AI Blocks › Build and ship software
Agent self-improvement loop builder
Give it an AI agent and a test set. Get a loop that rewrites its prompts and skills until scores rise.
Agent self-improvement loop builder is a free AI skill for the job "Build and ship software". Give the instructions to an AI assistant that supports the required files and tools, then review its output.
Free · No account · MIT licence · Made by Orchestra Research · Level: expert · Thinking: high
Made by Orchestra Research. See the original on GitHub; we copied it as it is, with its licence.
What it does
An AI agent that gets better on its own, because a loop keeps testing it and rewriting its instructions until scores go up. Each round changes one thing, and a change that makes results worse is rolled back. The agent's prompts, skills, memory and tools all live as plain files in one folder. The loop runs your agent on a set of test tasks, looks at where it failed, and asks a model to rewrite the files. It scores the new version, keeps it only if it wins, and saves every change in version history. This needs Python, a package install and model access, plus a test set that gives clear scores.
You give it
- Your AI agent, or its prompt and tool files
- A set of test tasks with a way to score them
- Model access and a Python setup
- How many rounds to run (optional)
You get
- A setup that runs the improve-test-keep loop
- Score results after each round
- Updated prompt, skill and memory files
- A saved history you can roll back
Use it when
- You have an AI agent and a set of test tasks, and you want its score to climb.
- You are tuning prompts by hand and keep breaking things that worked.
- You want every change to the agent saved so you can undo a bad one.
- You want to find which skills or notes help the agent most.
How to use it
- Add it to your AI once: download it and add the folder to Claude, ChatGPT or Codex as a skill.
- Ask for the job in your own words. Your AI opens the skill when it fits.
- It works best when your AI can run Python code and call an AI model.