perplexity-reference-architecture

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime path sends the outsider-provided “query/topic/claim” strings from the research orchestrator and fact-checker directly into Perplexity via client.chat.completions.create (e.g., Step 1 search(query, ...) and Step 3 research(topic)), making user-authored free text part of the LLM prompt input.