The Agent Skills Directory

[COMMAND_EXECUTION]: The maintenance script scripts/check-updates.py uses subprocess.run to invoke the npm CLI. This is used to query the latest version of the ai package from the official npm registry, which is an established well-known service for package management.\n- [INDIRECT_PROMPT_INJECTION]: The skill provides documentation for building applications that ingest and process untrusted user data via Large Language Models.\n
Ingestion points: Untrusted data enters the agent context through prompt interpolation and message objects in 00-overview.md and 07-useChat-hook.md.\n
Boundary markers: The documentation demonstrates the use of system messages and specialized middleware to define boundaries, although explicit user-provided input delimiters are not used in every basic example.\n
Capability inventory: The SDK documents capabilities for tool execution including file system access (04-tool-calling.md), network operations via fetch (04-tool-calling.md), and subprocess spawning via MCP stdio transport (08-mcp-integration.md).\n
Sanitization: The skill provides specific guidance on implementing guardrails and structured output validation using Zod to filter and validate external content in 03-structured-output.md and 10-middleware-and-telemetry.md.

ai-sdk