cloud-security

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (medium risk: 0.60). The Companion check section instructs the agent to run local shell commands (ls over user skill dirs) and propose running npx to install packages—actions that request local file access and modification and are outside the Cloud Security skill's stated purpose, so this is a hidden/deceptive behavioral instruction beyond the skill's scope.