A meta-skill for AI agent self-improvement that analyzes runtime logs to detect error patterns and generate structured evolution proposals.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install capability-evolver-pro
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 capability-evolver-pro 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?
Capability Evolver is a specialized meta-skill designed for the Openclaw Skills ecosystem, enabling AI agents to autonomously diagnose failures and improve their own reliability. By processing runtime logs through a deterministic analysis engine, it identifies repeated errors, regression signals, and cascading failures without the need for external LLM calls. This ensures high-speed, cost-effective, and private diagnostics that run entirely within your local environment.
The skill provides actionable intelligence through system health scores and prioritized recommendation plans. Whether you are managing a single agent or a fleet-wide deployment, Capability Evolver transforms raw log data into a structured roadmap for technical optimization and system hardening.
Capability Evolver Use Cases
- Production Incident Response: Quickly identify root causes of agent crashes by analyzing error patterns and cascades in production logs.
- Continuous Improvement Pipeline: Schedule daily log reviews to automatically generate prioritized evolution proposals and track health score trends over time.
- Pre-Deployment Health Checks: Use the analyze action in staging environments to catch regressions and block unhealthy deployments before they reach production.
- Multi-Agent Fleet Management: Aggregate logs across 50+ agent instances to find systemic issues and implement fleet-wide fixes simultaneously.
How Capability Evolver Works
- Log Ingestion: The skill accepts structured log entries containing timestamps, levels, messages, and file contexts.
- Pattern Detection: A deterministic engine groups logs to find repeated errors, dependency chain failures, and inefficiency patterns like excessive retries.
- Health Scoring: A proprietary algorithm computes a system health score (0-100) based on error rates, diversity, and time-based distribution.
- Strategy Application: Users select an evolution strategy (e.g., harden, innovate, or balanced) to shape the logic of the output.
- Proposal Generation: The skill outputs a structured JSON response including prioritized recommendations, risk assessments, and estimated health improvements.
Capability Evolver Setup
Install the skill via the OpenClaw CLI and integrate it directly into your agent's workflow using the following commands:
openclaw skill add capability-evolver
To run an analysis within your application:
const result = await agent.run('capability-evolver', {
action: 'analyze',
logs: userLogs
});
Capability Evolver Data Schema & Taxonomy
Capability Evolver utilizes a structured input/output schema to ensure compatibility across the Openclaw Skills library:
| Input Field | Type | Description |
|---|---|---|
| action | string | analyze, evolve, or status |
| logs | array | Array of objects with timestamp, level, and message |
| strategy | string | auto, balanced, innovate, harden, repair-only |
| Output Field | Type | Description |
|---|---|---|
| patterns | array | Detected issues with severity levels |
| health_score | number | 0-100 rating of system stability |
| evolution_id | string | Unique identifier for tracking proposals |
Capability Evolver Advanced Features
- Deterministic Engine: Guarantees reproducible results and zero hallucination risk with sub-100ms local processing.
- Evolution Strategies: Five distinct modes (Auto, Balanced, Innovate, Harden, Repair-Only) to tailor recommendations to your current project phase.
- Privacy-First Design: Processes all data locally within Openclaw Skills, requiring no external API keys or third-party data transfers.
- Fleet-Wide Diagnostics: Support for batch analysis to identify patterns across multiple distributed agent instances.
SKILL.md
Loading
Related Openclaw Skills
File Placement Enforcer
A high-precision protocol for Openclaw Skills to enforce zero-tolerance directory organization for documentation and utility scripts.
Eagle API Control
A powerful CLI-driven skill that enables automated control over the Eagle application for advanced digital asset management and organization.
RAG Production Engineering
A comprehensive methodology and toolset for architecting, optimizing, and operating production-grade Retrieval-Augmented Generation systems.
ArXiv CV Daily
An automated research assistant that fetches, filters, and summarizes computer vision papers from arXiv based on specific user topics.
MCP Builder
A comprehensive guide and toolset for creating Model Context Protocol (MCP) servers that allow LLMs to interact with external services via well-designed tools.
AI Agent Lending Protocol
A decentralized, wallet-based credit system that allows AI agents to borrow USDC on the Somnia network without requiring upfront collateral.
Featured*
