feat(agent): LLM-powered collection node with welcome stage and session-scoped delete (assignment 4)

- prompts/collection.md: versioned collection system prompt extracted from workflow node yQaBxfrWjbRr with source/version/porting notes - llm.py: LLMService protocol; FakeLLMService (scripted, records calls); OpenAILLMService (async, wait_for timeout with cancellation, retry only on timeout/429/connection, errors normalized) - collect.py: run_collect reference impl - one LLM call produces state prefix + body, controlled repair (max 1 retry) for missing prefix or illegal stage proposal, whitelist guard, CollectReplyError without raw model content; dialogue history + confirmed fields injected into prompt so the model skips known info (作业4-5) - fields.py / clock.py: three-state field candidates (unknown/ candidate/confirmed), per-type validation (count not inferred from description, future time rejected), injectable Clock - graph.py: welcome node - first message returns fixed 1001 greeting without LLM (mirrors workflow is_first_message -> Assigned Reply#3), route empty stage to welcome - endpoints/delete_session: backend switch - langgraph clears graph checkpoint, fastgpt path unchanged - config: AGENT_LLM_PROVIDER fake|openai with validation - untrack .env (secrets), move template to root .env.example Verified: - pytest 103 passed; smoke turns 1/2/1 - Real LLM e2e: POST /chat multi-turn context works - Golden eval collection_skip_known_info: 12/12 cases pass - Golden eval no_false: 8/10 cases pass; 4 confirm/photo cases fail at final step - model routes on short reply '对' (prompt-adherence gap, photo/confirm stages are assignment 5 scope); recorded in evals/results/20261020-144856 Known limits: field extraction from model reply not wired (stub); photo/confirm nodes still scaffold; clarify counting is approximate.
2026-10-20 14:53:46 +08:00
parent 8323d53a70
commit 1c6b1a513d
17 changed files with 1596 additions and 34 deletions
@@ -0,0 +1,189 @@
+"""作业 4 验收测试：LLM 接入采集节点。
+
+分组：
+- 接口与工具（提示词加载、FakeLLM 记录、前缀解析、固定时钟）：
+  现在就应该通过；
+- 关键代码（run_collect / validate_candidate / validate_accident_time）：
+  实现前会失败，属预期状态。
+"""
+
+from datetime import datetime
+
+import pytest
+
+from src.agent.clock import FixedClock, validate_accident_time
+from src.agent.collect import (
+    CollectReplyError,
+    parse_state_reply,
+    run_collect,
+)
+from src.agent.fields import FieldCandidate, FieldStatus, validate_candidate
+from src.agent.graph import build_graph
+from src.agent.llm import FakeLLMService
+from src.agent.prompts import COLLECTION, load_prompt
+
+
+def collect_state(**overrides) -> dict:
+    state = {
+        "session_id": "s1",
+        "input_text": "你好",
+        "stage_code": "1002",
+        "turn_count": 1,
+        "no_reply_count": 0,
+        "clarify_count": 0,
+        "accident_fields": {},
+        "turn_patch": {},
+        "client_event": "",
+    }
+    state.update(overrides)
+    return state
+
+
+# ---------- 接口与工具（现在即应通过） ----------
+
+
+def test_prompt_file_has_source_header():
+    text = load_prompt(COLLECTION)
+    assert "事故信息采集2025118.json" in text
+    assert "nodeId" in text
+    assert "原文开始" in text
+
+
+@pytest.mark.asyncio
+async def test_fake_llm_records_calls():
+    llm = FakeLLMService(lambda s, u: f"<state>1002</state>{u}")
+    out = await llm.complete("sys", "你好")
+    assert out == "<state>1002</state>你好"
+    assert len(llm.calls) == 1
+    assert llm.calls[0]["user_message"] == "你好"
+
+
+def test_parse_state_reply_valid():
+    code, body = parse_state_reply("<state>1002</state>请问有几辆车？")
+    assert code == "1002"
+    assert body == "请问有几辆车？"
+
+
+def test_parse_state_reply_missing_prefix_raises():
+    with pytest.raises(ValueError):
+        parse_state_reply("没有前缀的回复")
+
+
+def test_fixed_clock():
+    fixed = datetime(2026, 10, 19, 12, 0, 0)
+    assert FixedClock(fixed).now() == fixed
+
+
+# ---------- 关键代码（实现前失败，属预期） ----------
+
+
+@pytest.mark.asyncio
+async def test_collect_single_llm_call():
+    # 计划作业 4-3：一次轮次一次模型调用，同时产出前缀与正文
+    llm = FakeLLMService(
+        lambda s, u: "<state>1002</state>请问有几辆汽车卷入了这次事故呢？"
+    )
+    result = await run_collect(collect_state(), llm)
+    assert len(llm.calls) == 1
+    assert result["response_text"].startswith("<state>1002</state>")
+    assert result["turn_count"] == 2
+
+
+@pytest.mark.asyncio
+async def test_collect_routes_to_photo_when_llm_says_so():
+    # LLM 提议 2000（白名单合法）：写状态，下一轮路由到 single_photo
