press-release-writing

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

• Third-party content exposure detected (high risk: 0.80). The required runtime workflow uses a web-search/fact-checking step (e.g., belt app run tavily/search-assistant and exa/search/answer), which fetches external/public web content and can feed that outsider-authored free-form text into the LLM context for summarization/verification.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires the inference.sh CLI (https://inference.sh) and shows runtime "belt app run ..." commands that invoke remote apps on that service, which will execute remote code and supply prompts/instructions at runtime.