agent-sandbox

Pass

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides logic to execute shell commands using Node.js execSync within a local or Docker-based sandbox environment. Evidence: Found in the exec methods of both AgentSandbox and DockerSandbox.
  • [COMMAND_EXECUTION]: Security is enforced through a blocklist of restricted shell commands and patterns to prevent destructive actions or unauthorized access. Evidence: Defined in the DEFAULT_BLOCKED array in sandbox.ts.
  • [DATA_EXFILTRATION]: Path restriction logic is implemented to prevent the agent from accessing sensitive directories or escaping the designated workspace. Evidence: checkPathAllowed method and the deniedPaths configuration in sandbox.ts.
  • [EXTERNAL_DOWNLOADS]: The container-based strategy utilizes well-known Docker images from official registries to provide an isolated runtime. Evidence: Default use of node:20-slim in DockerSandbox.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 13, 2026, 09:15 PM