ai-sdk-core
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill documents the implementation of Model Context Protocol (MCP) clients that spawn local subprocesses via stdio transport. Examples provided use npx to run trusted packages like @modelcontextprotocol/server-filesystem for filesystem access.
- [INDIRECT_PROMPT_INJECTION]: The skill illustrates the processing of untrusted user input within API handlers (e.g., in cloudflare-worker-integration.ts and nextjs-server-action.ts) which is then passed to large language models. The documentation proactively provides security guidance to mitigate these risks, recommending static tool generation over dynamic MCP tools and demonstrating prompt sanitization techniques in the production patterns reference.
- [COMMAND_EXECUTION]: The documentation references CLI tools for project migration and maintenance, such as npx ai migrate for automated version updates and npx mcp-to-ai-sdk for generating static tool definitions from MCP servers.
Audit Metadata