For agents capable of spawning sub-agents or accessing external repositories, mandatory pre-execution constraints and hard checkpoints should be established rather than relying on post-hoc audits.