+    llm = FakeLLMService(
+        lambda s, u: "<state>2000</state>好的，信息已记录。接下来请对准车辆前方拍摄。"
+    )
+    result = await run_collect(collect_state(), llm)
+    assert result["stage_code"] == "2000"
+
+
+@pytest.mark.asyncio
+async def test_collect_missing_prefix_repaired_once():
+    # 第一次回复缺前缀 → 受控修复重试一次（带修复指令），第二次合法
+    replies = iter(["没有前缀的回复", "<state>1002</state>请问您现在还在事故现场吗？"])
+
+    def responder(s, u):
+        # 修复调用应携带修复指令（对前缀的要求）
+        if len(llm.calls) == 1:
+            assert "<state>" in u or "state" in s
+        return next(replies)
+
+    llm = FakeLLMService(responder)
+    result = await run_collect(collect_state(), llm)
+    assert len(llm.calls) == 2
+    assert result["response_text"].startswith("<state>1002</state>")
+
+
+@pytest.mark.asyncio
+async def test_collect_missing_prefix_unrepaired_is_recognizable_error():
+    # 修复后仍缺前缀：抛 CollectReplyError，不提交业务状态，
+    # 且调用次数 = 原调用 + 1 次修复（不得无限重试）
+    llm = FakeLLMService(lambda s, u: "还是没有前缀")
+    with pytest.raises(CollectReplyError):
+        await run_collect(collect_state(), llm)
+    assert len(llm.calls) == 2
+
+
+@pytest.mark.asyncio
+async def test_collect_illegal_stage_proposal_not_committed():
+    # LLM 提议非法阶段码（9999 未登记 / 非法）：不得写状态，
+    # 修复一次后仍非法 → 保持当前阶段并给出可识别错误
+    llm = FakeLLMService(lambda s, u: "<state>9999</state>正文")
+    with pytest.raises(CollectReplyError):
+        await run_collect(collect_state(), llm)
+    assert len(llm.calls) == 2
+
+
+@pytest.mark.asyncio
+async def test_graph_collect_node_with_llm():
+    # 图级：build_graph(llm=...) 时首轮走 welcome（不调模型），
+    # 第二轮起 collect 节点走 run_collect
+    llm = FakeLLMService(lambda s, u: "<state>1002</state>收到。")
+    graph = build_graph(llm=llm)
+    first = await graph.ainvoke(
+        {"session_id": "s1", "input_text": "追尾了"},
+        config={"configurable": {"thread_id": "s1"}},
+    )
+    assert len(llm.calls) == 0  # welcome 是确定性节点
+    assert first["stage_code"] == "1001"
+
+    result = await graph.ainvoke(
+        {"session_id": "s1", "input_text": "继续办理"},
+        config={"configurable": {"thread_id": "s1"}},
+    )
+    assert len(llm.calls) == 1
+    assert result["response_text"].startswith("<state>1002</state>")
+    assert result["stage_code"] == "1002"
+
+
+# ---------- 字段候选与时间校验（实现前失败，属预期） ----------
+
+
+def test_validate_candidate_unknown_key_rejected():
+    # 计划作业 4-4/6-7：未知 key 不得写入正式字段
+    with pytest.raises(ValueError):
+        validate_candidate(FieldCandidate(key="not_a_field", value=1))
+
+
+def test_validate_candidate_future_time_rejected():
+    clock = FixedClock(datetime(2026, 10, 19, 12, 0, 0))
+    result = validate_candidate(
+        FieldCandidate(key="sgfssj", value="明天早上八点"), clock=clock
+    )
+    assert result.status != FieldStatus.CONFIRMED
+
+
+def test_validate_candidate_count_not_inferred_from_description():
+    # 计划作业 4-8：数量未知不得从“追尾”等词推断
+    result = validate_candidate(FieldCandidate(key="jdcsl", value="追尾"))
+    assert result.status != FieldStatus.CONFIRMED
+
+
+def test_validate_accident_time_parses_absolute_and_relative():
+    now = datetime(2026, 10, 19, 12, 0, 0)
+    assert validate_accident_time("11点45分", now) is True
+    assert validate_accident_time("刚才", now) is True
+    assert validate_accident_time("明天早上八点", now) is False
+    assert validate_accident_time("乱七八糟", now) is False
@@ -46,11 +46,24 @@ async def test_different_threads_are_isolated() -> None:


@pytest.mark.asyncio
-async def test_reply_uses_fixed_template_with_stage_prefix() -> None:
+async def test_first_turn_is_welcome_stage() -> None:
+    # 新会话首条消息：welcome 节点固定问候，返回 1001（对照工作流
+    # work_start is_first_message → Assigned Reply#3），不调用模型
    graph = build_graph()
-
    result = await ask(graph, "thread-a", "你好")

-    assert result["stage_code"] == "1002"
-    assert result["response_text"].startswith("<state>1002</state>")
-    assert "你好" in result["response_text"]
+    assert result["stage_code"] == "1001"
+    assert result["response_text"].startswith("<state>1001</state>")
+    assert result["turn_count"] == 1
+
+
+@pytest.mark.asyncio
+async def test_second_turn_keeps_stage_without_event() -> None:
+    # 问候后的普通输入：脚手架版无事件则保持 1001，
+    # 【继续办理】等事件经白名单进入 1002（见 test_transitions.py）
+    graph = build_graph()
+    await ask(graph, "thread-a", "你好")
+    result = await ask(graph, "thread-a", "我出车祸了")
+
+    assert result["stage_code"] == "1001"
+    assert "我出车祸了" in result["response_text"]