antigravity-guide

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime workflow ingests outsider-authored free text when Antigravity 2.0/IDE chat/agent modes read the user’s live messages (and uploaded media text) in the chat canvas/sidebar agent for processing, which is an untrusted text submission path at runtime.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). SKILL.md explicitly requires the agent to "MUST dynamically fetch the live page from the official sitemap" at runtime (e.g., https://antigravity.google/docs and the listed subpages), so those antigravity.google URLs are runtime external dependencies that can directly control prompts.