pydantic-ai

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

• Third-party content exposure detected (high risk: 1.00). The documentation and examples (e.g., references/agents.md, references/tools.md, and SKILL.md) explicitly show the agent using WebSearch/WebFetch/FileSearchTools and URL inputs (ImageUrl, DocumentUrl, VideoUrl) and common web search tools (DuckDuckGo, Exa, Tavily) to fetch and ingest arbitrary public web content, which the agent is expected to read and use to drive tool calls and decisions.

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

• Potentially malicious external URL detected (high risk: 0.80). The example configures an MCPServerTool pointing at https://mcp.deepwiki.com/mcp which would be contacted at runtime to provide/expose tools (i.e., execute remote tool logic and influence agent prompts/results), meeting the criteria for a runtime external dependency that can execute code/control agent behavior.