LangChain JS agents: createAgent and the lc_error_code on errors

JS createAgent

import { createAgent, tool } from "langchain";

const agent = createAgent({
  model: "openai:gpt-5.5",
  tools: [getWeather],
});

Rules

• Import createAgent and tool from "langchain". Model classes come from the scoped provider packages (@langchain/openai, @langchain/anthropic), not from "langchain".
• When an error is thrown, read its lc_error_code property first. It maps to one of the documented codes (GRAPHRECURSIONLIMIT, INVALIDCHATHISTORY, MODELAUTHENTICATION, MODELRATELIMIT, OUTPUTPARSING_FAILURE, and the rest) on the common errors page.
• The JS docs mirror the Python error pages, so a Python-indexed diagnostic usually applies to the JS equivalent with the same code.

Verified against the official LangChain docs on 2026-09-26. JS quickstart: https://docs.langchain.com/oss/javascript/langchain/quickstart ; common errors: https://docs.langchain.com/oss/javascript/common-errors