runpod
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external media files (images and videos) using cloud GPUs. This creates a vulnerability surface where adversarial content embedded in those files could attempt to influence the agent or the processing model.
- Ingestion points: Input media files processed by the Python scripts in the
tools/directory (e.g.,image_edit.py,upscale.py). - Boundary markers: No explicit delimiters or instructions to ignore embedded content are documented in the skill instructions.
- Capability inventory: The skill executes local Python scripts, performs network requests to RunPod and Cloudflare R2 APIs, and modifies the
.envfile for configuration storage. - Sanitization: There is no mention of input validation or content sanitization for the processed media files.
- [COMMAND_EXECUTION]: The setup workflow involves executing local Python scripts using the
uv runcommand (e.g.,uv run tools/image_edit.py --setup). These scripts interact with the local environment to store configuration and manage cloud resources. - [EXTERNAL_DOWNLOADS]: The skill references and utilizes external Docker images hosted on GitHub Container Registry (
ghcr.io/conalmullan/video-toolkit-*) to provide the necessary environments for RunPod serverless endpoints.
Audit Metadata