tuzi-image-gen

SUSPICIOUS. The skill’s capabilities mostly match its image-generation purpose, but its trust model is weaker than a direct-provider integration: it defaults to a third-party gateway (Tuzi), supports custom base URLs, asks users to paste API keys, stores them locally in plaintext, and uses unpinned `npx bun` execution. I do not see confirmed malware or covert exfiltration beyond the declared provider/gateway flows.

The code has a concrete command-injection risk in its proxy-enabled curl path because environment-derived values are interpolated into a shell command. It also permits API-key disclosure if GOOGLE_BASE_URL is configured to an untrusted endpoint. No clear evidence of intentional malware is present; avoid the shell by using fetch or pass curl arguments without shell interpretation and validate trusted endpoints.