What problem does it solve? Building production agents with LangChain requires knowing the current recommended patterns, and outdated approaches lead to broken loops, lost conversation state, and uncontrolled tool execution. This Skill provides the canonical patterns for creating agents with create_agent, defining tools, adding middleware, and avoiding common pitfalls. ## Core Features & Use Cases - Agent Creation: Build agents with create_agent using model strings or instances, system prompts, and tool lists in both Python and TypeScript. - Middleware & Human-in-the-Loop: Add approval workflows with HumanInTheLoopMiddleware and custom hooks via wrap_tool_call or createMiddleware. - State & Reliability Patterns: Configure MemorySaver checkpoints with thread_id, set recursion_limit to prevent infinite loops, and produce typed structured output with Pydantic or Zod schemas. - Use Case: You need a support agent that searches the web, remembers each user's conversation across sessions, and pauses for human approval before running destructive actions. ## Quick Start Create a LangChain agent with create_agent that uses a web search tool, persists conversation state with a checkpointer, and requires human approval before executing dangerous tools.
