autoresearch

Automated prompt evaluation runs iterative loops that baseline a skill, mutate its prompts, and commit improvements until performance thresholds are met. It requires a SKILL.md with a clear description and binary evaluation criteria to drive the loop safely.

Before running an autonomous prompt improvement loop, you need a Git repository for versioning and a target skill defined in a SKILL.md file. The SKILL.md must contain a clear name and description to establish the baseline evaluation starting point.

Automated skill optimization handles version control by creating an isolated branch, applying targeted mutations to prompts, and running an automated commit-and-run loop. This records results, compares scores against the baseline, and preserves the best configuration for deployment.

Yes, you can use automated self-improvement loops to refine tool configurations and workflows. The loop applies targeted mutations and isolated changes to instructions and constraints, evaluating them against binary criteria to discover better configurations over time.

You should not use an autonomous mutation loop if your target skill lacks a SKILL.md with a clear name and description, or if you cannot define binary evaluation criteria. Without these safe guards and a Git repository, the loop cannot establish a baseline or measure improvements.

The best way to establish a performance baseline is to evaluate the target skill directly from its SKILL.md file. This baseline evaluation provides the starting point for the iterative mutation loop to compare subsequent scores and preserve the best configuration.