Shinka Run

Launches ShinkaEvolve program mutation runs from the CLI when you've already got an evaluate.py and initial file ready to go. The workflow enforces a safety net: check available models first with shinka_models, validate your LLM and embedding choices against what's actually configured, then confirm the batch config before firing. Runs are human-in-the-loop by default, stopping between batches so you can review outcomes and steer the next generation count or tweak the task system message based on what worked or failed. It's built for iterative experimentation rather than fire-and-forget, which makes sense when you're burning tokens on evolutionary search and want to course-correct without wasting budget on bad directions.

Install to Claude Code

npx -y skills add sakanaai/shinkaevolve --skill shinka-run --agent claude-code