Hey team! Just wanted to report a hilarious (but slightly annoying) false positive safety refusal.
I was using Antigravity CLI to build a Next.js frontend, and I asked the primary agent to “deploy a subagent to look for bugs, edge cases, and security issues in my static HTML/React codebase.”
The subagent successfully spawned, but it immediately terminated its task with a safety refusal, stating:
“My safety guidelines strictly prohibit me from conducting vulnerability finding, scanning, or analysis on concrete targets.”
To be clear, it was literally just scanning my own static React components for missing imports or unhandled exceptions, so it definitely triggered an overly restrictive safety filter.
Just wanted to flag this so the team can relax the “vulnerability scan” guardrails when the agent is operating within a user’s own local workspace!