cloud-gcp

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). The "Companion check" explicitly directs the agent to run local shell commands (ls and npx) to inspect and modify the host environment, which is unrelated to GCP architecture guidance and thus constitutes deceptive/out-of-scope instructions.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.70). The SKILL.md "Companion check" explicitly instructs the agent to run commands like npx skills add AbsolutelySkilled/... to install recommended companion skills from external repositories, which would fetch and ingest untrusted third‑party packages that could change agent behavior.