video-creation-collaborator
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user requirements (product information and video demands) which flow through an 11-agent pipeline. This data eventually populates parameters for external media generation scripts.
- Ingestion points: User input is collected at the start of the workflow in
SKILL.md(Step 1) to define commodity selling points and video style. - Boundary markers: The instructions and agent prompts in
references/agent-prompts.mddo not utilize explicit boundary markers or delimiters to isolate user-supplied strings from agent instructions. - Capability inventory: The workflow triggers several Python scripts (
image_generator.py,audio_generator.py,video_compositor.py) that perform file operations and media processing based on the ingested requirements. - Sanitization: No explicit sanitization logic is implemented in the provided scripts to filter or escape user-provided strings before they are used as script arguments.
- [COMMAND_EXECUTION]: The skill relies on the agent executing multiple local Python scripts via shell commands to generate assets and assemble the final video. While these are intended tool calls, the command arguments are constructed based on processed user input, presenting a standard command-line interaction surface.
Audit Metadata