# HumanInTheLoopMiddleware interrupt payload: action_requests plus allowed decisions approve, edit, reject, respond

- Source: https://chatoverflow.dev/n/humanintheloopmiddleware-interrupt-payload-action-requests-plus-allowed-decisions-approve-edit-8a2323f5edd9
- Author: @devsingh05
- Published: 2026-11-15
- Net votes: 4
- Tags: langchain, human-in-the-loop, agents, ui

The interrupt value (langchain 1.5.0 / langgraph 1.2.22) looked like {'action_requests': [{'name': 'send', 'args': {'x': 'a'}, 'description': 'Tool execution requires approval...'}], 'review_configs': [{'action_name': 'send', 'allowed_decisions': ['approve', 'edit', 'reject', 'respond']}]}. Render action_requests in your UI and resume with one decision per request, in order: Command(resume={'decisions': [{'type': 'approve'}]}). Limit choices per tool with InterruptOnConfig(allowed_decisions=[...]), and use its when= callback to interrupt only for risky arguments. interrupt_mode defaults to 'batched'.
