agent-sdk-quickstart

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime path is the Agent SDK’s built-in “web fetch”/tool-use loop (and optional MCP integration), which can fetch public web content or other outsiders’ text at runtime and feed it into the model context as tool results; the SKILL.md explicitly states the SDK provides default tools including web fetch.