Optimize

After you've run discover and committed your baseline, this is the loop that drives all subsequent optimization work. It orchestrates rounds of structured experimentation: spawns subagents with concrete briefs, runs cross-cutting analysis between rounds, dispatches ideator subagents on stall, enforces verifier hooks pre and post. The width argument (subagents=N) scales from 1 for serial workloads up to whatever your binding resource allows, but the real value is the structure itself. Frontier-based parent selection, annotation discipline, and the scan-subagent cycle apply whether you're running one experiment at a time on a single GPU or fanning out across cores. Don't skip it just because your benchmark forces serial execution. You'd lose the scaffolding that makes evo autoresearch instead of ad-hoc iteration.

Install to Claude Code

npx -y skills add evo-hq/evo --skill optimize --agent claude-code