ar-run
alirezarezvani/claude-skills
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.
