parse_resume_command

Build a resume :class:Command from request input items, if present.

Two input shapes are accepted, both keyed by interrupt.id:

• :class:FunctionCallOutputItemParam matched by call_id. Its output field is decoded — a JSON object with any of {"resume", "update", "goto"} populates the :class:Command; anything else (string, malformed JSON, list of content parts) is treated as the raw resume value.
• :class:MCPApprovalResponse matched by approval_request_id. For a recognized LangChain HumanInTheLoopMiddleware request, its boolean choice becomes one ordered decision per action. Otherwise, approve=True retains the original interrupt-value echo behavior and approve=False is surfaced by :func:validate_approval_responses. Recognition uses the complete action_requests / review_configs structure, which is reserved for this fast path. Use function_call_output to preserve an identically shaped custom value.

Conflict resolution: when both shapes target the same interrupt id in one request, the function_call_output wins (richer payload) and a warning is logged. This is a deliberate, deterministic departure from Agent Framework's order-dependent last-write-wins. Callers must run :func:validate_approval_responses first; it rejects conflicting MCP approval responses before this function builds a command.

Resume shape: with a single pending interrupt the resume value is passed through as-is (Command(resume=value)). With parallel pending interrupts LangGraph requires an id-keyed resume map — a bare value raises RuntimeError: When there are multiple pending interrupts, you must specify the interrupt id when resuming. — so every matched item that explicitly carries a resume value is folded into Command(resume={id: value}). Answering only some of the pending interrupts is allowed; the unanswered ones stay paused.