newsletter-curation

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md Content Sourcing section explicitly instructs the agent to run infsh apps (e.g., tavily/search-assistant, exa/search) against public web sources—including social/community sites like reddit.com—which the agent is expected to read and use to decide what to include in the newsletter, creating a clear path for untrusted third‑party content to influence behavior.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires and uses the inference.sh CLI (https://inference.sh) at runtime (e.g., "infsh app run tavily/search-assistant" and other infsh app runs), which executes remote apps whose content can control prompts or run code and are relied on by the skill.