feat: add response_schema to interrupt and Interrupt

[YassinNouh21]

close #5027

Summary

Adds optional response_schema parameter to interrupt() function and Interrupt dataclass to enable structured validation of resume values in human-in-the-loop workflows.

Changes

• Core functionality: Added response_schema: dict | None field to Interrupt dataclass
• API enhancement: Modified interrupt(value, *, response_schema=None) to accept Pydantic model classes
• Validation: Resume values are validated against the schema and converted to model instances
• Error handling: Invalid data triggers new interrupts with detailed validation error messages

Technical Details

• Stores JSON schema dictionaries (not Pydantic classes) for proper serialization
• Compatible with all checkpointers and graph state persistence
• Maintains backward compatibility - existing code works unchanged
• Validation errors include detailed field-level feedback

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
langgraph-docs-preview	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jun 10, 2025 6:32am

[YassinNouh21]

@sydney-runkle Can u take a look ?

[sydney-runkle]

Thanks for this - marked as deferred, want to clean up Interrupt first for V1, then we can add this new feature.

[sydney-runkle]

Closing as stale for now. We're doing a big push for V1, but we do want this feature eventually!!

Happy to sync back up in a few weeks to sprint on this :)

[YassinNouh21]

Closing as stale for now. We're doing a big push for V1, but we do want this feature eventually!!

Happy to sync back up in a few weeks to sprint on this :)

That would be nice when are u opening back the contributions?