novita-sandbox-skills

Pass

Audited by Gen Agent Trust Hub on Sep 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEDATA_EXFILTRATIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill documents the sandbox.commands.run API and novita sandbox exec CLI command for executing arbitrary shell commands within the sandbox environment. This is a core feature of the documented platform.\n- [EXTERNAL_DOWNLOADS]: Instructions are provided for installing the official novita-sandbox SDK and CLI from PyPI and npm. All external references point to the vendor's official documentation (docs.novita.ai) and GitHub repositories.\n- [CREDENTIALS_UNSAFE]: The documentation includes guidance on managing API keys and secrets. It specifically recommends using environment variables (NOVITA_API_KEY) and the platform's dedicated Secret service, which prevents real credential values from being exposed to the sandbox environment.\n- [DATA_EXFILTRATION]: The skill provides detailed instructions on configuring network egress rules (allow_out, deny_out) to control or disable outbound internet access from sandboxes, which is a key security measure against data exfiltration.\n- [DYNAMIC_EXECUTION]: The documentation describes how to dynamically build sandbox templates from OCI images and Dockerfiles, as well as how to execute code at runtime using the SDK's command and PTY modules.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 24, 2026, 05:37 AM