aisdkagents

AI SDK Agents

You are helping build production-grade AI SDK agents. Use this skill when a user asks about agent architecture, streaming chat, tool calling, human approval, structured outputs, artifact generation, evaluator-optimizer loops, orchestration, or choosing between AI SDK APIs.

Triage Protocol

1. Identify the user's implementation shape before reading long references: routing, fixed chain, dynamic orchestration, sub-agent delegation, human approval, human inquiry, artifact generation, structured output, or advanced tool API.
2. Read references/pattern-decision-tree.md when choosing between patterns.
3. Read references/implementation-protocol.md before writing or reviewing code. AI SDK APIs change quickly; do not trust model memory or stale excerpts.
4. Read the matching references/patterns/* file for architecture, tradeoffs, and embedded evidence.
5. Read references/current-ai-sdk-grounding.md and references/common-errors.md before writing or fixing AI SDK APIs.
6. Use references/current-skeletons/* for copy-paste starting points. Treat embedded source excerpts as evidence, not guaranteed current API syntax.
7. Check references/anti-patterns.md before recommending complex orchestration, sub-agents, human approval, artifacts, or filesystem tools.

How To Use This Skill

Start with the decision and grounding references:

• references/implementation-protocol.md for the verification workflow before writing code.
• references/pattern-decision-tree.md to choose the pattern.
• references/current-ai-sdk-grounding.md for current API usage and naming.
• references/common-errors.md when typecheck fails or an excerpt uses stale API names.
• references/framework-integration.md before wiring agents into React, Next.js, or route handlers.
• references/composition-recipes.md when combining multiple patterns.
• references/anti-patterns.md to avoid overbuilding or stale API choices.
• references/current-skeletons/ for minimal current implementation templates.

Then read the focused pattern references. They are self-contained playbooks with architecture guidance and embedded source excerpts:

• references/patterns/routing-agent-pattern.md for routing agent pattern.
• references/patterns/multi-step-tool-agent-pattern.md for multi-step tool agent pattern.
• references/patterns/orchestrator-worker-pattern.md for orchestrator worker pattern.
• references/patterns/sub-agent-orchestrator-pattern.md for sub-agent orchestrator pattern.
• references/patterns/sequential-agent-chain-pattern.md for sequential agent chain pattern.
• references/patterns/evaluator-optimizer-pattern.md for evaluator optimizer pattern.
• references/patterns/human-approval-pattern.md for human approval pattern.
• references/patterns/inquire-human-input-pattern.md for inquire human input pattern.
• references/patterns/plan-and-implement-agent-pattern.md for plan and implement agent pattern.
• references/patterns/artifact-agent-pattern.md for artifact agent pattern.
• references/patterns/skills-on-disk-agent-pattern.md for skills on disk agent pattern.
• references/patterns/advanced-tool-api-patterns.md for advanced tool api patterns.
• references/patterns/structured-agent-output-pattern.md for structured agent output pattern.

Use the broader references when you need a catalog view across the whole library:

• references/agent-architecture-patterns.md for routing, orchestration, sub-agents, parallel processing, and domain agents.
• references/tool-calling-patterns.md for tool definitions, multi-step loops, repair, approval, and typed tool UI.
• references/human-in-the-loop-patterns.md for confirmation, inquire, and approval workflows.
• references/artifact-patterns.md for chat-plus-artifact systems and streamed durable outputs.
• references/streaming-ui-patterns.md for data parts, transport setup, status states, and custom UI messages.
• references/evaluator-optimizer-patterns.md for critique/revise loops.
• references/ai-sdk-api-cookbook.md for choosing streamText, generateObject, streamObject, tool, ToolLoopAgent, and related APIs.
• references/provider-env-matrix.md for providers, external services, environment variables, and setup complexity.

When answering, prefer reusable architecture and code-shape guidance over telling users to install a specific pro block. Mention source blocks only as examples or proven references.