What problem does it solve? Manual optimization of code performance, prompts, or prose requires constant human attention and guesswork. This Skill automates the entire experiment cycle — editing, measuring, keeping improvements, and reverting regressions — so optimization runs autonomously for hours without supervision. ## Core Features & Use Cases - Autonomous Keep/Revert Loop: Edits in-scope files, commits each experiment, verifies against a metric command, and automatically reverts regressions via git. - Domain Routing: Ships reference guides for code performance, test speed, prompt optimization, ML training, code quality, agent harnesses, visual performance, and subjective prose via adversarial debate judging. - Fast vs Epoch-Based Loops: Distinguishes metrics measurable in seconds from metrics that accumulate over days of real-world agent runs, with protocols for each. - Use Case: Point it at a slow benchmark suite overnight — it runs dozens of atomic experiments, keeps the ones that reduce execution time, and hands you a clean branch of only winning commits. ## Quick Start Ask the agent to run autoresearch on your project with a goal, the files it may edit, a metric with direction, and a shell command that prints that metric.
