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

mcp_server/tools/__init__.py

@@ -19,6 +19,7 @@ from mcp_server.tools.diagnose import mobile_diagnose
 from mcp_server.tools.inspect_trace import mobile_inspect_trace
 from mcp_server.tools.task_manager import mobile_manage_task
 from mcp_server.tools.task_runner import mobile_run_task
+from mcp_server.tools.exploration import mobile_exploration
 
 __all__ = [
     "mobile_run_task",
@@ -32,6 +33,7 @@ __all__ = [
     "mobile_batch",
     "mobile_search_history",
     "mobile_export_report",
+    "mobile_exploration",
 ]
 
 from mcp_server.tools.automation import (

mcp_server/tools/automation.py

@@ -143,7 +143,7 @@ async def mobile_workspace(include_devices: bool = True) -> dict:
 
 @mcp.tool()
 async def mobile_config(
-    section: Literal["model", "capabilities", "reports"] = "model",
+    section: Literal["model", "capabilities", "reports", "exploration"] = "model",
     changes: dict[str, Any] | None = None,
     persist_defaults: bool = False,
 ) -> dict:
@@ -163,6 +163,8 @@ async def mobile_config(
     worker_cleanup_timeout_seconds (30), adb_auto_connect (true),
     adb_discovery_interval_seconds (10), adb_discovery_timeout_seconds (5).
     reports: formats=['docx','xlsx'] (at least one).
+    exploration: model (default gpt-6.1-sol), reasoning_effort (high), timeout_seconds
+    (1800 active seconds), demonstration_timeout_seconds (900), demonstration_warning_seconds (60).
     API keys are never accepted or returned: use the Web UI or .env for credentials.
     Saving settings does not change an already running task. Flash/Pro are execution
     profiles, not model names. Model retries do not repeat device test cases.
@@ -171,6 +173,16 @@ async def mobile_config(
         "model": ("/api/system/model-settings", MODEL_FIELDS),
         "capabilities": ("/api/system/capability-settings", CAPABILITY_FIELDS),
         "reports": ("/api/automation/preferences", {"formats"}),
+        "exploration": (
+            "/api/explorations/settings",
+            {
+                "model",
+                "reasoning_effort",
+                "timeout_seconds",
+                "demonstration_timeout_seconds",
+                "demonstration_warning_seconds",
+            },
+        ),
     }[section]
     if changes is not None:
         if not persist_defaults:
@@ -182,7 +194,9 @@ async def mobile_config(
         return current
     payload = {key: value for key, value in current.items() if key in allowed}
     payload.update(changes)
-    return await _request("PUT" if section == "reports" else "POST", route, json=payload)
+    return await _request(
+        "PUT" if section in {"reports", "exploration"} else "POST", route, json=payload
+    )
 
 
 @mcp.tool()

