GuardFall bypass slashes through open-source AI coding agent safety checks

Adversa AI's GuardFall attack exploits shell quoting differences to bypass command filters in 10 of 11 open-source AI coding agents; only Continue resisted. The technique uses decades-old shell tricks like empty quotes (e.g., r''m) to hide dangerous commands from text-based blocklists.