feat: add guided exploration and share user-started MCP backend

tests/unit/test_exploration_recovery.py

@@ -0,0 +1,123 @@
+import os
+import time
+import subprocess
+import sys
+from unittest.mock import AsyncMock, Mock
+
+import pytest
+
+from apps.admin_console.services.automation_store import AutomationStore
+from apps.admin_console.services.exploration_recovery import reconcile_explorations
+from artemis.config.exploration import ExplorationConfig
+from artemis.exploration.editorial import DiscussionReply, discuss
+from artemis.exploration.store import ExplorationStore
+from artemis.runtime import task_control
+
+
+def test_worker_can_import_exploration_without_importing_web_server():
+    result = subprocess.run(
+        [
+            sys.executable,
+            "-c",
+            "import sys; from artemis.exploration.runtime import ExplorationControl; assert 'apps.admin_console.server' not in sys.modules",
+        ],
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+        timeout=30,
+    )
+    assert result.returncode == 0, result.stderr
+
+
+def test_web_and_exploration_share_the_same_queue_state():
+    code = "from apps.admin_console import server; from apps.admin_console.services import task_queue_service, exploration_recovery; from apps.admin_console.routers import tasks; assert server.state is task_queue_service.state is exploration_recovery.state is tasks.state"
+    result = subprocess.run(
+        [sys.executable, "-c", code], capture_output=True, text=True, encoding="utf-8", timeout=30
+    )
+    assert result.returncode == 0, result.stderr
+
+
+@pytest.fixture
+def records(tmp_path, monkeypatch):
+    monkeypatch.setattr(task_control, "TRACES_PATH", tmp_path / "traces")
+    repo = ExplorationStore(AutomationStore(tmp_path / "history.db"))
+    session = repo.create("探索设置", ExplorationConfig(), {}, "create")
+    return repo, session
+
+
+def test_recovery_preserves_live_paused_owner_and_real_verdict(records, monkeypatch):
+    repo, session = records
+    run = repo.reserve_run(session["id"], "start", kind="exploration", goal="设置", task_config={})
+    repo.set_run_state(run["id"], "paused", reason="execution_budget")
+    monkeypatch.setattr(
+        "apps.admin_console.services.exploration_recovery.read_state",
+        lambda _: {"pid": os.getpid()},
+    )
+    reconcile_explorations(repo)
+    assert repo.get(session["id"])["status"] == "paused"
+    monkeypatch.setattr(
+        "apps.admin_console.services.exploration_recovery.read_worker_outcome",
+        lambda _: {
+            "task_status": "completed",
+            "test_result": {"status": "passed", "reason": "有证据"},
+        },
+    )
+    reconcile_explorations(repo)
+    assert repo.get(session["id"])["active_run_id"] is None
+    assert repo.run(run["id"])["outcome"]["test_result"]["status"] == "passed"
+
+
+def test_orphan_editorial_jobs_become_retryable_without_starting_tasks(records):
+    repo, session = records
+    for kind in ("exploration_generation", "exploration_discussion"):
+        repo.database.put(
+            kind,
+            "dead",
+            {
+                "id": "dead",
+                "status": "running",
+                "owner_pid": 0,
+                "lease_until": time.time() + 999,
+                "exploration_id": session["id"],
+            },
+        )
+    reconcile_explorations(repo)
+    assert repo.runs(session["id"]) == []
+    assert repo.database.records("exploration_generation")[0]["status"] == "failed"
+    assert repo.database.records("exploration_discussion")[0]["status"] == "failed"
+
+
+def test_validation_status_follows_actual_worker_without_creating_outcome(records, monkeypatch):
+    from artemis.exploration.models import CaseDraft
+
+    repo, session = records
+    repo.save_case(session["id"], CaseDraft(name="设置", goal="检查设置"), 0, "save")
+    run = repo.reserve_run(
+        session["id"], "validate", kind="validation", goal="复跑", task_config={}, case_version=1
+    )
+    monkeypatch.setattr(
+        "apps.admin_console.services.exploration_recovery.read_state",
+        lambda _: {"pid": os.getpid(), "status": "running"},
+    )
+    reconcile_explorations(repo)
+    assert repo.run(run["id"])["status"] == "running"
+    assert repo.run(run["id"])["outcome"] is None
+
+
+@pytest.mark.asyncio
+async def test_completed_discussion_is_idempotent_and_never_admits_device_work(
+    records, monkeypatch
+):
+    repo, session = records
+    message = repo.add_message(session["id"], "为什么选这个入口？", "m")
+    monkeypatch.setattr("artemis.exploration.editorial.editorial_context", Mock())
+    model = Mock()
+    model.with_structured_output.return_value.ainvoke = AsyncMock(
+        return_value=DiscussionReply(reply="已观察到的入口在右上角；可编辑提示词。")
+    )
+    monkeypatch.setattr("artemis.exploration.editorial.get_llm", Mock(return_value=model))
+    await discuss(repo, session["id"], message)
+    await discuss(repo, session["id"], message)
+    assert len([m for m in repo.messages(session["id"]) if m["role"] == "assistant"]) == 1
+    assert repo.runs(session["id"]) == []
+    model.with_structured_output.return_value.ainvoke.assert_awaited_once()