newsletter-curation

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 belt app run tavily/search-assistant / exa/search to fetch and ingest public web/social/community content (e.g., articles, Reddit threads) into the LLM context for curation, which is outsider-authored free text.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill invokes the belt CLI to run remote apps (e.g., belt app run tavily/search-assistant), which at runtime calls and executes apps on the inference.sh service (https://inference.sh) that control prompts/behavior, making that external URL a required runtime dependency.