positioning

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow ingests outsider-authored free text when it reads user-provided URLs/file paths (public web pages or downloaded documents) via WebFetch/scraping in Phase 1, and when it incorporates scraped competitor content from runtime web research tools (e.g., Perplexity/Exa/Firecrawl/WebFetch) in Phase 2.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly fetches arbitrary user-provided URLs and competitor websites at runtime via WebFetch/WebSearch and MCP tools (e.g., URLs fetched by WebFetch or mcp__perplexity__perplexity_search), and that remote content is injected into agent prompts/instructions to drive research and question generation, so those external URLs can directly control the agent's behavior.