How to make post_model_hook in create_react_agent check if system is going to END and enforce gaurdrails.

[nsgoneape]

async def clear_agent(state: CustomAgentState):
    """
    Post‑model hook:
    1.  keep the original “safe‑name” overwrite
    2.  emit the final prompt to the logger
    3.  ***NEW*** – if we are about to route to END **while**
        `state.from_agent.not_replied_yet` is still True,
        inject a SystemMessage that reminds the model to
        call `reply_to_agent` and loop back to the agent node
        (`goto="agent"`).
    """

    # ─────────── ORIGINAL LOGIC ───────────
    this_agent = state.get("this_agent", {}) or {}
    raw_name   = this_agent.get("agent_name", "")
    raw_id     = this_agent.get("agent_id", "")
    safe_name  = f"{''.join(c for c in raw_name if c.isalnum())}_{raw_id.replace('-', '')}" if raw_name and raw_id else "UnnamedAgent"

    messages = state.get("messages", [])
    if messages and isinstance(messages[-1], AIMessage):
        messages[-1].name = safe_name

    prompt = await prompt_for_agent(state)
    logger.info(f"[Dario Agent] Final Prompt Used:\n{prompt}")

    # ─────────── NEW REPLY‑ENFORCEMENT ───────────
    from_agent = state.get("from_agent") or {}
		must_reply = from_agent.get("not_replied_yet", False)

    # “About to finish”  ⇢ last LLM turn has **no** tool calls
    last_msg   = messages[-1] if messages else None
    finishing  = isinstance(last_msg, AIMessage) and not getattr(last_msg, "tool_calls", None)

    if must_reply and finishing:
        reminder = SystemMessage(
            content=(
                "⚠️  You are the **receiver** in an agent→agent thread and "
                "must call the `reply_to_agent` tool **before ending**."
            )
        )
        # Tell LangGraph to loop back to the agent node
        return Command(
            update={"messages": [reminder]},
            goto="agent"            # ← documented pattern:contentReference[oaicite:0]{index=0}
        )

    # If everything is fine, just return an empty dict (no state mutation)
    return {}

I’m implementing agent-to-agent messaging using a from_agent field in my LangGraph state, which includes a not_replied_yet: bool flag. If another agent has messaged the current agent, this flag is set to True until the reply_to_agent tool is used. I want to enforce that the graph cannot terminate while not_replied_yet is still True. Right now, I’m using the post_model_hook to check that flag and raise a ValueError if it's still True, which effectively blocks completion. Is this the recommended pattern for enforcing final-state invariants like this in LangGraph, or is there a more idiomatic way to prevent termination until certain tools/actions have been executed?