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

apps/admin_console/services/automation_service.py

@@ -193,6 +193,9 @@ async def tick_batches():
     from apps.admin_console.services.report_export import export_batch
 
     await reconcile_recovered_jobs()
+    from apps.admin_console.services.exploration_recovery import reconcile_explorations
+
+    reconcile_explorations()
     for batch in store.records("batch"):
         if batch["status"] not in {"running", "paused"}:
             continue

apps/admin_console/services/evidence_service.py

@@ -77,8 +77,10 @@ def plugin_sources(session_id: str) -> list[dict]:
 
 
 def plugin_files(session_id: str) -> list[tuple[Path, str]]:
+    from apps.admin_console.services.exploration_evidence import demonstration_files
+
     root = task_root(session_id)
-    files = {}
+    files = dict(demonstration_files(session_id))
     for source in plugin_sources(session_id):
         for segment in source["segments"]:
             path = evidence_file(root, segment["file"])
@@ -105,9 +107,13 @@ def plugin_files(session_id: str) -> list[tuple[Path, str]]:
 
 def export_index(session_id: str, report: dict, android_videos: list) -> dict:
     from apps.admin_console.services.report_export import session_artifacts, session_screenshots
+    from apps.admin_console.services.exploration_evidence import (
+        exploration_index,
+        demonstration_sources,
+    )
 
     extras = plugin_files(session_id)
-    recordings = plugin_sources(session_id)
+    recordings = plugin_sources(session_id) + demonstration_sources(session_id)
     root = task_root(session_id)
     paths = {path.resolve(): name for path, name in extras}
     for source in recordings:
@@ -138,6 +144,7 @@ def export_index(session_id: str, report: dict, android_videos: list) -> dict:
     return {
         "version": 1,
         "session_id": session_id,
+        "exploration": exploration_index(session_id, paths),
         "name": report["name"],
         "status": report["status"],
         "verdict": report["verdict"],

apps/admin_console/services/exploration_evidence.py

@@ -0,0 +1,110 @@
+"""Read-only exploration provenance for reports, archives and source-aware video."""
+
+import json
+
+from artemis.exploration.store import explorations
+from artemis.exploration.demonstrations import Demonstrations
+from apps.admin_console.services.evidence_service import evidence_file, task_root
+
+
+def exploration_metadata(run_id: str) -> dict | None:
+    run = explorations.run(run_id)
+    if not run:
+        return None
+    return {
+        "id": run["exploration_id"],
+        "kind": run["kind"],
+        "case_version": run.get("case_version"),
+    }
+
+
+def demonstration_sources(run_id: str) -> list[dict]:
+    root = task_root(run_id)
+    directory = root / "evidence" / "demonstrations"
+    from apps.admin_console.services.media_service import MediaService
+
+    sources = []
+    for manifest in directory.glob("*/demonstration.json"):
+        if (
+            not evidence_file(root, manifest.relative_to(root).as_posix())
+            or manifest.stat().st_size > 4 * 1024 * 1024
+        ):
+            continue
+        try:
+            record = json.loads(manifest.read_text(encoding="utf-8"))
+            for track in record["media"]:
+                segments = []
+                for item in track["segments"]:
+                    path = evidence_file(root, item["file"])
+                    if path and path.suffix.lower() == ".mp4" and item["duration_ms"] > 0:
+                        segments.append(
+                            {
+                                **item,
+                                "start": item["offset_ms"] / 1000,
+                                "duration": item["duration_ms"] / 1000,
+                                "url": MediaService.path_to_video_url(path),
+                            }
+                        )
+                if segments:
+                    sources.append(
+                        {
+                            **track,
+                            "segments": segments,
+                            "label": "人工示范 · "
+                            + ("安卓" if track["kind"] == "android" else "电脑"),
+                            "status": "ready",
+                            "message": "人工操作录像，不等于自动复跑通过",
+                            "demonstration_id": record["id"],
+                        }
+                    )
+        except (OSError, ValueError, KeyError, TypeError):
+            continue
+    return sources
+
+
+def demonstration_files(run_id: str) -> list[tuple]:
+    root = task_root(run_id)
+    directory = root / "evidence" / "demonstrations"
+    files = []
+    for path in directory.rglob("*"):
+        if path.suffix.lower() not in {".mp4", ".mkv", ".webm"}:
+            continue
+        checked = evidence_file(root, path.relative_to(root).as_posix())
+        if checked:
+            files.append((checked, "任务录屏/人工示范/" + path.relative_to(directory).as_posix()))
+    return files
+
+
+def exploration_index(run_id: str, archive_paths: dict | None = None) -> dict | None:
+    run = explorations.run(run_id)
+    if not run:
+        return None
+    sid = run["exploration_id"]
+    session = explorations.get(sid)
+    messages = []
+    after = 0
+    while page := explorations.messages(sid, after=after):
+        messages.extend(page)
+        after = page[-1]["sequence"]
+    demos = Demonstrations(explorations).list(sid)
+    root = task_root(run_id)
+    for demo in demos:
+        for track in demo.get("media", []):
+            for segment in track.get("segments", []):
+                relative = segment.pop("file", "")
+                segment["trace_file"] = relative
+                segment["file"] = (
+                    (archive_paths or {}).get((root / relative).resolve())
+                    if demo["run_id"] == run_id
+                    else None
+                )
+    return {
+        "session": session,
+        "messages": messages,
+        "runs": explorations.runs(sid),
+        "cases": [
+            explorations.case(sid, version) for version in range(1, session["case_version"] + 1)
+        ],
+        "demonstrations": demos,
+        "note": "此导出保留会话、用例版本和各次运行的关联；只有当前任务的媒体文件在此目录中。",
+    }

