A self-evolution engine that enables AI agents to autonomously analyze history, identify inefficiencies, and perform protocol-constrained self-improvement.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install evolver-1-17-1
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 evolver-1-17-1 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 Capability Evolver?
The Capability Evolver is a sophisticated meta-skill designed for Openclaw Skills that enables recursive self-improvement. By programmatically inspecting runtime history, memory files, and execution logs, the agent can autonomously identify failures or bottlenecks and write new code to improve its own performance. It operates using the Gene Evolution Protocol (GEP), ensuring that every mutation is structured, auditable, and based on reusable assets.
This skill transforms static agents into dynamic systems capable of self-repair and forced optimization. Whether running in a controlled review mode or a fully automated loop, Capability Evolver ensures that your Openclaw Skills adapt to their environment and improve over time without manual developer intervention. It is environment-agnostic and supports dynamic tool detection to enhance its reporting and execution capabilities based on available local resources.
Capability Evolver Use Cases
- Autonomous debugging and patching of AI agent crashes in production workflows.
- Continuous optimization of reasoning patterns and prompt logic based on previous success capsules.
- Implementing recursive self-improvement cycles for advanced Openclaw Skills.
- Scaling agent capabilities by allowing them to discover and integrate new local tools dynamically.
- Maintaining high-performance standards in autonomous research or data processing tasks through forced optimization.
How Capability Evolver Works
- The engine scans the agent's memory and history files to detect patterns of failure, errors, or inefficient reasoning.
- It cross-references these patterns against the Gene Evolution Protocol store to identify appropriate genes or success capsules.
- Based on the configured EVOLVE_STRATEGY, the system generates a mutation or patch to resolve the issue or optimize logic.
- In Review Mode, the system pauses for human confirmation; in Mad Dog Mode, it applies the changes to the source code or memory immediately.
- The lifecycle of the evolution event is logged to a structured, append-only repository to maintain a verifiable history of all Openclaw Skills modifications.
Capability Evolver Setup
To install and configure the evolver for your Openclaw Skills environment, use the following commands:
# Install via npm
npm install -g evolver@latest
# Or clone from source
git pull && npm install
# Run the standard evolution cycle
node index.js
# Run with human review before changes are applied
node index.js --review
Capability Evolver Data Schema & Taxonomy
The skill manages its evolution metadata through the GEP Protocol using the following file structure:
| File Path | Description |
|---|---|
assets/gep/genes.json |
Definitions of reusable Gene structures for evolution. |
assets/gep/capsules.json |
A store of success capsules to prevent repeating past reasoning errors. |
assets/gep/events.jsonl |
An append-only log of evolution events with parent IDs for tree-like tracking. |
.env |
Configuration file for variables like EVOLVE_STRATEGY and EVOLVE_REPORT_TOOL. |
Capability Evolver Advanced Features
- Mad Dog Mode: Enables fully automated, continuous evolution loops for background autonomous agents.
- GEP Protocol: A standardized framework for auditable evolution that uses structured asset stores.
- Dynamic Detection: Automatically identifies and upgrades reporting tools when specific Openclaw Skills are detected in the workspace.
- Strategy Profiles: Supports multiple modes including repair-only, harden, innovate, and early-stabilize to control the risk profile of mutations.
- Safety Guardrails: Built-in risk mitigation including load average backing off and identity injection to prevent infinite recursion.
SKILL.md
Loading
Related Openclaw Skills
Samsung Health Connect CLI
A command-line tool for analyzing and reporting on Samsung Health Connect data exported to Google Drive.
claw-todolist
A structured conversational task management system implementing GTD and Eisenhower Matrix principles for AI agents.
MangaDex CLI
A command-line interface for MangaDex that enables searching, chapter tracking, and personalized recommendations for AI agents and developers.
AniList CLI Skill
A command-line interface for AniList that enables AI agents to search for media, manage watchlists, and track seasonal airing schedules.
OpenClaw Security Audit
A local-only auditing tool for OpenClaw that identifies configuration vulnerabilities and provides AI-driven remediation steps.
Summarize File
An automated tool for generating concise 2-3 sentence summaries from local text files, logs, and reports.
Featured*
