technical-blog-writing

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

• Third-party content exposure detected (high risk: 0.70). The Quick Start workflow explicitly calls an external search app ("belt app run exa/search --input ...") to research topics, which implies the agent will fetch and read open/public web search results and thus ingest untrusted third-party content that can influence writing and follow-up actions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly relies on the inference.sh CLI (https://inference.sh) and shows runtime commands like belt app run infsh/python-executor which execute remote code via that service, meaning external content at inference.sh is used at runtime to execute code and control outputs.