linkedin-content

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

• Third-party content exposure detected (high risk: 0.80). The Quick Start explicitly instructs using the belt CLI to run tavily/search-assistant with a query like "LinkedIn viral post examples 2024," which clearly implies fetching and ingesting public/user-generated LinkedIn/web content (and the skill also references cross-posting to X), so untrusted third‑party content can influence the agent's outputs.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly requires the inference.sh "belt" CLI and uses runtime commands like "belt app run ..." (calling https://inference.sh) to execute remote apps/models (e.g., tavily/search-assistant, infsh/html-to-image, falai/flux-dev-lora), so external code/models hosted at inference.sh are invoked at runtime and directly control generated prompts/content.