Each Agent variation was completed in a separate branch. See the open PRs for each experiment's implementation. I also committed run artifacts and eval results to github.