hermes-agent/tests/agent/test_turn_finalizer_final_response_persistence.py

from types import SimpleNamespace
from typing import Any

from agent.turn_finalizer import finalize_turn


class FakeAgent:
    def __init__(self):
        self.max_iterations = 90
        self.iteration_budget = SimpleNamespace(remaining=10, used=1, max_total=90)
        self.quiet_mode = True
        self.model = "test-model"
        self.provider = "test-provider"
        self.base_url = ""
        self.session_id = "sess-test"
        self.context_compressor = SimpleNamespace(last_prompt_tokens=0)
        self.session_input_tokens = 0
        self.session_output_tokens = 0
        self.session_cache_read_tokens = 0
        self.session_cache_write_tokens = 0
        self.session_reasoning_tokens = 0
        self.session_prompt_tokens = 0
        self.session_completion_tokens = 0
        self.session_total_tokens = 0
        self.session_estimated_cost_usd = 0
        self.session_cost_status = "unknown"
        self.session_cost_source = "test"
        self._tool_guardrail_halt_decision = None
        self._interrupt_message = None
        self._response_was_previewed = True
        self._skill_nudge_interval = 0
        self._iters_since_skill = 0
        self.valid_tool_names = []
        self.persisted_messages: list[dict[str, Any]] | None = None
        self._persist_user_message_idx: int | None = None
        self._persist_user_message_override: Any = None
        self._persist_user_message_timestamp: float | None = None

    def _handle_max_iterations(self, messages, api_call_count):
        raise AssertionError("not expected")

    def _emit_status(self, *_args, **_kwargs):
        pass

    def _safe_print(self, *_args, **_kwargs):
        pass

    def _save_trajectory(self, *_args, **_kwargs):
        pass

    def _cleanup_task_resources(self, *_args, **_kwargs):
        pass

    def _drop_trailing_empty_response_scaffolding(self, messages):
        pass

    def _persist_session(self, messages, conversation_history):
        # Capture the durable write before finalization restores API-local
        # guidance to the returned/live transcript.
        self.persisted_messages = [dict(message) for message in messages]

    def _apply_persist_user_message_override(self, messages):
        idx = self._persist_user_message_idx
        override = self._persist_user_message_override
        if idx is not None and override is not None:
            messages[idx]["content"] = override

    def _file_mutation_verifier_enabled(self):
        return False

    def _turn_completion_explainer_enabled(self):
        return False

    def _drain_pending_steer(self):
        return None

    def clear_interrupt(self):
        pass

    def _sync_external_memory_for_turn(self, **_kwargs):
        pass


def test_finalizer_restores_clean_api_local_text_before_return(monkeypatch):
    """One-shot CLI notes do not replay through same-process history."""
    monkeypatch.setattr("hermes_cli.plugins.invoke_hook", lambda *_a, **_kw: [])
    agent = FakeAgent()
    messages = [
        {"role": "user", "content": "[MODEL SWITCH NOTE]\n\nclean prompt"},
        {"role": "assistant", "content": "Done."},
    ]
    agent._persist_user_message_idx = 0
    agent._persist_user_message_override = "clean prompt"
    agent._persist_user_message_timestamp = None

    result = finalize_turn(
        agent,
        final_response="Done.",
        api_call_count=1,
        interrupted=False,
        failed=False,
        messages=messages,
        conversation_history=[],
        effective_task_id="task",
        turn_id="turn",
        user_message="[MODEL SWITCH NOTE]\n\nclean prompt",
        original_user_message="clean prompt",
        _should_review_memory=False,
        _turn_exit_reason="text_response(finish_reason=stop)",
    )

    assert agent.persisted_messages is not None
    assert agent.persisted_messages[0]["content"] == "clean prompt"
    assert result["messages"][0]["content"] == "clean prompt"


def test_finalizer_restores_clean_api_local_multimodal_before_return(monkeypatch):
    """A queued note does not remain in the next-turn native image payload."""
    monkeypatch.setattr("hermes_cli.plugins.invoke_hook", lambda *_a, **_kw: [])
    agent = FakeAgent()
    clean_content = [
        {"type": "text", "text": "Describe the image"},
        {"type": "image_url", "image_url": {"url": "data:image/png;base64,AAAA"}},
    ]
    api_content = [
        {"type": "text", "text": "[MODEL SWITCH NOTE]\n\nDescribe the image"},
        {"type": "image_url", "image_url": {"url": "data:image/png;base64,AAAA"}},
    ]
    messages = [
        {"role": "user", "content": api_content},
        {"role": "assistant", "content": "Done."},
    ]
    agent._persist_user_message_idx = 0
    agent._persist_user_message_override = clean_content
    agent._persist_user_message_timestamp = None

    result = finalize_turn(
        agent,
        final_response="Done.",
        api_call_count=1,
        interrupted=False,
        failed=False,
        messages=messages,
        conversation_history=[],
        effective_task_id="task",
        turn_id="turn",
        user_message=api_content,
        original_user_message=clean_content,
        _should_review_memory=False,
        _turn_exit_reason="text_response(finish_reason=stop)",
    )

    assert agent.persisted_messages is not None
    assert agent.persisted_messages[0]["content"] == clean_content
    assert result["messages"][0]["content"] == clean_content


def test_final_response_closes_tool_tail_before_persistence(monkeypatch):
    """A recovered/previewed final response must be durable in session history.

    Regression for turns where the caller receives a non-empty final_response,
    but the message transcript still ends at a tool result. If persisted that
    way, the next turn reloads a stale/malformed history and can appear to loop
    because the assistant's visible final answer is missing from durable state.
    """
    monkeypatch.setattr("hermes_cli.plugins.invoke_hook", lambda *_a, **_kw: [])
    agent = FakeAgent()
    messages = [
        {"role": "user", "content": "do it"},
        {
            "role": "assistant",
            "content": "I'll check.",
            "tool_calls": [
                {"id": "call-1", "function": {"name": "terminal", "arguments": "{}"}}
            ],
        },
        {"role": "tool", "tool_call_id": "call-1", "name": "terminal", "content": "ok"},
    ]

    result = finalize_turn(
        agent,
        final_response="Done.",
        api_call_count=2,
        interrupted=False,
        failed=False,
        messages=messages,
        conversation_history=[],
        effective_task_id="task",
        turn_id="turn",
        user_message="do it",
        original_user_message="do it",
        _should_review_memory=False,
        _turn_exit_reason="fallback_prior_turn_content",
    )

    assert result["messages"][-1] == {"role": "assistant", "content": "Done."}
    assert agent.persisted_messages is not None
    assert agent.persisted_messages[-1] == {"role": "assistant", "content": "Done."}