cloudinary
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION] (MEDIUM): The skill provides templates that use
bash -cto wrap commands for environment variable persistence. This dynamic construction of shell strings using parameters likePUBLIC_IDandTIMESTAMPpresents a command injection risk if the agent does not properly sanitize these inputs. - [DATA_EXFILTRATION] (LOW): As its primary function, the skill reads local media files and uploads them to an external third-party domain (
api.cloudinary.com). While intended, this constitutes a data exposure surface. - [EXTERNAL_DOWNLOADS] (SAFE): Network activity is restricted to the official Cloudinary API endpoints, and no external code or scripts are downloaded for execution.
- [CREDENTIALS_UNSAFE] (SAFE): The skill correctly uses
vm0_secretsfor managing API keys and secrets, avoiding hardcoded credentials in the instructions.
Audit Metadata