A sophisticated meta-skill that allows AI agents to autonomously improve their own capabilities through feedback-driven execution analysis.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install auto-skill-evolver-1-5-1
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 auto-skill-evolver-1-5-1 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 Auto Skill Evolver?
Auto Skill Evolver is a meta-utility designed to close the feedback loop for AI agents. It functions as a specialized optimizer that observes how an agent performs a task, evaluates the execution logs (traces), and then rewrites the underlying skill definition to fix errors or improve performance. By treating skills as living documents, this tool ensures that your library of Openclaw Skills stays sharp and adapts to edge cases encountered during real-world usage.
The skill is built with a security-first mindset, employing atomic writes and strict section whitelisting to ensure that only intended parts of a skill file are modified. It bridges the gap between manual developer updates and fully autonomous agent evolution, providing a robust framework for continuous skill deployment with human-in-the-loop approval.
Auto Skill Evolver Use Cases
- Iterative performance tuning for complex coding or research tasks where initial prompts may fail.
- Automating the generation of refined usage examples based on real-world execution traces.
- Self-healing skill definitions that adapt to changes in external APIs or command-line interfaces.
- Managing and approving agent-proposed improvements via mobile chat interfaces using Openclaw Skills.
How Auto Skill Evolver Works
- The agent executes a target task using its current skill definition.
- Execution traces and user feedback are captured in local log files.
- A specialized optimizer sub-agent analyzes the trace to identify bottlenecks or failures.
- The optimizer generates a proposal for an updated SKILL.md file.
- Built-in security scanners check the proposal for high-risk commands or prompt injection markers.
- A formal proposal is created, requiring a human approval token (hash or interactive yes).
- Upon approval, the skill is updated atomically, and a versioned backup is stored for safety.
Auto Skill Evolver Setup
Ensure you have Python 3.8+ and the OpenClaw CLI installed. To begin optimizing a skill, use the training loop:
python skills/auto-skill-evolver/scripts/train_loop.py \
--skill-path "skills/target-skill/SKILL.md" \
--command "[\"your-agent-command\", \"--task\", \"do the thing\"]" \
--iterations 5 \
--interactive-each-iteration
For chat-based triggers, ensure your environment can route natural language intents to the optimize_skill.py script.
Auto Skill Evolver Data Schema & Taxonomy
The skill manages its lifecycle through a structured directory and file system within the target Openclaw Skills root:
| File/Directory | Purpose |
|---|---|
SKILL.md |
The primary skill definition being optimized. |
.skill_versions/ |
Stores historical backups of the skill for restoration. |
.proposed |
The staging file for the new skill version before approval. |
.proposed.meta.json |
Stores proposal metadata including SHA256 hashes and risk levels. |
.secure_workspace/ |
An isolated directory for temporary processing of traces and security scans. |
Auto Skill Evolver Advanced Features
- Multi-language intent recognition supporting both English and Chinese conversational triggers.
- Comprehensive version control system allowing for instant rollbacks to any previous state.
- Proposal-first workflow that enforces a mandatory human-in-the-loop security gate.
- Secure section whitelisting to protect core skill frontmatter and metadata from accidental modification.
- Machine-readable JSON output mode for seamless integration with mobile agent dashboards and UIs.
SKILL.md
Loading
Related Openclaw Skills
Video Editing with Canva AI
A powerful AI-driven video editing skill that automates transitions, text overlays, and background music using a cloud rendering pipeline.
High-Tempo Experiment Cycle
A disciplined weekly experimentation framework that transforms ad-hoc testing into a high-velocity, compounding growth engine.
Monetization Experiment Planner
A data-driven framework for growth teams to classify business models, analyze revenue cohorts, and execute high-impact pricing and upsell experiments.
AI Consulting System
A comprehensive AI consulting framework for SMEs that leverages McKinsey methodologies to diagnose business needs, design technical solutions, and track ROI.
Growth Team Structure Planner
A strategic AI tool for designing cross-functional growth teams and drafting executive-ready proposals for scaling startups.
Acpx Harness Delegation
A standardized protocol for delegating tasks from a coordinator agent to external executors like Claude Code or Codex via the ACP runtime.
Featured*
