vision-support

The code is a recognizable image-to-AI-provider CLI with proxy support and configuration management. It contains no clear malware, credential harvesting routine, persistence mechanism, destructive behavior, reverse shell, or covert exfiltration endpoint. Security risks include plaintext API-key storage and replication, sending sensitive image data and credentials through automatically detected or user-configured proxies, arbitrary custom API destinations, Google keys in URLs, and possible SSRF/resource abuse through remote image URLs. These behaviors are largely consistent with the stated functionality but should be reviewed and constrained in security-sensitive deployments.

The manifest itself does not contain overtly malicious indicators (no http:// URLs, no non-registry dependency specifiers), but it registers a postinstall step that runs local JS (bin/postinstall.mjs) and ships install scripts (install.mjs, install.sh). That makes arbitrary code execution during install possible. You should review the contents of bin/postinstall.mjs, install.mjs, and install.sh (and any files they call) before installing. Treat this as a moderate security risk until those scripts are audited.