linkedin-outbound-angle

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

• Third-party content exposure detected (high risk: 0.85). The skill’s workflow ingests a LinkedIn profile via web_fetch (Format A) or extracted text from a user-provided URL/screenshot/PDF (Formats B/C), which can include outsider-authored free text from the prospect’s public profile page or other non-user documents; that readable text is then used as LLM context for analysis and angle generation.

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

• Potentially malicious external URL detected (high risk: 0.70). The skill explicitly fetches arbitrary LinkedIn profile pages at runtime using web_fetch (e.g., https://www.linkedin.com/...), and that fetched content is injected into the agent's analysis and used to generate prompts, so external page content can directly control the agent's instructions.