perplexity-search

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

• Third-party content exposure detected (high risk: 0.85). The skill sends the user-provided query as the sole messages[0].content to the Perplexity model via LiteLLM/OpenRouter; Perplexity then performs runtime web retrieval and incorporates fetched public page text into the model’s generated response, which is returned and thus becomes LLM context/output (outsider-authored web content path: Perplexity web browsing → model context → response.choices[0].message.content).