blog-image

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly supports "Search-Grounded Generation" (see references/prompt-engineering-blog.md) and notes MCP support for "Search grounding" (references/mcp-tools.md and references/gemini-models.md), indicating it can fetch and ingest open web/search results via the MCP/googleSearch feature and use those third‑party results to shape prompts and image generation decisions, allowing untrusted web content to materially influence tool behavior.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill config and setup invoke an external MCP package via npx (package @ycse/nanobanana-mcp, referenced at https://github.com/YCSE/nanobanana-mcp), which will fetch and execute remote code at runtime to provide the gemini_* tooling used for image generation and thus represents a runtime-executed external dependency.