hf-cloud-serving-image-selection
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- Command Execution via Subprocess: The
scripts/mirror_image.pyscript utilizes thesubprocessmodule to automate interactions with the AWS CLI and Docker. The script employs security considerations by passing arguments as lists and avoiding shell execution, which helps prevent command injection vulnerabilities during the mirroring process. - External Resource Acquisition: The skill provides instructions and a script to download Deep Learning Containers from official AWS and Hugging Face registries. These operations target well-known infrastructure and are necessary for the skill's primary purpose of model deployment.
- Indirect Prompt Injection Surface: The workflow involves retrieving model configuration files (
config.json) from the Hugging Face Hub to determine architectural compatibility. This ingestion of external data is a potential surface for indirect prompt injection; however, the skill limits the data's impact by using it for specific structural validation (e.g., checking architecture strings) to select the correct container. - Credential Management: The skill describes the use of authentication tokens for accessing gated models and private registries. These values are handled via environment variables or managed through official CLI authentication mechanisms rather than being hardcoded within the skill's files.
Audit Metadata