TraderAI/tests/test_agent.py

import pytest
import asyncio
import itertools

from traderai.agent import OllamaAgent, SYSTEM_PROMPT
from traderai.memory import MemoryStore


class EmptyTools:
    schemas = []

    @property
    def pending_actions(self):
        return {}


class WakeTools(EmptyTools):
    def __init__(self):
        self.calls = []

    async def execute(self, name, arguments):
        self.calls.append((name, arguments))
        return {"count": 1, "notifications": [{"message": "Buyer replied"}]}


class WakeAgent(OllamaAgent):
    def __init__(self, memory):
        super().__init__("http://127.0.0.1:1", "missing-model", WakeTools(), memory=memory)
        self.responses = [
            {
                "message": {
                    "role": "assistant",
                    "content": "",
                    "tool_calls": [
                        {
                            "function": {
                                "name": "check_uex_notifications",
                                "arguments": {},
                            }
                        }
                    ],
                }
            },
            {"message": {"role": "assistant", "content": "I checked notifications: Buyer replied."}},
        ]

    async def ensure_available(self):
        return None

    async def _ollama_chat(self, *args, **kwargs):
        return self.responses.pop(0)


class TitleAgent(OllamaAgent):
    def __init__(self, memory):
        super().__init__("http://127.0.0.1:1", "missing-model", EmptyTools(), memory=memory)

    async def ensure_available(self):
        return None

    async def _generate_chat_title(self, first_message):
        return "UEX Market Check"

    async def _ollama_chat(self, *args, **kwargs):
        return {"message": {"role": "assistant", "content": "Done"}}


class ImageCaptureAgent(OllamaAgent):
    def __init__(self, memory):
        super().__init__("http://127.0.0.1:1", "missing-model", EmptyTools(), memory=memory)
        self.last_messages = None

    async def ensure_available(self):
        return None

    async def _chat_once(self, query="", messages=None, **kwargs):
        self.last_messages = messages
        return {"message": {"role": "assistant", "content": "Seen"}}


class SlowToolTools(EmptyTools):
    schemas = [
        {
            "type": "function",
            "function": {
                "name": "slow_tool",
                "description": "Slow fake tool.",
                "parameters": {"type": "object", "properties": {}},
            },
        }
    ]

    def __init__(self):
        self.calls = 0

    async def execute(self, name, arguments):
        self.calls += 1
        await asyncio.sleep(0.01)
        return {"status": "ok", "value": "slow result"}


class SlowStreamingAgent(OllamaAgent):
    def __init__(self, memory):
        super().__init__("http://127.0.0.1:1", "missing-model", SlowToolTools(), memory=memory)
        self.stream_calls = 0

    async def health(self):
        return {"online": True, "model": "test", "base_url": self.base_url}

    async def _ollama_chat_stream(self, *args, **kwargs):
        self.stream_calls += 1
        if self.stream_calls == 1:
            yield {
                "message": {
                    "role": "assistant",
                    "content": "",
                    "tool_calls": [{"function": {"name": "slow_tool", "arguments": {}}}],
                },
                "done": True,
            }
            return
        yield {"message": {"role": "assistant", "content": ""}, "done": True}


class FailingAfterToolAgent(SlowStreamingAgent):
    async def _ollama_chat_stream(self, *args, **kwargs):
        self.stream_calls += 1
        if self.stream_calls == 1:
            yield {
                "message": {
                    "role": "assistant",
                    "content": "",
                    "tool_calls": [{"function": {"name": "slow_tool", "arguments": {}}}],
                },
                "done": True,
            }
            return
        raise RuntimeError("ollama timed out")


@pytest.mark.asyncio
async def test_chat_events_warns_when_ollama_offline():
    agent = OllamaAgent("http://127.0.0.1:1", "missing-model", EmptyTools())
    events = []
    async for event in agent.chat_events("hello"):
        events.append(event)

    assert events[0]["type"] == "warning"
    assert "Ollama is offline" in events[0]["message"]
    assert events[-1]["type"] == "done"


def test_runtime_context_uses_previous_interaction_not_current_message(tmp_path):
    memory = MemoryStore(str(tmp_path / "memory.sqlite3"))
    memory.add_conversation("assistant", "older answer")
    previous = memory.last_interaction()
    assert previous is not None

    memory.add_conversation("user", "current question")
    current = memory.last_interaction()

    agent = OllamaAgent("http://127.0.0.1:1", "missing-model", EmptyTools(), memory=memory)
    context = agent._runtime_context("current question", previous_interaction=previous)

    assert f"Previous interaction before this message: {previous['created_at']}" in context
    assert f"Previous interaction before this message: {current['created_at']}" not in context
    assert "Current local date/time:" in context


def test_runtime_context_includes_uex_user_identity(tmp_path):
    memory = MemoryStore(str(tmp_path / "memory.sqlite3"))
    memory.set_profile(
        "uex_user",
        {
            "username": "pilot_hudson",
            "name": "Hudson",
            "email": "hudson@example.test",
            "timezone": "America/New_York",
            "specializations": "trading,hauling",
        },
    )

