research

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 loads and follows the research_workflow from the external deep-research MCP server, which (per its purpose) can fetch and ingest public web content / scraped pages into the agent’s LLM context as free-form text.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs the agent to "Load the research_workflow MCP prompt from the deep-research server" at runtime, meaning it fetches remote prompt content that directly controls agent instructions and is required for execution (resource: deep-research MCP server).