design-prd

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

• Third-party content exposure detected (high risk: 0.85). The skill’s required workflow performs competitive research by calling Firecrawl to fetch and scrape web pages at runtime, and then uses that scraped markdown content as “detailed patterns,” which is outsider-authored free text flowing into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill calls an MCP scrape at runtime to fetch external web content using the URL "<BEST_RESULT_URL>" (CallMcpTool firecrawl_scrape), which would be retrieved and injected into the agent context to directly influence prompts and outputs.