    agent = OllamaAgent("http://127.0.0.1:1", "missing-model", EmptyTools(), memory=memory)
    context = agent._runtime_context("")

    assert "You are speaking with UEX user pilot_hudson (Hudson)." in context
    assert "timezone: America/New_York" in context
    assert "specializations: trading,hauling" in context
    assert "hudson@example.test" not in context


def test_stream_metrics_include_reading_and_writing_rates():
    metrics = OllamaAgent._stream_metrics(
        {
            "prompt_eval_count": 20,
            "prompt_eval_duration": 2_000_000_000,
            "eval_count": 30,
            "eval_duration": 3_000_000_000,
        }
    )

    assert metrics["reading_tokens"] == 20
    assert metrics["reading_tokens_per_second"] == 10
    assert metrics["writing_tokens"] == 30
    assert metrics["writing_tokens_per_second"] == 10


def test_system_prompt_prefers_current_marketplace_data():
    assert "open/current" in SYSTEM_PROMPT
    assert "Do not use historical sale data" in SYSTEM_PROMPT
    assert "aUEC/UEC credits" in SYSTEM_PROMPT
    assert "never real-world dollars" in SYSTEM_PROMPT


def test_ollama_options_include_num_ctx():
    agent = OllamaAgent("http://127.0.0.1:1", "missing-model", EmptyTools(), num_ctx=64000)

    assert agent._ollama_options() == {"num_ctx": 64000}


def test_deepseek_tool_rounds_are_not_capped_at_ten():
    agent = OllamaAgent("https://api.deepseek.com", "deepseek-v4-flash", EmptyTools(), provider="deepseek", api_key="test")

    rounds = list(itertools.islice(agent._tool_rounds(), 12))

    assert len(rounds) == 12


def test_plan_draft_normalization_extracts_json_and_defaults():
    seed = {"title": "Wikelo Polaris", "objective": "Find parts", "kind": "buying", "constraints": {}, "items": []}
    raw = 'draft:\n{"title":"Wikelo Polaris Parts","objective":"Find and draft deals for the parts below","kind":"buying","cadence":"0 */3 * * *","constraints":{"message_tone":"casual","instructions":"Prioritize cheap listings first."},"items":[{"item_name":"RCMBNT-RGL-1","desired_quantity":2}]}'

    draft = OllamaAgent._normalize_plan_draft(raw, seed)

    assert draft["title"] == "Wikelo Polaris Parts"
    assert draft["cadence"] == "0 */3 * * *"
    assert draft["constraints"]["message_tone"] == "casual"
    assert draft["items"][0]["item_name"] == "RCMBNT-RGL-1"
    assert draft["items"][0]["desired_quantity"] == 2


def test_plan_draft_heuristic_fills_in_basic_instructions():
    seed = {"title": "Watch open negotiations", "objective": "", "kind": "custom", "constraints": {}, "items": []}

    draft = OllamaAgent._heuristic_plan_draft(seed)

    assert draft["kind"] == "custom"
    assert draft["cadence"] == "0 */4 * * *"
    assert "summarize" in draft["constraints"]["instructions"].casefold()
    assert draft["constraints"]["message_tone"] == "friendly and direct"


def test_codex_prompt_mentions_tools_and_images(tmp_path):
    memory = MemoryStore(str(tmp_path / "memory.sqlite3"))
    agent = OllamaAgent("codex", "gpt-5.3-codex", EmptyTools(), memory=memory, provider="codex")

    prompt = agent._codex_cli_prompt(
        "check listing",
        [
            {"role": "system", "content": SYSTEM_PROMPT},
            {
                "role": "user",
                "content": "Look at this",
                "images": ["ZmFrZQ=="],
                "image_content_types": ["image/png"],
            },
            {
                "role": "assistant",
                "content": "",
                "tool_calls": [
                    {
                        "id": "call_123",
                        "type": "function",
                        "function": {"name": "search_marketplace_listings", "arguments": "{\"commodity\":\"gold\"}"},
                    }
                ],
            },
            {
                "role": "tool",
                "tool_name": "search_marketplace_listings",
                "tool_call_id": "call_123",
                "content": "{\"ok\":true}",
            },
        ],
    )

    assert "Available tools" in prompt
    assert "attached images: 1" in prompt
    assert "search_marketplace_listings" in prompt
    assert "tool search_marketplace_listings" in prompt


def test_deepseek_openai_messages_include_reasoning_content_for_tool_turns():
    agent = OllamaAgent("https://api.deepseek.com", "deepseek-v4-flash", EmptyTools(), provider="deepseek", api_key="test")

    messages = agent._openai_messages(
        "check listing",
        [
            {"role": "system", "content": SYSTEM_PROMPT},
            {"role": "user", "content": "Check this listing"},
            {
                "role": "assistant",
                "content": "",
                "reasoning_content": "I should check the current listing first.",
                "tool_calls": [
                    {
                        "id": "call_123",
                        "type": "function",
                        "function": {"name": "search_marketplace_listings", "arguments": "{\"query\":\"panel\"}"},
                    }
                ],
            },
            {"role": "tool", "tool_name": "search_marketplace_listings", "tool_call_id": "call_123", "content": "{\"ok\":true}"},
        ],
    )

    assistant_turn = next(message for message in messages if message["role"] == "assistant")
    assert assistant_turn["reasoning_content"] == "I should check the current listing first."


