lookalike-audience-builder

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

• Third-party content exposure detected (high risk: 0.85). The skill’s Website seed path uses WebFetch to retrieve a runtime page (e.g., “/customers” or “Trusted by” content) and then ingests the extracted customer/company text into the LLM context for pattern analysis, which is outsider-authored free text from public web content.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill performs a runtime WebFetch of user-supplied websites (example: https://acme-analytics.com/customers) and parses the fetched page to extract customer companies that directly control subsequent enrichment/search prompts and results, so this is a runtime external dependency that influences agent behavior.