qwencloud-vision

SUSPICIOUS. The core vision/OCR behavior, credentials, and official API destinations are coherent with the stated purpose, and there is no evidence of credential theft or hidden exfiltration. The main concern is transitive trust: after running, the skill may prompt installation of another skill via an unpinned npx/skills workflow, which expands remote trust beyond the original vision task. Overall this looks like a legitimate vendor integration with moderate supply-chain and transitive-install risk, not malware.