Adversarial Code Review for Pipeline Hardening

Summary

This skill facilitates the systematic hardening of BFCL training and evaluation pipelines through iterative adversarial review rounds. It employs an external LLM to identify potential bugs, which are then verified against the codebase to ensure accurate fixes and prevent regression.