Skip to content

DeepTutor: Lifelong Personalized Tutoring. https://deeptutor.info/.

7.4 KB
Raw Blame
from __future__ import annotations

from types import SimpleNamespace

import pytest

from deeptutor.agents.chat.agentic_pipeline import AgenticChatPipeline
from deeptutor.agents.loop.prompt_blocks import ChatPromptAssembler
from deeptutor.capabilities.mastery.pipeline import MasteryLoopPipeline
from deeptutor.core.context import TurnRuntimeContext, UnifiedContext, WorkspaceRuntimeContext
from deeptutor.runtime.agentic.tool_dispatch import MAX_PARALLEL_TOOL_CALLS


@pytest.fixture(autouse=True)
def _fake_llm_config(monkeypatch: pytest.MonkeyPatch) -> None:
    cfg = SimpleNamespace(
        binding="openai",
        model="gpt-test",
        api_key="sk-test",
        base_url="https://example.test/v1",
        api_version=None,
    )
    monkeypatch.setattr(
        "deeptutor.agents.loop.pipeline.get_llm_config",
        lambda: cfg,
    )
    monkeypatch.setattr("deeptutor.agents.base_agent.get_llm_config", lambda: cfg)


@pytest.mark.parametrize("language", ["en", "zh"])
@pytest.mark.parametrize("mode", ["chat", "immersive_reading", "mastery_path"])
def test_tool_call_limit_reaches_each_loop_prompt(language: str, mode: str) -> None:
    pipeline_type = MasteryLoopPipeline if mode == "mastery_path" else AgenticChatPipeline
    pipeline = pipeline_type(language=language)
    context = UnifiedContext(active_capability=mode)
    prompt = pipeline._build_system_prompt([], context)

    assert MAX_PARALLEL_TOOL_CALLS == 15
    assert prompt.count("## tool_call_policy\n") == 1
    assert "{limit}" not in prompt
    if language == "zh":
        assert "最多请求 15 次工具调用" in prompt
        assert "不是用户整个请求的总次数" in prompt
        assert "超出上限的调用不会执行" in prompt
    else:
        assert "at most 15 tool calls" in prompt
        assert "not a total for the user's request" in prompt
        assert "Calls beyond the limit\nare not executed" in prompt


def test_agentic_chat_final_prompt_uses_selected_language(
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    class FakeRegistry:
        def build_prompt_text(self, *_args, **_kwargs) -> str:
            return "- tool"

    monkeypatch.setattr(
        "deeptutor.agents.loop.pipeline.get_tool_registry",
        lambda: FakeRegistry(),
    )

    from deeptutor.core.context import UnifiedContext

    ctx = UnifiedContext()
    zh_prompt = AgenticChatPipeline(language="zh")._build_system_prompt([], ctx)
    en_prompt = AgenticChatPipeline(language="en")._build_system_prompt([], ctx)

    # Prompt blocks are phase-specific, but the shared language directive
    # still runs at the end, so per-language imperatives must surface.
    assert "请严格使用中文" in zh_prompt
    assert "Write ALL reader-facing text" in en_prompt
    # Persona phrasing differs by language so the prompts are not just
    # English text with a Chinese tail appended.
    assert "你是 DeepTutor" in zh_prompt
    assert "You are DeepTutor" in en_prompt


def test_mastery_plugin_system_prompt_uses_localized_fallback(
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    class FakeRegistry:
        def build_prompt_text(self, *_args, **_kwargs) -> str:
            return "- tool"

    monkeypatch.setattr(
        "deeptutor.agents.loop.pipeline.get_tool_registry",
        lambda: FakeRegistry(),
    )

    from deeptutor.core.context import UnifiedContext

    ctx = UnifiedContext(metadata={"mastery_mode": True, "mastery_path_id": "p1"})
    zh_prompt = AgenticChatPipeline(language="zh")._build_system_prompt([], ctx)
    en_prompt = AgenticChatPipeline(language="en")._build_system_prompt([], ctx)

