Does this require constant human supervision?

No, the framework is designed for autonomous, open-ended evolution. Once the fitness criteria and benchmarks are set, the agent can iterate through generations of code improvements independently.

How does the self-evolving agent skill improve code?

The skill provides a structure where an agent acts as its own developer, using feedback from execution benchmarks to guide an LLM-based mutator in rewriting its code for better performance.

What role does Long-Term Memory play in this skill?

Long-Term Memory allows the agent to store past experiences and successful mutations, ensuring that evolution is cumulative rather than random and preventing the loss of useful learned behaviors.

What is a Darwin Gödel Machine (DGM)?

A Darwin Gödel Machine is a research-backed architecture for self-improving AI that uses an evolutionary loop—sampling, mutating with an LLM, and validating against benchmarks—to autonomously upgrade its own capabilities.

Self-Evolving AI Agent Patterns

Name: Self-Evolving AI Agent Patterns
Author: plurigrid

byplurigrid

•

数据科学与机器学习

Implements Darwin Gödel Machine patterns to enable autonomous code evolution and self-improvement for AI agents.

This skill provides a comprehensive framework for building AI agents capable of modifying their own source code and logic through open-ended evolution. By leveraging Darwin Gödel Machine (DGM) architectures, agents can autonomously sample, mutate via LLMs, and evaluate their own codebases to achieve higher performance on specific benchmarks. It is an essential tool for developers and researchers building autonomous systems that need to adapt to complex environments, maintain long-term memory, and improve their internal logic without manual human intervention.

主要功能

01Automated fitness evaluation and archive pruning for agent diversity

027 GitHub stars

03Darwin Gödel Machine (DGM) architecture for autonomous code modification

04LLM-based mutation loops for iterative logic and capability improvement

05Multi-agent co-evolution patterns for shared knowledge transfer

06Long-term memory (LTM) integration for persistent agent learning

使用场景

01Developing autonomous coding agents that refactor and optimize their own source code

02Creating continual learning systems that adapt to new tasks through feedback loops

03Researching Artificial Superintelligence (ASI) pathways via open-ended evolution

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add plurigrid/asi self-evolving-agent

For use in Claude.ai and ChatGPT

Download Skill