Skip to content
Get packs

A-Evolve Agent Optimizer

Guides an agent to iteratively evolve and optimize prompts, skills and memory against benchmarks

Source Download skill
Repository
Orchestra-Research/AI-Research-SKILLs
Stars
14,195
Licence
MIT
Files
9 files

What it does

This skill teaches an AI agent how to use A-Evolve, a framework that automatically improves another AI agent. The agent's prompts, skills, memory and tools are stored as files, and A-Evolve runs repeated cycles of solving benchmark tasks, scoring the results, mutating the files with an LLM, and keeping changes that are validated with git snapshots. Once installed, the agent can set up these evolution loops, write benchmark adapters, and explain the results.

Great for

  • Developers building self-improving agents who want automated prompt and skill tuning against a measurable benchmark
  • Teams that already have a working agent and a test suite or benchmark they want to optimize
  • Engineers who want git-versioned history of every agent change

Skip it if

  • Simple one-shot agent tasks or building multi-agent orchestration from scratch, where the skill itself says other frameworks such as CrewAI or LangGraph fit better
  • RAG pipeline tuning, which the skill says is better handled by LlamaIndex or Chroma
  • Anyone expecting a guaranteed score gain: evolution runs many LLM calls, can be costly, and the reported benchmark results come from the authors and are not independently verified here

Try asking

  • “Set up an A-Evolve loop to optimize my agent's system prompt against SWE-bench for 10 cycles”
  • “Write a custom benchmark adapter that scores my agent on code review quality”
  • “Show me the git diff between the first and last evolution cycle of my agent workspace”
  • “Explain how the solve-observe-evolve-gate-reload cycle works in A-Evolve”
Step-by-step guide: how to use A-Evolve Agent OptimizerInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 9 original files, the MIT licence and an attribution note.

    a-evolve.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip a-evolve.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip a-evolve.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Start with a small cycles count and batch size to limit API cost before running long evolutions
  • Keep a held-out holdout split of tasks so you can check whether gains generalize rather than overfit
  • Review the evolved prompts and skills in git before deploying the optimized agent

What's inside

  • references/api.md
  • references/architecture.md
  • references/design-patterns.md
  • references/examples.md
  • references/issues.md
  • references/README.md
  • references/releases.md
  • references/tutorials.md
  • SKILL.md

Redistributed unmodified from Orchestra-Research/AI-Research-SKILLs under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.