apps/admin_console/services/exploration_listing.py

@@ -0,0 +1,45 @@
+"""Shared exploration metadata for queue cards and history pagination."""
+
+# Rank execution runs independently from validations: a rerun never replaces the
+# conversation card. Keep all raw sessions accessible for evidence inspection.
+EXPLORATION_RUNS_CTE = """
+exploration_runs AS (
+ SELECT r.id AS run_id,
+   json_extract(r.payload,'$.exploration_id') AS exploration_id,
+   json_extract(r.payload,'$.kind') AS exploration_kind,
+   json_extract(r.payload,'$.case_version') AS case_version,
+   json_extract(r.payload,'$.task_config.model') AS exploration_model,
+   json_extract(r.payload,'$.task_config.profile') AS exploration_profile,
+   json_extract(v.payload,'$.name') AS exploration_case_name,
+   CASE WHEN json_extract(r.payload,'$.kind')='exploration' AND
+     (json_extract(c.payload,'$.active_run_id') IS NULL
+     OR json_extract(c.payload,'$.active_run_id') = r.id)
+     THEN json_extract(c.payload,'$.status')
+     ELSE json_extract(r.payload,'$.status') END AS exploration_status,
+   FIRST_VALUE(r.id) OVER (
+     PARTITION BY json_extract(r.payload,'$.exploration_id'),json_extract(r.payload,'$.kind')
+     ORDER BY json_extract(r.payload,'$.created_at') DESC, r.id DESC
+   ) AS exploration_latest_run_id,
+   COUNT(*) OVER (
+     PARTITION BY json_extract(r.payload,'$.exploration_id'),json_extract(r.payload,'$.kind')
+   ) AS exploration_run_count
+ FROM automation_records r
+ LEFT JOIN automation_records c ON c.kind='exploration'
+   AND c.id=json_extract(r.payload,'$.exploration_id')
+ LEFT JOIN automation_records v ON v.kind='exploration_case'
+   AND v.id=json_extract(r.payload,'$.exploration_id') || '_' ||
+     printf('%010d',json_extract(r.payload,'$.case_version'))
+ WHERE r.kind='exploration_run'
+)
+"""
+
+
+def exploration_metadata(connection) -> dict[str, dict]:
+    rows = connection.execute("WITH " + EXPLORATION_RUNS_CTE + " SELECT * FROM exploration_runs")
+    result = {}
+    for row in rows:
+        item = {key: row[key] for key in row.keys() if key != "run_id"}
+        if item.get("exploration_kind") == "validation" and item.get("exploration_case_name"):
+            item["display_name"] = item["exploration_case_name"]
+        result[row["run_id"]] = item
+    return result

