Lask 0.15.0 Lets Developers Declare and Statically Check What AI-Written Tasks Can Access

Coding agents often generate tasks containing shell commands that few people review in full, leaving unclear what files or networks those commands can touch. The author builds Lask, a task runner that runs each command in a declared Docker container, and version 0.15.0 adds a feature called within. It lets developers declare permitted files and networks, which the type checker verifies before anything executes. The article argues that manual code review is an unreliable defense, citing the xz Utils backdoor and studies showing most dependency updates are only partially reviewed. It also notes that AI tools have increased code volume and review load, making human checking harder. The author proposes verifying the boundary around code rather than the code itself, while acknowledging that this static check is not a formal proof and has limitations.