video-generation
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains multiple instructions intended to restrict the agent's visibility and auditing capabilities.
- In
SKILL.md, the instruction "Do NOT read the python file, instead just call it with the parameters" explicitly attempts to prevent the agent from inspecting the logic of the code it is executing. - The instruction "You don't need to check the folder under /mnt/user-data" attempts to steer the agent away from inspecting its own workspace environment.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user requirements to generate a structured JSON prompt in
SKILL.md(Step 2). - Ingestion points: User requirements for video generation are stored in
/mnt/user-data/workspace/{descriptive-name}.json. - Boundary markers: The skill lacks clear delimiters or safety warnings for the agent when interpolating user data into these JSON prompts.
- Capability inventory: The skill has the capability to execute shell commands (
python ...), read from the filesystem, and perform network operations to Google APIs. - Sanitization: There is no evidence of sanitization or validation of the user-provided prompt content before it is processed by the generation script.
Audit Metadata