Maintain a private global AI-work benchmark across repositories by checkpointing code before AI mutation, harvesting reusable difficulties afterward, validating privacy and capture policy, and running candidates in restored evaluator-isolated workspaces. Use before the first file mutation in a subst
Maintain a private global AI-work benchmark across repositories by checkpointing code before AI mutation, harvesting reusable difficulties afterward, validating privacy and capture policy, and running candidates in restored evaluator-isolated workspaces. Use before the first file mutation in a substantive repository task when an initialized global Bench enables auto or suggest capture; use after s