What problem does it solve? Manual performance tuning is slow and error-prone: developers guess at optimizations, lose track of what worked, and accumulate bloated changes. This Skill automates metric-driven experimentation so every code change is measured, and only genuine improvements are kept. ## Core Features & Use Cases - Automated Experiment Loop: Repeatedly measures a baseline metric, asks an executor agent for one focused change, re-measures, and commits or resets based on the result. - Circuit Breaker & Simplicity Gate: Stops the loop after N consecutive non-improvements and discards changes whose metric gain does not justify the added code complexity. - Crash & Scope Handling: Resets trivially broken builds without penalizing the approach, enforces allowed file scope, and shuts down gracefully on SIGINT with a summary. - Use Case: Optimize a Go benchmark by pointing the loop at pkg/foo/foo.go with a go test -bench metric command, letting it iterate up to 30 times and report the best result via auto experiment summary. ## Quick Start Initialize an experiment session with a metric command and target files, then let the agent run the optimization loop and show the summary of kept improvements.
