ai-content-pipeline

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

• Third-party content exposure detected (low risk: 0.10). The required workflows generate content from the agent's own prompts, but the Blog to Video pipeline (line 150) and the optional Tavily research step (lines 45, 188) have the agent summarize or search outsider-authored web content that it must actively fetch or is given, which is low risk rather than medium.

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

• Potentially malicious external URL detected (medium risk: 0.30). The npx skills add belt-sh/cli setup command at line 7 is an instructed remote-repository fetch that installs skill material, and the belt-sh org's first-party ownership of that CLI is plausible but not established, so it is treated as common/reputable unofficial (medium) under the recall-first override; the other setup pointers (the raw inference-sh instructions link and inference-sh/skills installs) come from the platform vendor and are strong, while belt app run calls are control-plane and documentation links are inert.