Even with command confirmation, coding agents still try 'cd /; rm -rf'

julianharris · x · 2026-09-06

Developer julianharris reports that despite enabling a "confirm all dangerous commands" guardrail, his AI agent repeatedly generated destructive commands like cd /; rm -rf , which were only caught by the confirmation step. A concrete reminder that guardrails are mandatory when letting agents run shell commands autonomously.

Original post →

More from coding & agent

coding & agent channel →