spark-local-media-generation
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to perform multiple local shell operations using a specialized toolset and Python scripting.
- Evidence: Uses the
cinefoundryCLI for image/video generation (cinefoundry image,cinefoundry video) and system diagnostics (cinefoundry doctor). - Evidence: Includes a Python one-liner in
SKILL.mdused to evict the Linux page cache for model files to manage GPU memory constraints. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user content that is directly interpolated into generation commands, creating a potential surface for prompt injection.
- Ingestion points: User-provided
<product>descriptions,<image_prompt>, and<video_prompt>strings defined inSKILL.md. - Boundary markers: The instructions lack explicit delimiters or "ignore embedded instructions" warnings for the interpolated user content.
- Capability inventory: The agent can execute arbitrary CLI commands, write files to project directories, and manage system memory caches.
- Sanitization: No sanitization or validation of the input strings is mentioned in the workflow.
Audit Metadata