Write, check and dry-run AI agent programs in Lugh, where permissions, budgets and human approval are part of the code. Every action passes a policy gate and each run ends with a policy log.