autoagent

Why we included this project

AutoAgent automates the slowest part of building an LLM harness by letting the machine grind through it overnight. You write a Markdown file describing the kind of agent you want, and a meta-agent edits the system prompt, tools, agent configuration, and orchestration, runs the benchmark suite, checks the score, and keeps or discards each change before trying again. The harness under test lives in a single Python file, so you can see exactly what the agent is changing rather than trusting a black box. It suits teams that already develop against benchmarks and want systematic hill-climbing over prompt and tool variations instead of manual tweaks. If you have Docker and can run evaluation tasks locally, this gives you a concrete loop to hand to the machine for the night shift.