novita-sandbox

Pass

Audited by Gen Agent Trust Hub on Sep 24, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads its own instructions and reference documents from the novitalabs/novita-skills GitHub repository during installation and update processes.
  • [REMOTE_CODE_EXECUTION]: The skill includes instructions to download and execute the NodeSource setup script (https://deb.nodesource.com/setup_22.x) to install Node.js on Linux systems.
  • [COMMAND_EXECUTION]: The skill executes various shell commands to install dependencies, manage files, and interact with the novita-sandbox-cli wrapper.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection through the agent invoke command.
  • Ingestion points: The <payload> argument passed to the agent invoke command in SKILL.md and references/cli-reference.md.
  • Boundary markers: No explicit delimiters or boundary markers are defined for the payload string within the skill's instructions.
  • Capability inventory: The skill enables the agent to invoke deployed agents within an isolated sandbox environment, which may include network access or file system operations depending on the agent's configuration.
  • Sanitization: The skill does not implement input sanitization or validation, relying instead on the underlying CLI tool and agent implementation to handle the payload securely.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 24, 2026, 05:19 AM