Purpose

This Skill is the source of truth for implementing AI-related logic with Vercel AI SDK v6 in production codebases (TypeScript/Next.js).

When to Use (Trigger Conditions)

Use this Skill when the task involves any of the following:

• •generateText or streamText
• •Streaming endpoints / stream protocol handling
• •Tool calling (function tools) and schema validation
• •Structured output generation (schema-constrained JSON)
• •Next.js Route Handlers (App Router) for AI endpoints
• •Chat UI integration (e.g., tool usage, message handling)
• •Refactoring older AI SDK code to v6 patterns
• •Building AI Agents with ToolLoopAgent class
• •Agent loop control (stepCountIs, custom stop conditions)
• •Multi-step tool orchestration workflows
• •createAgentUIStreamResponse for agent API routes
• •Type-safe agent definitions (InferAgentUIMessage)

Non-Negotiable Rules

1. •Use only APIs/options shown in REFERENCE.md / snippets/. If not present, do not guess.
2. •If the user appears to be on v5 or earlier, call out version mismatch and propose v6 migration.
3. •Prefer type-safe schemas for tools and structured output.
4. •For side-effect tools (DB writes, payments, emails), require explicit approval step in the flow.
5. •Always consider runtime constraints: Node vs Edge.

Navigation

• •Core reference (APIs & signatures): REFERENCE.md
• •Best practices & architecture decisions: PLAYBOOK.md
• •Pre-ship checklist: CHECKLIST.md
• •Copy/paste templates: snippets/
• •Curated official links: resources/links.md

“Good Output” Definition

• •Correct v6 imports and usage
• •Clear separation of concerns: prompt -> tools -> execution -> output
• •Deterministic structured outputs when the result is consumed by code
• •Robust error handling, timeouts, and safe defaults