linkedin

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 uses ClawLink to “read and act on” LinkedIn by browser/OAuth, which necessarily fetches outsider-authored LinkedIn profile/page/post content (public or user-accessible) and feeds it into the agent’s LLM context via the action outputs (e.g., linkedin_get_person, linkedin_get_post_content, linkedin_get_company_info).

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

• Potentially malicious external URL detected (high risk: 0.90). This skill invokes npx @useclawlink/cli at runtime, which fetches and executes remote code from the npm package (https://www.npmjs.com/package/@useclawlink/cli), meaning external code controls the agent's actions.