    assert "## mastery_tutor" in zh_prompt
    assert "掌握式导师" in zh_prompt
    assert "## mastery_tutor" in en_prompt
    assert "mastery tutor" in en_prompt


def test_ask_questions_plugin_system_prompt_uses_localized_fallback(
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    class FakeRegistry:
        def build_prompt_text(self, *_args, **_kwargs) -> str:
            return "- tool"

    monkeypatch.setattr(
        "deeptutor.agents.loop.pipeline.get_tool_registry",
        lambda: FakeRegistry(),
    )

    from deeptutor.core.context import UnifiedContext

    ctx = UnifiedContext(metadata={"ask_questions_mode": True})
    zh_prompt = AgenticChatPipeline(language="zh")._build_system_prompt([], ctx)
    en_prompt = AgenticChatPipeline(language="en")._build_system_prompt([], ctx)

    assert "## ask_questions" in zh_prompt
    assert "主动提问模式" in zh_prompt
    assert "必须以一次 `ask_user` 调用开始" in zh_prompt
    assert "第 2、3、10 轮" in zh_prompt
    assert "此前所有对话" in zh_prompt
    assert "## ask_questions" in en_prompt
    assert "Ask Questions mode" in en_prompt
    assert "second, third, tenth" in en_prompt
    assert "calling `ask_user` exactly once" in en_prompt


def test_prompt_blocks_include_localized_optional_context() -> None:
    from deeptutor.core.context import UnifiedContext

    prompts = {
        "general": "通用",
        "runtime_policy": "策略",
        "loop": {
            "system": "循环",
            "user": "用户说:{user_message}",
            "finish_exhausted": "预算已用完,请直接回答。",
        },
    }
    ctx = UnifiedContext(
        user_message="解释光合作用",
        sidebar_context="[选中内容]\n把代码和静态数据加载进内存",
        persona_context="用苏格拉底式提问",
        memory_context="学生喜欢例子",
    )
    assembler = ChatPromptAssembler(prompts=prompts, language="zh")

    blocks = assembler.blocks(context=ctx, tool_manifest="", workspace_note="工作区可用")

    names = [block.name for block in blocks]
    assert names[:4] == ["general", "runtime_context", "runtime_policy", "loop"]
    assert "sidebar_tutor_context" in names
    assert "persona_style" in names
    assert "memory" in names
    assert "workspace" in names
    assert assembler.user_message(context=ctx) == "用户说:解释光合作用"
    assert assembler.finish_exhausted_instruction() == "预算已用完,请直接回答。"
    system_prompt = assembler.render(blocks)
    assert "## sidebar_tutor_context" in system_prompt
    assert "把代码和静态数据加载进内存" in system_prompt


@pytest.mark.parametrize("language", ["en", "zh"])
@pytest.mark.parametrize(
    "mode",
    [
        "chat",
        "deep_solve",
        "ask_questions",
        "immersive_reading",
        "visualize",
        "mastery_path",
    ],
)
def test_workspace_instructions_reach_shared_loop_prompts_and_stay_in_the_system(language, mode):
    pipeline_type = MasteryLoopPipeline if mode == "mastery_path" else AgenticChatPipeline
    pipeline = pipeline_type(language=language)
    text = "先讲直觉。Keep {literal braces} and line breaks.\nThen show examples."
    context = UnifiedContext(
        active_capability=mode,
        runtime=TurnRuntimeContext(workspace=WorkspaceRuntimeContext(instructions=text)),
    )
    prompt = pipeline._build_system_prompt([], context)
    assert prompt.count("## workspace_instructions\n") == 1
    assert text in prompt
    assert "workspace_instructions" not in pipeline._runtime_snapshots
    assert (
        "当前能力必需的工作流程" in prompt
        if language == "zh"
        else "capability's required workflow" in prompt
    )
    context.runtime.workspace = WorkspaceRuntimeContext()
    assert "## workspace_instructions\n" not in pipeline._build_system_prompt([], context)