deepagents-python-quickstart

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

• Third-party content exposure detected (high risk: 0.85). SKILL.md:5 and SKILL.md:7 instruct the runtime workflow to fetch LangChain live docs from the public web and ingest that page content into the agent “research-agent shape it shows,” which is outsider-authored free text via runtime HTTP/Docs MCP.

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

• Potentially malicious external URL detected (high risk: 0.90). The SKILL explicitly tells the agent to "Fetch that page" and implement the research-agent shape from https://docs.langchain.com/oss/python/deepagents/quickstart, so runtime-fetched content would directly control agent prompts/instructions.