cyberneticlibrary

Automatically improve AI agents

evolving-ai-agentsskillsetup L39,423
Orchestra-Research/AI-Research-SKILLs
What it does

Evolve agent prompts against benchmarks

Best for

When optimizing agent performance through iterative evolution cycles that automatically mutate prompts, skills, and memory.

Inputs
  • · agent workspace
  • · benchmark tasks
Outputs
  • · evolved prompts
  • · fitness scores
Requires
  • · a-evolve
  • · pyyaml
Preconditions
  • · agent workspace is a git repo
  • · benchmark adapter implemented
  • · manifest.yaml exists
Failure modes
  • · evolution diverges from benchmark goal
  • · prompt mutations reduce performance
  • · skill memory exceeds limits
Trust signals
  • · 76.8% SWE-bench Verified
  • · 79.4% MCP-Atlas #1
  • · open-source A-Evolve