gemini-antigravity

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill includes examples that embed API tokens verbatim in command-line/network flags (e.g., Authorization:Basic TOKEN), which would require the LLM to output secret values directly.

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

• Third-party content exposure detected (high risk: 0.70). SKILL.md describes mounting external repositories/inline content at runtime via --source REPO_URL:TARGET_DIR and --agent-file ... "CONTENT", which can cause outsider-authored free text (e.g., repo files or provided inline content) to be read by the agent’s LLM during execution.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly runs a runtime "clone https://github.com/octocat/Spoon-Knife, run the test suite" and shows mounting a repo via "--source "https://github.com/my-org/backend:/workspace/repo\"", meaning external GitHub repositories are fetched at runtime and can supply/execute remote code (tests), so these Git URLs pose a runtime-executed dependency.