image-gen
Warn
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DATA_EXFILTRATION]: The skill defines a mandatory "LIBRARY-FIRST PROTOCOL" that directs the agent to search for and extract code from "D:/Projects/*". This directory is broad and likely contains sensitive developer assets, including source code, environment files, and local configuration.
- [INDIRECT_PROMPT_INJECTION]: User-supplied prompts are passed directly as arguments to a Python CLI tool executed via the Bash tool. There is no evidence of boundary markers or sanitization to prevent shell metacharacter injection. Ingestion points include chat-based prompts interpolated into commands like "python scripts/multi-model/image-gen/cli.py {prompt}".
- [EXTERNAL_DOWNLOADS]: The setup command "python scripts/multi-model/image-gen/cli.py --setup local" triggers a remote download of approximately 7GB of data, presumably model weights, without explicit source verification in the instructions.
Audit Metadata