shinka-run

Automate batch runs of program mutations using the ShinkaEvolve CLI by specifying the task directory, results directory, and generation count. The process validates your run configuration, launches the main run, and verifies outputs with built-in guardrails.

To launch a ShinkaEvolve batch run, your task directory must contain an evaluate.py file, an initial.* file, and the required run configuration. The process validates these inputs before proceeding with the mutation experiments.

Yes, you can override mutation and embedding models during an automated evolution run by passing explicit flags. The launch process accepts model overrides alongside the required task directory, results directory, and generation count arguments.

The best way to ensure reproducible program mutation experiments is to use explicit task directories and results directories. The automation validates the run configuration and applies built-in guardrails to check outputs for consistent evolution exploration.

Your batch evolution run stops before processing generations because the validation phase failed. The process checks for the presence of evaluate.py, initial.* files, and the run configuration, halting the launch if any required input is missing.