Agent skill · software engineering · aradotso
hyperagents-self-improving
hyperagents-self-improving — an agent skill for coding agents.
Why this skill is useful
Adds executable scripts and a framework for self-improving agents that optimize tasks using meta-agents, which is not commonly found in public resources.
What it needs
Requires bzip2-devel, cmake, graphviz, graphviz-devel, libffi-devel, ncurses-devel, ninja-build, python3.12-devel, zlib-devel installed locally. Requires anthropic, gemini, openai account access. About 5k tokens when loaded. Last updated 2026-07-09. 70 stars on the source repository.
What this skill does
generateloop.py └── metaagent.py ← proposes improvements (diffs) to task agent code └── taskagent.py ← executes tasks in a target domain └── agent/ ← foundation model wrappers (OpenAI, Anthropic, Gemini) └── domains/ ← domain-specific evaluation code └── runmetaagent.py ← helper to run meta agent and get diffs bash sudo dnf install -y python3.12-devel sudo dnf install -y graphviz graphviz-devel cmake ninja-build \ bzip2-devel zlib-devel ncurses-devel libffi-devel bash python3.12 -m venv venvnat source venvnat/bin/activate pip install -r requirements.txt pip install -r requirementsdev.txt bash OPENAIAPIKEY=youropenaikeyhere ANTHROPICAPIKEY=youranthropickeyhere GEMINIAPIKEY=yourgeminikeyhere bash bash ./setupinitial.sh bash docker build --network=host -t hyperagents . bash Basic run on a domain python generateloop.py --domains <domain> Examples: python generateloop.py --domains coding python generateloop.py --domains math python generateloop.py --domains reasoning bash python runmetaagent.py bash Combine multi-part ZIP archive zip -s 0 outputsosparts.zip --out unsplitlogs.zip unzip unsplitoutputs.zip python Typical invocation pattern (from source) import argparse from metaagent import MetaAgent from taskagent import TaskAgent parser = argparse.ArgumentParser() parser.addargument("--domains", type=str, required=True) parser.addargument("--outputdir", type=str, default="outputs/") parser.addargument("--numiterations", type=int, default=10) parser.addargument("--model", type=str, default="gpt-4o") args = parser.parseargs() python from metaagent import MetaAgent meta = MetaAgent( model="claude-3-5-sonnet-20241022", # or "gpt-4o", "gemini-pro" taskdescription="Solve competitive programming problems", currentagentcode=open("taskagent.py").read() ) Get proposed diff/improvement diff = meta.proposeimprovement() print(diff) python from taskagent import TaskAgent agent = TaskAgent( model="gpt-4o", domain="coding" ) Run agent on a task result = agent.run(taskinput="Solve: given …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills aradotso/hyperagents-self-improving