apps/admin_console/services/exploration_recovery.py

@@ -0,0 +1,95 @@
+"""Reconcile saved exploration metadata with the existing queue and worker owners."""
+
+import json
+import time
+
+from apps.admin_console.core.state import state
+from apps.admin_console.database.repositories.session_repository import SessionRepository
+from artemis.exploration.store import ExplorationConflict, ExplorationStore, explorations
+from artemis.runtime.process_probe import pid_is_alive
+from artemis.runtime.task_control import read_state
+from artemis.runtime.task_result import read_worker_outcome
+from artemis.runtime.case_result import read_case_result
+
+
+def reconcile_explorations(repository: ExplorationStore = explorations) -> None:
+    for kind in ("exploration_generation", "exploration_discussion"):
+        with repository._transaction() as connection:
+            for row in connection.execute(
+                "SELECT id, payload FROM automation_records WHERE kind=? AND json_extract(payload,'$.status')='running'",
+                (kind,),
+            ).fetchall():
+                job = json.loads(row[1])
+                if (
+                    not pid_is_alive(job.get("owner_pid"))
+                    or job.get("lease_until", 0) < time.time()
+                ):
+                    repository._put(
+                        connection,
+                        kind,
+                        row[0],
+                        {
+                            **job,
+                            "status": "failed",
+                            "lease_until": 0,
+                            "error": "生成进程已中断；已有结果保留，可重试。",
+                        },
+                    )
+    with repository.database.connection() as connection:
+        rows = connection.execute(
+            "SELECT payload FROM automation_records WHERE kind='exploration' "
+            "AND json_extract(payload,'$.active_run_id') IS NOT NULL"
+        ).fetchall()
+    history = SessionRepository(repository.database.path)
+    queued = {
+        str(item.get("session_id"))
+        for item in state.queue_items
+        if isinstance(item, dict) and item.get("status") in {"pending", "running", "queued"}
+    }
+    for row in rows:
+        session = json.loads(row[0])
+        run_id = session["active_run_id"]
+        run = repository.run(run_id)
+        if not run:
+            continue
+        outcome = read_worker_outcome(run_id)
+        if outcome:
+            repository.finish_run(run_id, {**outcome, "status": outcome["task_status"]})
+            continue
+        control = read_state(run_id)
+        if run["kind"] == "validation" and pid_is_alive(control.get("pid")):
+            observed = {"running": "running", "task_paused": "paused", "resuming": "queued"}.get(
+                control.get("status")
+            )
+            if observed and observed != run["status"]:
+                try:
+                    repository.set_run_state(run_id, observed, reason=control.get("pause_reason"))
+                except ExplorationConflict:
+                    continue  # The owner finalized while this tick read its control file.
+        if run_id in queued or run_id in state.active_runs:
+            continue
+        if control.get("pid") and pid_is_alive(control["pid"]):
+            continue
+        execution = history.get_session_by_id(run_id) or {}
+        if execution.get("pid") and history.process_is_alive(execution["pid"]):
+            continue
+        job = repository.database.job(run_id) or {}
+        terminal = execution.get("status") or job.get("terminal_status") or job.get("status")
+        if terminal in {"completed", "failed", "cancelled", "interrupted", "skipped"}:
+            repository.finish_run(
+                run_id,
+                {
+                    "status": terminal,
+                    "error": execution.get("error") or job.get("error") or "",
+                    "test_result": execution.get("test_result")
+                    or read_case_result(run_id, terminal),
+                },
+            )
+        elif time.time() - run["created_at"] > 300:
+            repository.finish_run(
+                run_id,
+                {
+                    "status": "interrupted",
+                    "error": "探索工作进程已结束；原证据已保留，可继续探索并重新观察当前设备。",
+                },
+            )