mcp_server/tools/exploration.py

@@ -0,0 +1,97 @@
+"""Guided exploration through the same HTTP service used by the workspace."""
+
+from typing import Any, Literal
+
+from mcp_server.base import mcp
+from mcp_server.tools.automation import _id, _request
+
+
+@mcp.tool()
+async def mobile_exploration(
+    action: Literal[
+        "list",
+        "get",
+        "create",
+        "start",
+        "message",
+        "pause",
+        "continue",
+        "stop",
+        "demonstrate",
+        "demo_finish",
+        "demo_cancel",
+        "demo_extend",
+        "case",
+        "save_case",
+        "generate",
+        "validate",
+        "batch",
+    ],
+    exploration_id: str | None = None,
+    request_id: str | None = None,
+    payload: dict[str, Any] | None = None,
+    demonstration_id: str | None = None,
+    after: int = 0,
+    version: int | None = None,
+) -> Any:
+    """Explore ONE complex test with user guidance and produce a reusable case.
+
+    Read-only list/get/case never execute tasks. create stores {goal,config?,task_config?};
+    start/continue explicitly enqueue or resume Pro exploration. Independent defaults:
+    gpt-6.1-sol/high, 1800 active seconds; timeout pauses, never fails by itself.
+    message payload={text}; answer pending questions or guide without discarding history.
+    demonstrate payload={source:'android'|'computer'|'both'} pauses at a safe point and
+    records while retaining the device/plugin lease. Wait for recording before the user
+    operates. demo_finish analyzes the SAVED clip; demo_extend payload={seconds:900};
+    demo_cancel preserves evidence. Default 900s, warning 60s before expiry, auto-pause.
+    save_case payload={expected_version,case:{name,goal,preconditions?,evidence?}};
+    generate payload={source_run_id,expected_version} starts evidence-based synthesis,
+    poll get.generations for completion. validate payload={version,profile,task_config?}
+    explicitly runs the saved version in Flash/Pro; batch additionally accepts name and
+    creates a DRAFT batch, never starts it. Exploration pass and rerun pass are separate.
+    Use stable request_id for every mutation except pause/stop, reusing it on transport
+    retries. No API keys; configure secrets in the Web UI. mobile_config(section='exploration')
+    controls non-secret defaults. After errors call mobile_diagnose before resubmitting.
+    """
+    base = "/api/explorations"
+    if action in {"list", "get", "case"}:
+        if payload or request_id:
+            raise ValueError("查询不接受写入内容")
+        if action == "list":
+            return await _request("GET", base)
+        path = f"{base}/{_id(exploration_id)}"
+        if action == "get":
+            if after < 0:
+                raise ValueError("after 不能小于零")
+            return await _request("GET", path, params={"after": after})
+        return await _request(
+            "GET", path + "/case", params={"version": version} if version is not None else {}
+        )
+    if action not in {"pause", "stop"} and (not request_id or not request_id.strip()):
+        raise ValueError("写入探索需提供稳定的 request_id")
+    body = dict(payload or {})
+    if "request_id" in body:
+        raise ValueError("request_id 请使用单独参数")
+    if request_id:
+        body["request_id"] = request_id
+    if action == "create":
+        return await _request("POST", base, json=body, params={"source": "mcp"})
+    suffix = {
+        "start": "start",
+        "message": "messages",
+        "pause": "pause",
+        "continue": "continue",
+        "stop": "stop",
+        "demonstrate": "demonstrations",
+        "save_case": "case",
+        "generate": "generate",
+        "validate": "validate",
+        "batch": "batch",
+    }.get(action)
+    if action.startswith("demo_"):
+        suffix = f"demonstrations/{_id(demonstration_id)}/{action.removeprefix('demo_')}"
+    if suffix is None:
+        raise ValueError("不支持的探索操作")
+    return await _request(
+        "POST", f"{base}/{_id(exploration_id)}/{suffix}", json=body, params={"source": "mcp"}
+    )

mcp_server/tools/task_runner.py

@@ -40,6 +40,7 @@ from artemis.runtime.task_limits import ExecutionBudget, task_timeout
 from artemis.runtime.task_control import read_state
 from artemis.runtime.task_result import save_failure_result, read_worker_outcome
 from artemis.config.settings import settings
+from artemis.runtime.daemon_client import local_daemon_url
 
 # Seconds the spawned runner gets to finish its imports and open its log files
 # before the spawn is declared dead. Normal startup creates stdout.log within a
@@ -393,7 +394,7 @@ def mobile_run_task(
             return {"trace_id": trace_id, **rejection}
 
     # 3. Dispatch via unified Artemis Daemon scheduler if available (unless standalone forced)
-    if os.environ.get("ARTEMIS_STANDALONE") != "1":
+    if local_daemon_url() or os.environ.get("ARTEMIS_STANDALONE") != "1":
         try:
             is_running, base_url = ensure_daemon_running(timeout=2.0, wait_ready=True)
             if is_running: