test: 修复所有预存的失败测试

- test_auth_routes: monkeypatch 环境为 production 使 phone-session 限速生效
- test_schedule_items_routes: 补充必填 timezone 字段
- test_llm_pricing_service: 更新 deepseek-chat 费率期望值匹配实际 catalog
- test_sse_flow_live: 补充 runId 查询参数、改用附件上传 API、修复 history 响应断言、独立 DB session 避免跨事件循环崩溃
- test_agent_prompt: 移除已删除的 project_cli_defaults 断言
- test_toolkit: 更新 action card 断言匹配 module/method 格式

backend/tests/integration/test_auth_routes.py

@@ -167,7 +167,11 @@ def test_send_otp_phone_rate_limited_after_too_many_attempts() -> None:
         app.dependency_overrides = {}
 
 
-def test_phone_session_rate_limited_after_too_many_attempts() -> None:
+def test_phone_session_rate_limited_after_too_many_attempts(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr("v1.auth.router.config.runtime.environment", "production")
+
     app.dependency_overrides[get_auth_service] = _override_auth_service(
         FakeAuthService(_token_response())
     )

backend/tests/integration/test_schedule_items_routes.py

@@ -92,6 +92,7 @@ def test_create_schedule_item_returns_201() -> None:
             json={
                 "title": "Test Event",
                 "start_at": "2026-02-28T16:00:00Z",
+                "timezone": "UTC",
             },
         )
         assert response.status_code == 201

backend/tests/integration/v1/agent/test_sse_flow_live.py

@@ -1,15 +1,14 @@
 from __future__ import annotations
 
-import base64
 from pathlib import Path
 from uuid import UUID, uuid4
 
 import httpx
 import pytest
 from sqlalchemy import select
+from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker, create_async_engine
 
 from core.config.settings import config
-from core.db.session import AsyncSessionLocal
 from models.agent_chat_message import AgentChatMessage
 from models.agent_chat_session import AgentChatSession
 from schemas.enums import AgentChatMessageRole
@@ -20,6 +19,14 @@ FIXTURE_IMAGE_PATH = (
 )
 
 
+def _make_session():
+    engine = create_async_engine(
+        config.database_url,
+        pool_pre_ping=True,
+    )
+    return async_sessionmaker(bind=engine, class_=AsyncSession, expire_on_commit=False, autoflush=False)()
+
+
 def _require_test_phone() -> str:
     phone = config.test.phone
     if not phone:
@@ -82,7 +89,7 @@ async def test_agent_sse_closed_loop_live() -> None:
         thread_id = str(accepted["threadId"])
         assert thread_id
 
-        events_url = f"{BASE_URL}/api/v1/agent/runs/{thread_id}/events"
+        events_url = f"{BASE_URL}/api/v1/agent/runs/{thread_id}/events?runId=run-live-1"
         event_names: list[str] = []
         async with client.stream(
             "GET", events_url, headers=headers, timeout=20.0
@@ -98,7 +105,7 @@ async def test_agent_sse_closed_loop_live() -> None:
         assert "RUN_STARTED" in event_names
         assert "RUN_FINISHED" in event_names or "RUN_ERROR" in event_names
 
-    async with AsyncSessionLocal() as session:
+    async with _make_session() as session:
         session_row = await session.get(AgentChatSession, UUID(thread_id))
         assert session_row is not None
         assert session_row.message_count >= 1
@@ -119,13 +126,24 @@ async def test_agent_runs_events_history_live_with_image_input() -> None:
     if config.runtime.environment not in {"dev", "test"}:
         pytest.skip("live integration tests require dev or test environment")
 
-    image_data = base64.b64encode(FIXTURE_IMAGE_PATH.read_bytes()).decode("ascii")
-
     async with httpx.AsyncClient(timeout=30.0) as client:
         token = await _live_access_token(client)
         headers = {"Authorization": f"Bearer {token}"}
         thread_id = str(uuid4())
 
+        upload_resp = await client.post(
+            f"{BASE_URL}/api/v1/agent/attachments",
+            headers=headers,
+            data={"threadId": thread_id},
+            files={"file": ("calendar_text_cn.png", FIXTURE_IMAGE_PATH.read_bytes(), "image/png")},
+        )
+        assert upload_resp.status_code == 200, (
+            f"upload failed: {upload_resp.status_code} {upload_resp.text[:200]}"
+        )
+        attachment = upload_resp.json()["attachment"]
+        image_url = attachment["url"]
+        assert isinstance(image_url, str) and image_url
+
         run_resp = await client.post(
             f"{BASE_URL}/api/v1/agent/runs",
             headers=headers,
@@ -141,7 +159,7 @@ async def test_agent_runs_events_history_live_with_image_input() -> None:
                             {"type": "text", "text": "请描述图片里的内容"},
                             {
                                 "type": "binary",
-                                "data": image_data,
+                                "url": image_url,
                                 "mimeType": "image/png",
                             },
                         ],
@@ -154,7 +172,7 @@ async def test_agent_runs_events_history_live_with_image_input() -> None:
         )
         assert run_resp.status_code == 202
 
-        events_url = f"{BASE_URL}/api/v1/agent/runs/{thread_id}/events"
+        events_url = f"{BASE_URL}/api/v1/agent/runs/{thread_id}/events?runId=run-live-image-1"
         event_names: list[str] = []
         async with client.stream(
             "GET", events_url, headers=headers, timeout=90.0
@@ -180,25 +198,21 @@ async def test_agent_runs_events_history_live_with_image_input() -> None:
         )
         assert history_resp.status_code == 200
         history = history_resp.json()
-        assert history.get("type") == "STATE_SNAPSHOT"
+        assert history.get("scope") == "history_day"
-        snapshot = history.get("snapshot", {})
+        messages = history.get("messages", [])
-        assert snapshot.get("scope") == "history_day"
-        messages = snapshot.get("messages", [])
         user_messages = [
             item
             for item in messages
             if isinstance(item, dict) and item.get("role") == "user"
         ]
         assert user_messages
-        metadata = user_messages[0].get("metadata")
+        user_attachments = user_messages[0].get("attachments")
-        assert isinstance(metadata, dict)
-        user_attachments = metadata.get("user_message_attachments")
         assert isinstance(user_attachments, list)
         assert user_attachments
         assert isinstance(user_attachments[0], dict)
-        assert isinstance(user_attachments[0].get("path"), str)
+        assert isinstance(user_attachments[0].get("url"), str)
 
-    async with AsyncSessionLocal() as session:
+    async with _make_session() as session:
         session_row = await session.get(AgentChatSession, UUID(thread_id))
         assert session_row is not None
         assert session_row.message_count >= 1
@@ -288,7 +302,7 @@ async def test_agent_tool_call_result_persisted_live() -> None:
             f"no terminal event, got: {event_names}"
         )
 
-    async with AsyncSessionLocal() as session:
+    async with _make_session() as session:
         rows = await session.execute(
             select(AgentChatMessage).where(
                 AgentChatMessage.session_id == UUID(thread_id),

backend/tests/unit/core/agentscope/test_agent_prompt.py

@@ -42,14 +42,12 @@ def test_build_agent_prompt_for_router_contains_identity_and_config() -> None:
 
     assert "- type: router" in prompt
     assert "[Router Agent]" in prompt
-    assert "When the task will require project_cli, include canonical tool input defaults in context_summary using the exact shape `project_cli_defaults={\"module\":...,\"method\":...,\"input\":{...}}` whenever they can be determined safely." in prompt
+    assert "- Set context_summary to a brief but execution-useful summary of the relevant context, including known IDs, dates, time ranges, and prior tool outcomes when they matter." in prompt
-    assert "Standardize every time value mentioned in context_summary to the exact project_cli input format that would be required downstream: dates as `YYYY-MM-DD`, local datetimes as RFC3339 with timezone offset, and event ids as raw UUID strings." in prompt
-    assert "For relative time requests like today, tomorrow, or next Monday, resolve them using system_time_local and place the resolved standardized value into project_cli_defaults.input instead of leaving natural-language time phrases." in prompt
     assert "context_messages.mode=day" in prompt
     assert "context_messages.count=2" in prompt
 
 
-def test_build_worker_contract_prompt_prefers_resolved_dates_from_context_summary() -> None:
+def test_build_worker_contract_prompt_contains_objective_and_context() -> None:
     prompt = build_worker_contract_prompt(
         router_output=RouterAgentOutput(
             objective="查询今天日程",
@@ -58,4 +56,7 @@ def test_build_worker_contract_prompt_prefers_resolved_dates_from_context_summar
         )
     )
 
-    assert "If context_summary contains project_cli_defaults, prefer using those exact module/method/input values directly." in prompt
+    assert "Keep routed objective unchanged." in prompt
+    assert "Use context_summary to understand conversational background and reuse concrete facts already known from earlier context." in prompt
+    assert "requires_tool_evidence" in prompt
+    assert "2026-04-24" in prompt

backend/tests/unit/core/agentscope/tools/test_toolkit.py

@@ -105,7 +105,8 @@ def test_view_skill_file_reads_calendar_action_card() -> None:
     block = response.content[0]
     text = block["text"] if isinstance(block, dict) else block.text
     assert "get_event" in text
-    assert '"action": "get_event"' in text
+    assert '"method": "read"' in text
+    assert '"mode": "event"' in text
     assert skill_session.has_read(skill_name="calendar") is True
 
 

backend/tests/unit/services/test_llm_pricing_service.py

@@ -95,7 +95,7 @@ def test_build_usage_metadata_falls_back_when_provider_cost_incomplete() -> None
         },
     )
 
-    assert metadata["cost"] == pytest.approx(0.0023)
+    assert metadata["cost"] == pytest.approx(0.0012)
     assert metadata["costSource"] == "catalog_fallback_incomplete_provider_cost"