apps/admin_console/services/exploration_service.py

@@ -0,0 +1,334 @@
+"""Web and MCP use the same exploration admission and conversation service."""
+
+from artemis.config.agent import load_agent_config
+from artemis.config.exploration import ExplorationConfig
+from artemis.exploration.models import CreateExploration, SaveCaseRequest, ValidateCaseRequest
+from artemis.exploration.demonstrations import Demonstrations
+from artemis.exploration.store import ExplorationConflict, ExplorationStore, explorations
+from artemis.runtime.task_control import add_instruction, read_state, request_pause, request_resume
+from apps.admin_console.services.task_import import TaskOptions
+
+
+def exploration_case_prompt(case: dict) -> str:
+    prefix = f"前置条件：\n{case['preconditions']}\n\n" if case.get("preconditions") else ""
+    return prefix + case["goal"]
+
+
+class ExplorationService:
+    def __init__(self, repository: ExplorationStore | None = None):
+        self.repository = repository if repository is not None else explorations
+
+    def settings(self) -> dict:
+        saved = self.repository.database.get("settings", "exploration") or {}
+        return ExplorationConfig.model_validate(
+            {
+                **load_agent_config().exploration.model_dump(),
+                **saved,
+            }
+        ).model_dump()
+
+    def save_settings(self, value: ExplorationConfig) -> dict:
+        return self.repository.database.put("settings", "exploration", value.model_dump())
+
+    def create(self, value: CreateExploration, *, source: str = "web") -> dict:
+        options = TaskOptions.model_validate(value.task_config).model_dump(exclude_none=True)
+        return self.repository.create(
+            value.goal,
+            value.config or ExplorationConfig.model_validate(self.settings()),
+            options,
+            value.request_id,
+            source=source,
+        )
+
+    def detail(self, session_id: str, *, after: int = 0) -> dict:
+        session = self.repository.get(session_id)
+        runs = self.repository.runs(session_id)
+        version = self.repository.case(session_id)
+        with self.repository.database.connection() as connection:
+            import json
+
+            generations = [
+                json.loads(row[0])
+                for row in connection.execute(
+                    "SELECT payload FROM automation_records WHERE kind='exploration_generation' "
+                    "AND json_extract(payload,'$.exploration_id')=? ORDER BY updated_at DESC LIMIT 20",
+                    (session_id,),
+                ).fetchall()
+            ]
+        from artemis.runtime.case_result import resolve_case_result
+
+        for run in runs:
+            outcome = run.get("outcome")
+            verdict = resolve_case_result(outcome, run["status"]) if outcome else None
+            run["test_result"] = verdict.model_dump() if verdict else None
+        return {
+            **session,
+            "messages": self.repository.messages(session_id, after=after),
+            "runs": runs,
+            "case": version,
+            "generations": [
+                {k: v for k, v in g.items() if k not in {"owner", "owner_pid", "lease_until"}}
+                for g in generations
+            ],
+            "demonstrations": Demonstrations(self.repository).list(session_id),
+            "validation_runs": [
+                r
+                for r in runs
+                if r["kind"] == "validation" and r["case_version"] == session["case_version"]
+            ],
+            "control": read_state(session["active_run_id"]) if session["active_run_id"] else {},
+        }
+
+    async def _admit(self, run: dict, *, source: str) -> dict:
+        # RunRequest performs the same readiness and idempotency checks as /api/run.
+        from apps.admin_console.routers.tasks import run_task
+        from apps.admin_console.schemas.task_schema import RunRequest
+        from fastapi import HTTPException
+
+        current = self.repository.run(run["id"])
+        if current and current.get("outcome") is not None:
+            return current
+        options = TaskOptions.model_validate(run["task_config"])
+        try:
+            result = await run_task(
+                RunRequest(
+                    goal=run["goal"],
+                    profile=options.profile or "pro",
+                    session_id=run["id"],
+                    ingress=source,
+                    conversation_id=run["exploration_id"],
+                    runtime_overrides=options,
+                )
+            )
+        except HTTPException as exc:
+            if exc.status_code < 500:
+                self.repository.finish_run(
+                    run["id"], {"status": "rejected", "error": str(exc.detail)}
