video-director
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied prompts and descriptors from the production manifest (e.g.,
brief.prompt,shot.still.prompt) and interpolates them into calls for other skills and models. - Ingestion points: Untrusted data enters the agent context through the
manifest.jsonfile, specifically the prompt and description fields. - Boundary markers: No explicit delimiters or instructions are provided to the downstream agents/models to ignore embedded instructions within these prompts.
- Capability inventory: The skill has the capability to execute shell commands (ffmpeg, ffprobe) and invoke other Python scripts via
subprocess.runinscripts/direct.py. - Sanitization: There is no explicit sanitization or validation of the prompt content before it is passed to the ComfyUI client or the VLM judge.
- [COMMAND_EXECUTION]: The driver script
scripts/direct.pyusessubprocess.run()to execute various tasks. - Evidence: The
shmethod inscripts/direct.pyexecutes commands such asffmpeg,ffprobe, and variouscomfy_client.pyorjudge.pyscripts located in sibling skill folders. - Context: While these calls use
shlex.quotefor logging and pass arguments as a list (reducing shell injection risk), they represent the primary mechanism through which the skill interacts with the host system.
Audit Metadata