e2b
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to install packages from standard registries (pip, npm). These packages (e2b, e2b-code-interpreter) are part of the SDK for E2B, a well-known service for agent sandboxing. All references point to official domains (e2b.dev) and GitHub repositories (e2b-dev).
- [COMMAND_EXECUTION]: The skill demonstrates how to execute shell commands within an E2B sandbox microVM. This is the primary and intended purpose of the tool, providing secure isolation for executing code generated by AI agents.
- [INDIRECT_PROMPT_INJECTION]: The skill enables the execution of externally generated code, creating an attack surface for indirect prompt injection. However, the skill specifically advocates for the use of Firecracker microVM isolation to mitigate the impact of executing untrusted content.
- Ingestion points: Untrusted data enters the context through AI-generated code passed to
sandbox.commands.run()orrun_code(). - Boundary markers: The infrastructure itself (Firecracker microVM) acts as a hardware-level boundary.
- Capability inventory: Provides full shell and filesystem access within the isolated sandbox environment.
- Sanitization: Relies on system-level isolation rather than string-based sanitization.
Audit Metadata