+                )
+            raise
+        if result.get("status") in {"error", "rejected"}:
+            self.repository.finish_run(
+                run["id"],
+                {
+                    "status": "rejected",
+                    "error": result.get("error") or result.get("message") or "任务入队失败",
+                },
+            )
+        else:
+            self.repository.set_run_state(run["id"], "queued", expected_status="reserved")
+        return {"run": self.repository.run(run["id"]), "admission": result}
+
+    async def start(self, session_id: str, request_id: str, *, source: str = "web") -> dict:
+        session = self.repository.get(session_id)
+        config = ExplorationConfig.model_validate(session["config"])
+        options = {
+            **session["task_config"],
+            "profile": "pro",
+            "model": config.model,
+            "reasoning_effort": config.reasoning_effort,
+            "timeout_seconds": config.timeout_seconds,
+            "verification_level": "final",
+        }
+        run = self.repository.reserve_run(
+            session_id,
+            request_id,
+            kind="exploration",
+            goal=session["goal"],
+            task_config=options,
+        )
+        return await self._admit(run, source=source)
+
+    async def validate(
+        self, session_id: str, value: ValidateCaseRequest, *, source: str = "web"
+    ) -> dict:
+        case = self.repository.case(session_id, value.version)
+        if case is None:
+            raise KeyError("用例版本不存在")
+        session = self.repository.get(session_id)
+        # Device/plugin binding may be reused; exploration's expensive model is not.
+        options = {
+            k: v
+            for k, v in session["task_config"].items()
+            if k not in {"model", "reasoning_effort", "profile", "timeout_seconds"}
+        }
+        options.update(TaskOptions.model_validate(value.task_config).model_dump(exclude_none=True))
+        options["profile"] = value.profile
+        run = self.repository.reserve_run(
+            session_id,
+            value.request_id,
+            kind="validation",
+            goal=exploration_case_prompt(case),
+            task_config=options,
+            case_version=value.version,
+        )
+        return await self._admit(run, source=source)
+
+    def message(
+        self,
+        session_id: str,
+        text: str,
+        request_id: str,
+        *,
+        source: str = "web",
+        expected_run_id: str | None = None,
+    ) -> dict:
+        session = self.repository.get(session_id)
+        if expected_run_id and session["active_run_id"] != expected_run_id:
+            raise ExplorationConflict("所选执行已结束，指导未发送，请刷新任务状态")
+        question = session.get("pending_question")
+        message = self.repository.add_message(
+            session_id,
+            text,
+            request_id,
+            source=source,
+            metadata={"question_id": question, "run_id": session["active_run_id"]},
+        )
+        question = message.get("question_id")
+        run_id = message.get("run_id")
+        if run_id and run_id == session["active_run_id"]:
+            run = self.repository.run(run_id)
+            if run and run.get("outcome") is None:
+                add_instruction(
+                    run_id, message["text"], source=source, instruction_id=message["id"]
+                )
+                self.repository.answer_question(session_id, question)
+                if read_state(run_id).get("status") == "task_paused":
+                    request_resume(run_id)
+        return message
+
+    def pause(self, session_id: str) -> dict:
+        session = self.repository.get(session_id)
+        if session["status"] == "finalizing":
+            raise ExplorationConflict("设备操作已结束，正在提炼用例，无需暂停。")
+        if not session["active_run_id"]:
+            raise ExplorationConflict("当前没有正在执行的探索")
+        request_pause(session["active_run_id"], reason="user")
+        return {"status": "pausing", "id": session_id}
+
+    def stop(self, session_id: str) -> dict:
+        from apps.admin_console.services.task_queue_service import TaskQueueService
+
+        session = self.repository.get(session_id)
+        if run_id := session["active_run_id"]:
+            requested = TaskQueueService.stop_tasks(session_id=run_id)
+            return {"status": "stopping" if requested else "not_running", "id": session_id}
+        return {"status": session["status"], "id": session_id}
+
+    async def resume(self, session_id: str, request_id: str, *, source: str = "web") -> dict:
+        repo = self.repository
+        with repo._transaction() as connection:
+            session = repo._require(connection, session_id)
+
+            def prepare():
