What problem does it solve? Iteratively improving code, prompts, or scripts against a measurable objective is slow and unstructured when done by hand. This Skill automates a self-referential improvement loop: a meta-agent proposes code modifications, evaluates each variant against a benchmark, and keeps only changes that measurably improve the target metric. ## Core Features & Use Cases - Evolutionary variant archive: Every evaluated variant is logged with hypothesis, metrics, and disposition (keep/discard/crash), with parent selection weighted by performance and exploration. - Empirical evaluation gates: Warmup and measured trials parse METRIC lines from task scripts, with optional correctness checks that must pass before a variant is kept. - Reporting and plateau detection: Generates CSV and HTML reports with lineage trees, trend charts, and automatic plateau warnings. - Use Case: Optimize a text summarizer's word-overlap score — the baseline scores 0.33, the meta-agent proposes keyword extraction, and the kept variant reaches 0.94, a 183% improvement tracked in the archive. ## Quick Start Ask the agent to run hyperagent on a script or prompt you want improved, specifying the metric to optimize and the command that benchmarks it.
