What is a Darwin Gödel Machine (DGM)?

A DGM is a framework for self-improving AI where a population of agents is evolved through mutation (via LLMs) and selection based on fitness evaluations against specific benchmarks.

How does this skill handle agent evolution?

It implements an iterative loop: sampling a parent agent from an archive, mutating its code using an LLM mutator, evaluating the new variant's performance, and adding it back to the archive if it meets novelty and fitness thresholds.

Is this suitable for production codebases?

This skill is intended for research and the development of experimental autonomous systems; because it involves agents modifying their own code, it should be used within secure, sandboxed environments.

Does it support long-term memory?

Yes, it includes patterns for Long-Term Memory (LTM) using the OMNE framework, allowing agents to store, retrieve, and consolidate experiences to facilitate lifelong learning.

Self-Evolving AI Agent Framework

Name: Self-Evolving AI Agent Framework
Author: plurigrid

byplurigrid

•

数据科学与机器学习

Implements Darwin Gödel Machine patterns to build AI agents that autonomously improve their own code and capabilities through open-ended evolution.

The Self-Evolving Agent skill provides a sophisticated framework for building AI systems capable of recursive self-improvement. Based on Darwin Gödel Machine (DGM) research, it enables agents to mutate their own codebase, evaluate performance against benchmarks, and maintain a diverse archive of high-performing variants. This skill is ideal for developers creating autonomous systems, lifelong learning agents, or AI researchers exploring open-ended evolution and Artificial Superintelligence (ASI) foundations within the Claude Code environment.

主要功能

017 GitHub stars

02Darwin Gödel Machine (DGM) architecture implementation

03LLM-based code mutation and evolution loops

04Long-term memory (OMNE) for lifelong learning

05Support for multi-agent co-evolution patterns

06Automated fitness evaluation and archive management

使用场景

01Building coding agents that refactor and optimize their own logic based on execution feedback

02Developing autonomous research agents that adapt their tools and strategies over time

03Creating persistent AI assistants with long-term memory and evolving skill sets

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