The Agent Skills Directory

[COMMAND_EXECUTION]: The SDK supports the creation of function tools that execute business logic. These tools are designed to run in isolated sandboxes (Native Node.js child processes or Vercel MicroVMs) with restricted file system access and resource limits to ensure environment security.
[EXTERNAL_DOWNLOADS]: The inkeep add CLI command allows users to fetch project templates and MCP server code from Inkeep's official GitHub repository. These downloads are part of the intended project initialization workflow and target trusted vendor sources.
[INDIRECT_PROMPT_INJECTION]: The SDK provides features for ingesting external data via webhooks and HTTP headers for use in agent prompts, which constitutes a potential injection surface.
Ingestion points: Webhook triggers (rules/overview-triggers-overview.md) and HTTP headers (rules/headers-passing-context.md) provide data that is interpolated into system prompts.
Boundary markers: Data is interpolated using template syntax; while the SDK provides type-safe templating helpers, it does not explicitly enforce the use of XML-style or other delimiters for untrusted strings in the examples.
Capability inventory: Agents built with the SDK can perform file operations via MCP tools or execute sandboxed code via function tools.
Sanitization: Input payloads are validated against Zod schemas for structural correctness, ensuring the agent receives data in the expected format.

typescript-sdk