google-agents-cli-eval

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (medium risk: 0.30). The skill is a legitimate eval guide, but the mock-mode snippet returns fabricated "success" data whenever the API key is missing or a dummy, without disclosing that to the user, which is sketchy mock data rather than malicious intent.

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

• Third-party content exposure detected (low risk: 0.10). The required eval loop runs the user's own agent over the user's dataset and then has the coding agent read the resulting traces and results files, and those traces can contain web content that the evaluated agent fetched through tools like google_search or load_web_page, so outsider-authored text reaches the agent only when the agent under test searches or fetches it.