Skill UI

This tool validates AI agent changes by running statistical regression tests to ensure performance doesn't degrade. It grades tool trajectories, detects flaky cases, and integrates with CI to block merges when significant regressions occur. Supports pass@k metrics and paired statistics for reliable evaluation.

This skill executes a single iteration of an automated research experiment. It reviews historical results, decides on a specific code modification, edits the target file, commits changes via git, and evaluates the outcome. Users can specify an experiment or select from a list. The process ensures only one change is made per iteration to isolate variables, preventing evaluation script modification. It supports strategy escalation based on run count and reports keep, discard, or crash status.