw Skills configuration
Active Self-Improvement Data Schema & Taxonomy
The skill organizes and updates data across a specific set of files to maintain the improvement loop:
| File Name | Purpose |
|---|---|
.learnings/ERRORS.md |
Stores logs of failures used for pattern detection. |
.learnings/LEARNINGS.md |
Records successful outcomes and valuable workflow wins. |
SOUL.md |
Updated to refine the agent's core identity and behavioral rules. |
AGENTS.md |
Updated to improve specific task-based instructions and capabilities. |
| Memory Files | General project context and facts that are refreshed periodically. |
Active Self-Improvement Advanced Features
- Autonomous risk-aware update engine that distinguishes between direct application and proposal-only changes.
- Pattern recognition across multiple Openclaw Skills to optimize cross-skill interactions.
- Deep integration with agent memory to prune stale data and prioritize relevant facts.
- Support for reversible updates, allowing for safe experimentation with new protocols.
- Automated playbook generation that translates raw logs into actionable instructions.
SKILL.md
Loading
Related Openclaw Skills
Task Extractor
A robust workflow skill designed to parse complex user messages into discrete, trackable, and verifiable action items.
UniProfit Mail Assistant
A specialized AI agent skill for sending verified emails through the UniProfit API using Openclaw Skills.
Tomato Fiction AI Short Drama Batch Generator
A comprehensive AI automation tool that converts multi-chapter novels into serialized short-form videos with automatic episode splitting and asset generation.
File Management
A robust toolkit for performing safe, dependency-free filesystem operations using native Unix and macOS utilities.
Spawner
The Spawner skill enables the delegation of complex, long-running tasks to independent sub-agents using efficient model routing.
Podcast Digest
An advanced AI tool that converts podcast audio and transcripts into structured knowledge assets, interactive visual reports, and cross-episode matrices.
Featured*
