javascript-sdk
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The SDK provides a notable attack surface for indirect prompt injection. It facilitates the ingestion and processing of untrusted data (user messages, file uploads, and external tool outputs) by AI agents.\n
- Ingestion points: The skill documents ingestion methods such as
agent.sendMessage(SKILL.md) andclient.uploadFile(references/files.md), and processing data through tools defined inreferences/tool-builder.md.\n - Boundary markers: Documentation examples lack delimiters or instructions to prevent agents from inadvertently executing commands embedded in untrusted content.\n
- Capability inventory: Agents can be configured with powerful capabilities including
codeExecution(references/agent-patterns.md),webhookToolfor external network calls, and file system interaction through Node.js (references/files.md).\n - Sanitization: The documentation does not provide examples of input validation or sanitization, leaving these critical security tasks to the end developer.\n- [DYNAMIC_EXECUTION]: The
references/tool-builder.mdfile contains code examples that utilize theeval()function to execute logic provided by an AI agent. While presented for educational purposes (e.g., a calculator tool), this pattern is inherently risky and can lead to arbitrary code execution if adopted without robust security controls.
Audit Metadata