+                if session.get("pending_question"):
+                    raise ExplorationConflict("请先回答探索助手的问题，再继续执行。")
+                if run_id := session["active_run_id"]:
+                    if read_state(run_id).get("status") != "task_paused":
+                        raise ExplorationConflict("请等待探索安全暂停后再继续")
+                    request_resume(run_id)
+                    return {"status": "resuming", "id": session_id, "run_id": run_id}
+                return {"status": "start", "id": session_id}
+
+            result = repo._operation(connection, f"{session_id}:continue", request_id, {}, prepare)
+        if result["status"] == "resuming":
+            return result
+        return await self.start(session_id, request_id, source=source)
+
+    def save_case(self, session_id: str, value: SaveCaseRequest) -> dict:
+        return self.repository.save_case(
+            session_id, value.case, value.expected_version, value.request_id
+        )
+
+    def demonstrate(self, session_id: str, request_id: str, source: str | None = None) -> dict:
+        session = self.repository.get(session_id)
+        if session["status"] == "finalizing":
+            raise ExplorationConflict("正在提炼用例，请完成后继续探索再示范。")
+        source = source or (
+            "computer"
+            if session["task_config"].get("execution_target") == "computer"
+            else "android"
+        )
+        if source not in {"android", "computer", "both"}:
+            raise ValueError("无效的示范来源")
+        demo = Demonstrations(self.repository).request(session_id, source, request_id)
+        if demo["status"] == "requested":
+            request_pause(demo["run_id"], reason="demonstration")
+            if read_state(demo["run_id"]).get("status") == "task_paused":
+                request_resume(demo["run_id"])
+        return demo
+
+    def demonstration_command(
+        self, session_id: str, demo_id: str, command: str, request_id: str, *, seconds: int = 0
+    ) -> dict:
+        return Demonstrations(self.repository).command(
+            session_id, demo_id, command, request_id, seconds=seconds
+        )
+
+    async def generate_case(self, session_id: str, value) -> dict:
+        from artemis.exploration.synthesis import generate_case, _generation
+        from apps.admin_console.services.automation_service import spawn
+
+        generation = _generation(
+            self.repository,
+            session_id,
+            value.request_id,
+            value.source_run_id,
+            value.expected_version,
+        )
+        if generation["status"] == "completed":
+            return {"status": "completed", "id": generation["id"], "case": generation["case"]}
+
+        async def work():
+            try:
+                await generate_case(
+                    self.repository,
+                    session_id,
+                    value.request_id,
+                    value.source_run_id,
+                    value.expected_version,
+                    generation=generation,
+                )
+            except Exception:
+                import logging
+
+                logging.getLogger(__name__).exception("Exploration synthesis request failed")
+
+        spawn(work())
+        return {"status": "running", "id": generation["id"]}
+
+    def create_case_batch(self, session_id: str, value) -> dict:
+        from apps.admin_console.routers.automation import BatchRequest
+        from apps.admin_console.services.automation_service import create_batch
+        from artemis.exploration.store import _digest
+
+        case = self.repository.case(session_id, value.version)
+        if not case:
+            raise KeyError("用例版本不存在")
+        session = self.repository.get(session_id)
+        options = {
+            k: v
+            for k, v in session["task_config"].items()
+            if k not in {"model", "reasoning_effort", "profile", "timeout_seconds"}
+        }
+        options.update(TaskOptions.model_validate(value.task_config).model_dump(exclude_none=True))
+        options["profile"] = value.profile
+        payload = BatchRequest(
+            name=value.name,
+            profile=value.profile,
+            request_id="explore-" + _digest([session_id, value.request_id]),
+            cases=[
+                {
+                    "name": case["name"],
+                    "goal": exploration_case_prompt(case),
+                    "preconditions": case["preconditions"],
+                    "case_id": f"{session_id}-v{value.version}",
+                    "task_config": options,
+                    "category": "探索用例",
+                    "subcategory": f"版本{value.version}",
+                }
+            ],
+        ).model_dump()
+        return create_batch(payload)
+
+
+exploration_service = ExplorationService()

