skills.rest
Orchestra-Research
Orchestra-ResearchOFFICIAL
Explore all skills by this author →
GitHub

evolving-ai-agents

Evolves and optimizes AI agents through LLM-driven mutation of prompts, skills, and memory against benchmarks.

13.0k|930|Updated Nov 3, 2025
Search Tags:#prompt optimization#llm#skill discovery#agent evolution#self-improving agents#benchmark evaluation#a-evolve
One-click install
npx skills add https://github.com/Orchestra-Research/AI-research-SKILLs --skill evolving-ai-agents-orchestra-research
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: evolving-ai-agents
Source: https://github.com/Orchestra-Research/AI-research-SKILLs/tree/main/14-agents/a-evolve
Command: npx skills add https://github.com/Orchestra-Research/AI-research-SKILLs --skill evolving-ai-agents-orchestra-research

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires a-evolve, pyyaml, and includes references (resource) components.

What problem does it solve? Improving an AI agent's performance normally requires manual prompt tuning, hand-written skills, and trial-and-error iteration. This Skill automates that process by running iterative solve-observe-evolve cycles that mutate an agent's workspace files (prompts, skills, memory) based on benchmark feedback, with git-versioned snapshots for rollback. ## Core Features & Use Cases - Automated Evolution Loop: Runs solve, observe, evolve, gate, and reload phases per cycle, mutating workspace files with an LLM-driven engine and stopping on score convergence. - Pluggable Interfaces: Implement custom BaseAgent, BenchmarkAdapter, and EvolutionEngine classes to evolve any agent on any domain-specific benchmark. - Built-in Seeds and Benchmarks: Ships with preconfigured agents and benchmarks for SWE-bench Verified, MCP-Atlas, Terminal-Bench 2.0, SkillsBench, and ARC-AGI-3. - Use Case: You have a code review agent that underperforms. Point the Evolver at its workspace and a custom review benchmark, run 10 cycles, and inspect the evolved system prompt and discovered skills via git tags. ## Quick Start Install the a-evolve package with pip, then ask the agent to run an evolution of the built-in swe seed against the swe-verified benchmark for 10 cycles and report the final score.

Frequently Asked Questions about evolving-ai-agents

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I evolve an AI agent against a benchmark in Python?▼

Install a-evolve, then create an Evolver with an agent workspace and benchmark name, and call run with a cycle count. The loop solves task batches, evaluates trajectories, mutates workspace files, and returns an EvolutionResult with the final score and history.

What is the difference between A-Evolve and DSPy or LangGraph?▼

A-Evolve optimizes existing agents by evolving prompts, skills, and memory against benchmarks, while LangGraph and CrewAI build agent orchestration from scratch. DSPy handles prompt-only optimization without skill or memory evolution.

Can I use A-Evolve with OpenAI or AWS Bedrock models?▼

Yes, the evolver model is configurable via EvolveConfig.evolver_model. Supported providers include Anthropic, OpenAI, AWS Bedrock, and LiteLLM for multi-provider access, with authentication through the corresponding API keys or AWS credentials.

Why does my evolution score plateau after a few cycles?▼

Plateaus usually come from a batch size too small to expose failure diversity or from the evolver not seeing scores. Increase batch_size to 15-20, keep trajectory_only as False, and adjust egl_threshold and egl_window for your domain.

How do I add a custom benchmark for agent evolution?▼

Subclass BenchmarkAdapter and implement get_tasks to return Task objects and evaluate to return Feedback with success, a 0.0-1.0 score, and a detail string. Rich detail text helps the evolver decide what to mutate.

When should I not use agent evolution frameworks?▼

Avoid evolution for one-shot agent tasks with no iteration, multi-agent orchestration building, RAG pipeline tuning, or prompt-only optimization. Evolution requires a measurable benchmark and iterative cycles to be worthwhile.

Related Skills

Explore More Software Engineering Skills→
Supporter09Supporter09

evolving-ai-agents

Mutate prompts, skills, memory, and tools against benchmark feedback.

Community
Advanced
qcmuuqcmuu

evolving-ai-agents

Evolve AI agents across prompts, skills, and memory using LLM-guided mutation cycles.

Community
Advanced
box755box755

evolving-ai-agents

Run automated solve–observe–evolve cycles to mutate versioned agent workspace files.

Community
Advanced
VYRE-StudiosVYRE-Studios

fine-tuning-with-trl

Fine-tune HuggingFace Transformers language models with TRL for RLHF alignment.

Official
Advanced