def test_codex_structured_response_extracts_text_and_tool_calls():
    agent = OllamaAgent("codex", "gpt-5.3-codex", EmptyTools(), provider="codex")

    result = agent._codex_structured_response(
        {
            "kind": "tool_call",
            "message": "",
            "tool_name": "search_marketplace_listings",
            "arguments_json": "{\"commodity\":\"gold\"}",
        }
    )

    assert result["message"]["content"] == ""
    assert result["message"]["tool_calls"] == [
        {
            "id": result["message"]["tool_calls"][0]["id"],
            "type": "function",
            "function": {
                "name": "search_marketplace_listings",
                "arguments": "{\"commodity\":\"gold\"}",
            },
        }
    ]


def test_parse_codex_exec_output_reads_final_json():
    agent = OllamaAgent("codex", "gpt-5.3-codex", EmptyTools(), provider="codex")

    result = agent._parse_codex_exec_output(
        {
            "returncode": 0,
            "stdout": "",
            "stderr": "",
            "events": [
                {"type": "thread.started", "thread_id": "abc"},
                {"type": "item.completed", "item": {"type": "agent_message", "text": "{\"kind\":\"final\",\"message\":\"hello\",\"tool_name\":\"\",\"arguments_json\":\"{}\"}"}},
                {"type": "turn.completed"},
            ],
        }
    )

    assert result == {"kind": "final", "message": "hello", "tool_name": "", "arguments_json": "{}"}


@pytest.mark.asyncio
async def test_wake_response_executes_tool_calls(tmp_path):
    memory = MemoryStore(str(tmp_path / "memory.sqlite3"))
    agent = WakeAgent(memory)

    response = await agent.generate_wake_response("Scheduled wake job fired. Check notifications.")

    assert response == "I checked notifications: Buyer replied."
    assert agent.tools.calls == [("check_uex_notifications", {})]
    wake_rows = memory.recent_conversation(thread_id="wake")
    assert wake_rows[-1]["content"] == "I checked notifications: Buyer replied."


@pytest.mark.asyncio
async def test_first_chat_message_generates_thread_title(tmp_path):
    memory = MemoryStore(str(tmp_path / "memory.sqlite3"))
    thread = memory.create_thread()
    agent = TitleAgent(memory)

    result = await agent.chat("Check UEX market listings", thread_id=thread["id"])

    assert result["message"] == "Done"
    assert memory.get_thread(thread["id"])["title"] == "UEX Market Check"


@pytest.mark.asyncio
async def test_chat_includes_pasted_images_and_memory_note(tmp_path):
    memory = MemoryStore(str(tmp_path / "memory.sqlite3"))
    agent = ImageCaptureAgent(memory)

    result = await agent.chat(
        "",
        images=[{"name": "listing.png", "content_type": "image/png", "image_data": "ZmFrZS1pbWFnZQ=="}],
    )

    assert result["message"] == "Seen"
    user_message = next(message for message in reversed(agent.last_messages) if message.get("role") == "user")
    assert user_message["images"] == ["ZmFrZS1pbWFnZQ=="]
    assert user_message["content"] == "Please analyze the attached image."
    assert "[Attached 1 pasted image]" in memory.recent_conversation()[-2]["content"]


@pytest.mark.asyncio
async def test_chat_events_returns_fallback_after_slow_tool_and_empty_final_response(tmp_path):
    memory = MemoryStore(str(tmp_path / "memory.sqlite3"))
    agent = SlowStreamingAgent(memory)

    events = [event async for event in agent.chat_events("run a slow tool")]
    text = "".join(event.get("content", "") for event in events if event["type"] == "token")

    assert agent.tools.calls == 1
    assert "I completed the tool call" in text
    assert "slow result" in text
    assert events[-1]["type"] == "done"


@pytest.mark.asyncio
async def test_chat_events_returns_tool_result_when_model_fails_after_slow_tool(tmp_path):
    memory = MemoryStore(str(tmp_path / "memory.sqlite3"))
    agent = FailingAfterToolAgent(memory)

    events = [event async for event in agent.chat_events("run a slow tool")]
    text = "".join(event.get("content", "") for event in events if event["type"] == "token")

    assert agent.tools.calls == 1
    assert "local model stopped after the tool call" in text
    assert "slow result" in text
    assert events[-1]["type"] == "done"