zeroboot-vm-sandbox

Pass

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of external packages zeroboot (Python) and @zeroboot/sdk (Node.js/NPM) to interface with the sandbox service.
  • [EXTERNAL_DOWNLOADS]: Provides instructions to clone the project source code from https://github.com/adammiribyan/zeroboot for self-hosting scenarios.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and execute code which may be provided by untrusted external sources or generated by the AI agent.
  • Ingestion points: Code snippets are passed as string arguments to the sb.run() method in the SDKs and via the code field in the REST API request body.
  • Boundary markers: The provided examples do not include explicit prompt delimiters or instructions for the agent to ignore potentially malicious content embedded within the code to be executed.
  • Capability inventory: The skill facilitates remote code execution within a virtual machine environment and makes network requests to api.zeroboot.dev.
  • Sanitization: Security relies primarily on the hardware-enforced KVM and Firecracker isolation provided by the Zeroboot platform backend.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 12, 2026, 08:38 PM
Security Audit — agent-trust-hub — zeroboot-vm-sandbox