autoresearch

Autonomously optimize a skill by running an evaluation loop of binary criteria and mutations, enabling self-improvement across multiple iterations without constant human input.

Binary evaluation criteria are 3-6 strict pass/fail checks used to evaluate prompt mutations, ensuring targeted improvements and measurable results during the autonomous optimization loop.

Start the autonomous loop by creating a dedicated branch, loading the target skill's SKILL.md baseline configuration, and running iterative mutations until reaching the target score or max iterations.

Yes, autonomous skill optimization integrates with your git workflow by enforcing strict commit rules, managing dedicated branches, and recording mutation results for audit throughout the iteration process.

Safety limits enforced during the mutation loop include strict guidance rules, cost controls, and maximum iteration thresholds to prevent uncontrolled changes while targeting a defined baseline score.

Yes, you can target any skill defined in a SKILL.md file, load its baseline configuration, and apply the autonomous mutation loop to iteratively improve its prompts and parameters.