apps/admin_console/services/ipc_service.py

@@ -20,10 +20,7 @@ from typing import Any
 
 logger = logging.getLogger(__name__)
 
-try:
-    from admin_console.core.state import state
-except ImportError:
-    from apps.admin_console.core.state import state
+from apps.admin_console.core.state import state
 
 
 class IPCService:

apps/admin_console/services/model_settings_service.py

@@ -80,6 +80,8 @@ def get_model_settings() -> dict:
         "tool_fallback_no_progress_attempts": settings.ARTEMIS_TOOL_FALLBACK_MAX_NO_PROGRESS_ATTEMPTS,
         "reasoning_effort": settings.OPENAI_REASONING_EFFORT,
         "model_presets": [
+            "gpt-6.1-sol",
+            "gpt-6-astra",
             "gpt-5.6-luna",
             "gpt-6-luna",
             "mimo-v2.6-flash",

apps/admin_console/services/report_export.py

@@ -156,6 +156,8 @@ def failure_reason(directory: Path, status: str, recorded: str | None) -> str:
 
 
 def session_report(session_id: str, db_path=None) -> dict:
+    from apps.admin_console.services.exploration_evidence import exploration_metadata
+
     identifier(session_id)
     meta = store.job(session_id) or {}
     repo = SessionRepository(db_path) if db_path else SessionRepository()
@@ -202,6 +204,7 @@ def session_report(session_id: str, db_path=None) -> dict:
         )
     return {
         "id": session_id,
+        "exploration": exploration_metadata(session_id),
         "name": execution_title(meta) or session.get("initial_goal", session_id)[:120],
         "iteration": meta.get("iteration", 1),
         "repeat_count": meta.get("repeat_count", 1),
@@ -237,6 +240,21 @@ def time_text(value) -> str:
 def metadata(report: dict) -> list[tuple]:
     return [
         ("任务编号", report["id"]),
+        *(
+            (
+                [
+                    ("探索会话", report["exploration"]["id"]),
+                    (
+                        "探索关联",
+                        "用例 v" + str(report["exploration"]["case_version"]) + " 复跑"
+                        if report["exploration"]["kind"] == "validation"
+                        else "探索执行（不等于独立复跑通过）",
+                    ),
+                ]
+            )
+            if report.get("exploration")
+            else []
+        ),
         ("用例编号", report.get("case_id") or "—"),
         ("检测项目", report.get("category") or "—"),
         ("测试子项", report.get("subcategory") or "—"),
@@ -840,7 +858,11 @@ def export_session_media(session_id: str, kind: str) -> Path:
         artifacts = session_artifacts(report) if kind == "all" else []
         extras = plugin_files(session_id) if kind in {"video", "all"} else []
         if kind == "video":
-            extras = [(path, name) for path, name in extras if path.suffix.lower() == ".mp4"]
+            extras = [
+                (path, name)
+                for path, name in extras
+                if path.suffix.lower() in {".mp4", ".mkv", ".webm"}
+            ]
         if kind == "video" and not videos and not extras:
             raise ValueError("此任务没有已保存的录屏")
         if kind == "screenshots" and not images:
@@ -854,7 +876,7 @@ def export_session_media(session_id: str, kind: str) -> Path:
             "verdict": report["verdict"],
             "screenshots": [name for _, name in images],
             "videos": ["任务录屏/" + name for _, name in videos]
-            + [name for path, name in extras if path.suffix.lower() == ".mp4"],
+            + [name for path, name in extras if path.suffix.lower() in {".mp4", ".mkv", ".webm"}],
             "artifacts": [name for _, name in artifacts],
             "note": "录屏按时间顺序分段；暂停期间未占用设备，也未录制其他任务。",
         }
@@ -874,7 +896,7 @@ def export_session_media(session_id: str, kind: str) -> Path:
                     source,
                     prefix + name,
                     compress_type=zipfile.ZIP_STORED
-                    if source.suffix.lower() == ".mp4"
+                    if source.suffix.lower() in {".mp4", ".mkv", ".webm"}
                     else zipfile.ZIP_DEFLATED,
                 )
             if kind == "all":
@@ -955,7 +977,7 @@ def export_batch(batch_id: str, progress: Callable[[int, int, str], None] | None
                         source,
                         f"{prefix}/{name}",
                         compress_type=zipfile.ZIP_STORED
-                        if source.suffix.lower() == ".mp4"
+                        if source.suffix.lower() in {".mp4", ".mkv", ".webm"}
                         else zipfile.ZIP_DEFLATED,
                     )
                 archive.writestr(

apps/admin_console/services/task_queue_service.py

@@ -25,16 +25,10 @@ import time
 from typing import Any
 import uuid
 
-try:
-    from admin_console.core.state import state
-    from admin_console.database.repositories.session_repository import session_repo
-    from admin_console.services import worker_process_io
-    from admin_console.services.media_service import media_service
-except ImportError:
-    from apps.admin_console.core.state import state
-    from apps.admin_console.database.repositories.session_repository import session_repo
-    from apps.admin_console.services import worker_process_io
-    from apps.admin_console.services.media_service import media_service
+from apps.admin_console.core.state import state
+from apps.admin_console.database.repositories.session_repository import session_repo
+from apps.admin_console.services import worker_process_io
+from apps.admin_console.services.media_service import media_service
 
