tanstack-ai-sandbox
Pass
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill documents a developer library designed to improve the security of AI agents through sandboxing and capability isolation.
- [COMMAND_EXECUTION]: The sandbox provider exposes process execution capabilities that are governed by a configurable policy engine, allowing users to restrict or require approval for commands like sudo or curl.
- [EXTERNAL_DOWNLOADS]: Workspace provisioning involves cloning Git repositories and interacting with external MCP servers, but these operations are restricted to the sandbox environment.
- [CREDENTIALS_UNSAFE]: The library provides a createSecrets utility that uses symbol-keyed registries to prevent API keys and tokens from being accidentally written to snapshots or event logs.
- [PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection by processing external data from Git and MCP, but it provides explicit boundary markers and policies to mitigate these risks.
Audit Metadata