evolve

Automated code optimization generates a pull request by running an evolutionary supervisor, explorer, and reviewer loop on isolated candidate branches, finalizing improvements without merging into main. You provide a natural-language goal or manifest path to start the process.

Evolutionary code optimization iteratively improves a code target through a multi-phase lifecycle: problem specification, baseline measurement, scoring, and reviewer gating. A supervisor orchestrates explorer agents to generate candidates and a reviewer gates their approval on isolated branches.

Start an evolutionary optimization run by providing a natural-language goal or an agent-evolve manifest path. The orchestration creates isolated evolve branches to test candidates, measures baselines, scores results, and opens a final PR for human review.

Automated code evolution supports both local and cloud backends for execution. You can restrict runs to specific module scopes, apply runtime-mode equivalence testing, and visualize the process using evolution graphs.

Safety boundaries in evolutionary optimization enforce scope-based pruning and ensure candidate branches never merge into main. The process finalizes by opening a pull request that remains open for mandatory human review before any integration.

The best way to orchestrate code improvements without risking the main branch is using an evolutionary loop on isolated candidate branches. This approach applies scope rules and reviewer gating to finalize a safe, human-reviewed pull request.