newsletter-curation

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

• Third-party content exposure detected (high risk: 0.90). The skill's "Content Sourcing" section instructs the agent to run search apps (e.g., infsh app run tavily/search-assistant and exa/search) including queries like "site:reddit.com [your niche] discussion this week" and references social/community sources (Twitter, Reddit, HN), which clearly pulls untrusted, user-generated third‑party content that the agent is expected to read and use to drive curation decisions.

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 calls the inference.sh CLI (https://inference.sh) at runtime (e.g., "infsh app run ...") to execute external apps (tavily/search-assistant, exa/search, etc.), meaning remote code and prompt-controlling behavior is executed and relied on as a required dependency.