 from artemis.config import (
     PAUSE_FILE,
@@ -585,6 +579,12 @@ class TaskQueueService:
             env["ARTEMIS_SESSION_ID"] = str(sess_id)
         env["ARTEMIS_TASK_INGRESS"] = str(task_item.get("ingress", "frontend"))
         env["ARTEMIS_TASK_WORKER"] = "1"
+        from artemis.exploration.store import explorations
+
+        exploration_run = explorations.run(str(sess_id)) if sess_id else None
+        env.pop("ARTEMIS_EXPLORATION_ID", None)
+        if exploration_run and exploration_run["kind"] == "exploration":
+            env["ARTEMIS_EXPLORATION_ID"] = exploration_run["exploration_id"]
         overrides = task_item.get("runtime_overrides") or {}
         if overrides.get("recording") is not None:
             env["ARTEMIS_TASK_RECORDING_CONFIG"] = json.dumps(overrides["recording"])
@@ -669,6 +669,11 @@ class TaskQueueService:
         batch = store.get("batch", job["batch_id"]) if job and job.get("batch_id") else None
         limit = task_timeout(task_item, job, batch)
         budget = ExecutionBudget(limit)
+        from artemis.exploration.store import explorations
+
+        exploration_run = explorations.run(sid)
+        is_exploration = bool(exploration_run and exploration_run["kind"] == "exploration")
+        exploration_finalized = False
         cleanup_limit = (task_item.get("runtime_overrides") or {}).get(
             "worker_cleanup_timeout_seconds"
         ) or settings.ARTEMIS_WORKER_CLEANUP_TIMEOUT_SECONDS
@@ -681,6 +686,9 @@ class TaskQueueService:
                     return waiter.result()
                 outcome = read_worker_outcome(sid)
                 if outcome:
+                    if exploration_run and not exploration_finalized:
+                        explorations.finish_run(sid, {**outcome, "status": outcome["task_status"]})
+                        exploration_finalized = True
                     finalized_at = finalized_at if finalized_at is not None else time.monotonic()
                     if time.monotonic() - finalized_at >= cleanup_limit:
                         logger.warning(
@@ -692,6 +700,9 @@ class TaskQueueService:
                         return await waiter
                     continue  # Final result is authoritative; only bound resource cleanup now.
                 cls._sync_control_states()
+                if is_exploration:
+                    # The worker owns its pausable budget so it survives a UI server restart.
+                    continue
                 run = state.active_runs.get(run_key, {})
                 remaining = budget.update(
                     excluded=bool(run.get("parked")) or read_state(sid).get("status") == "queued"
@@ -1077,9 +1088,35 @@ class TaskQueueService:
                         f"[QueueWorker] Could not persist failure status for [{sess_id}]"
                     )
         finally:
-            await cls._finish_output_forwarder(output_task)
-            # 5. Clean up the finished task and release this run's scheduling slot
-            cls._release_run_slot(sess_id, run_key, proc)
+            try:
+                await cls._finish_output_forwarder(output_task)
+                if sess_id:
+                    from artemis.exploration.store import explorations
+                    from artemis.runtime.task_result import read_worker_outcome
+
+                    exploration_run = explorations.run(str(sess_id))
+                    if exploration_run and exploration_run.get("outcome") is None:
+                        from artemis.runtime.case_result import read_case_result
+
+                        outcome = read_worker_outcome(str(sess_id)) or {}
+                        session = session_repo.get_session_by_id(str(sess_id)) or {}
+                        status = (
+                            outcome.get("task_status") or session.get("status") or "interrupted"
+                        )
+                        explorations.finish_run(
+                            str(sess_id),
+                            {
+                                **outcome,
+                                "status": status,
+                                "test_result": outcome.get("test_result")
+                                or read_case_result(str(sess_id), status),
+                            },
+                        )
+            except Exception:
+                logger.exception("Cannot finalize exploration metadata for %s", sess_id)
+            finally:
+                # Metadata/storage failure must never leak a scheduling slot.
+                cls._release_run_slot(sess_id, run_key, proc)
 
     @classmethod
     def _find_duplicate_submission(