LangGraph 1.2.12 lets interrupt() take a response_schema, so the human's resume value is validated before the node commits

GitHub·medium signalXBlueskyLinkedInCopy link

PR #8886, shipped in langgraph 1.2.12 on Sept 21, adds an optional keyword-only response_schema to interrupt(). It accepts a JSON Schema dict, a Pydantic model, a TypedDict or a dataclass. Pydantic-style types are validated on resume, and an invalid value raises from the node before it is committed. LangGraph Studio uses the schema to render a typed form instead of a free-form JSON box. For human-in-the-loop graphs this closes the gap where a malformed approval payload flowed straight into downstream state. The default is None, so existing graphs are unaffected.