open-agent-sdk-typescript
Warn
Audited by Gen Agent Trust Hub on May 17, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill documents and provides examples for tools that execute shell commands (
Bash,BashSession). While intended for development tasks, this provides a direct path for command execution if the agent processes untrusted input without proper sanitization. - [DYNAMIC_EXECUTION]: An example
calculatortool utilizes theFunction()constructor to evaluate mathematical expressions. This pattern is equivalent to usingeval()and can lead to arbitrary JavaScript execution if the input string is not strictly validated. - [EXTERNAL_DOWNLOADS]: The documentation demonstrates the integration of external Model Context Protocol (MCP) servers using
npx(e.g.,@modelcontextprotocol/server-filesystem). This involves downloading and executing third-party code from the NPM registry at runtime. - [PROMPT_INJECTION]: Multiple examples configure the agent with
permissionMode: "bypassPermissions". This setting suppresses user confirmation prompts for sensitive operations like file deletion or shell execution, increasing the impact of potential prompt injection attacks.
Audit Metadata