google-colab-runner
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Downloads machine learning models and tensors from Hugging Face resolve URLs. Hugging Face is a well-known service for hosting AI models.
- [EXTERNAL_DOWNLOADS]: Fetches the
cloudflaredbinary from Cloudflare's official GitHub releases to establish a tunnel for the orchestration mode. Cloudflare is a well-known service provider. - [REMOTE_CODE_EXECUTION]: Clones the ComfyUI repository and associated custom node packs from GitHub using pinned commits for version stability. GitHub is a well-known service.
- [COMMAND_EXECUTION]: Utilizes system utilities including
ffmpegfor media assembly andnvidia-smifor GPU hardware detection during the runtime environment setup. - [DYNAMIC_EXECUTION]: Assembles a self-contained Google Colab notebook (
.ipynb) by embedding the runner logic and workflows as a base64-encoded payload. This payload is unpacked and executed on the Google Colab instance at runtime. - [CREDENTIALS_UNSAFE]: Accesses the Hugging Face token through the standard Google Colab
userdata(secrets) mechanism. The skill correctly handles these credentials by reading them only into memory for request headers without logging or persisting them to disk. - [INDIRECT_PROMPT_INJECTION]: The skill ingests manifest files containing natural language prompts for video generation. These prompts are passed directly to machine learning models as parameters and are not interpreted as executable code or shell commands.
Audit Metadata