nanobanana-skill

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

• Third-party content exposure detected (high risk: 0.80). The script conditionally enables a Google Search tool that will let the model fetch and consume public web content—see nanobanana.py where tools = [types.Tool(google_search=types.GoogleSearch())] is passed into the model when --enable-google-search is used—allowing untrusted third-party (web) content to be read and interpreted by the agent.