citation-deep-dive

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

• Third-party content exposure detected (high risk: 0.85). The required workflow calls Promptwatch MCP tools like get_citations / get_citation_top_pages / get_citation_llm_sources, which at runtime will ingest citation/page text or metadata fetched from third-party web sources (public pages and other outsiders’ content) into the agent’s LLM context via the MCP tool responses.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires adding the Promptwatch MCP server URL (https://server.promptwatch.com/mcp) to the agent config so the agent will call that external control-plane at runtime to obtain tool/instruction data, making it a required runtime dependency that can control prompts.