Hey, sorry I didn't see your comment until now. Yes I just added a Claude Code plugin that allows LoopGain to work with /goal.
/goal's checker knows when it's done but not when it has stopped improving. The plugin sets a Stop hook that, after each turn, runs your own verifier (such as failing-test count) and stops the loop when LoopGain sees it's converged or stalled instead of a guessed cap.
Any skill that iterates can just call the LoopGain library from a small script to decide when to stop instead of a hardcoded cap. I do this in my own skills. It recurses until the number of unresolved pieces stops dropping, no fixed depth limit.
You can also point any AI agent to the repo so it learns how LoopGain works and it will be able to help you set it up how you like. It's open-source.
/goal's checker knows when it's done but not when it has stopped improving. The plugin sets a Stop hook that, after each turn, runs your own verifier (such as failing-test count) and stops the loop when LoopGain sees it's converged or stalled instead of a guessed cap.
- two install commands:
Any skill that iterates can just call the LoopGain library from a small script to decide when to stop instead of a hardcoded cap. I do this in my own skills. It recurses until the number of unresolved pieces stops dropping, no fixed depth limit.You can also point any AI agent to the repo so it learns how LoopGain works and it will be able to help you set it up how you like. It's open-source.