gemini-image-generation
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The Python script
scripts/generate-image.pycommunicates with Google's Generative Language API to submit generation requests and retrieve resulting image data. - [DATA_EXFILTRATION]: The skill attempts to load API credentials by reading configuration files from the local directory and the user's home directory, specifically targeting
~/.envand~/.env.local. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input that is subsequently interpolated into a shell command execution pattern.
- Ingestion points: User-provided image descriptions are extracted in
SKILL.md(Step 0) and passed as the--promptargument to the script. - Boundary markers: No explicit boundary markers or shell-escaping instructions are provided in the command templates within
SKILL.md. - Capability inventory: The skill has the capability to perform network operations (Google API) and write files to the local filesystem (
--output). - Sanitization: The skill lacks instructions for the agent to sanitize or escape user input before including it in the bash execution command.
Audit Metadata