chore: 迁移服务器前同步本地修改

- control_node 标注 DEPRECATED：保留目录壳子供未来远程探针节点复用，删除调用路径与相关测试
- 新增 task 表：极简元数据持久化 regulatory_node 完成的短任务（出报告/写文件/查询整理）
- regulatory_node 自标注：MessageResponse 扩展 task_action/title/summary，_run 末尾非阻塞落库
- query_task_list 改查 task 表，符合用户对"任务列表"的直觉，与 workflow 体系解耦
- 新增 /api/v1/task/list|/{id} 只读 API（task 由 regulatory 内部触发，不开放对外创建）

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

.gitignore

@@ -1,18 +1,25 @@
-# Python-generated files
-__pycache__/
-*.py[oc]
-build/
-dist/
-wheels/
-*.egg-info
-
-# Virtual environments
-.venv
-.idea
-# Local runtime data (MCP registry, etc.)
-data/
+# 项目运行时数据：默认全部忽略，仅显式开放需要纳入版本控制的子目录
+data/*
+!data/toolset/
 !data/plugin/
 data/plugin/skill/
-!data/toolset/
+
+# 插件运行时 SQLite / 状态
+data/plugin/*/_data/
+
 tmp/
 .env
+
+.idea/
+.venv/
+
+# Python
+__pycache__/
+*.pyc
+*.pyo
+
+# Node / Frontend
+node_modules/
+frontend/dist/
+data/plugin/*/frontend/dist/
+data/plugin/*/frontend/node_modules/

Dockerfile

@@ -2,12 +2,24 @@
 FROM node:22-alpine AS frontend-builder
 WORKDIR /app/frontend
 
-# Install dependencies and build the static assets
+# 主前端
 COPY frontend/package*.json ./
 RUN npm ci
 COPY frontend/ .
 RUN npm run build
 
+# 重型插件前端：每个插件独立 vite lib build，输出 dist/plugin-element.js + wc-manifest.json
+# 这一段通过 sh 循环：复制源码 → npm install → build。任何一个插件失败都会让镜像构建失败，
+# 用 || true 兜底过于宽松——这里选择硬失败，便于第一时间发现 build 问题。
+COPY data/plugin /app/data/plugin
+RUN set -e; \
+    for d in /app/data/plugin/*/frontend; do \
+      if [ -f "$d/package.json" ]; then \
+        echo "==> Building plugin frontend: $d"; \
+        cd "$d" && npm install && npm run build; \
+      fi; \
+    done
+
 # Stage 2: Build the Python backend and serve
 FROM python:3.13-slim
 WORKDIR /app
@@ -36,9 +48,11 @@ COPY . .
 
 # Copy the built frontend static assets from Stage 1
 COPY --from=frontend-builder /app/frontend/dist /app/frontend/dist
+# 重型插件前端 build 产物（让 /plugin-ui/<name>/ 静态挂载有内容可挂）
+COPY --from=frontend-builder /app/data/plugin /app/data/plugin
 
-# Expose FastAPI and Ray Dashboard ports
-EXPOSE 8000 8265
+# Expose FastAPI port
+EXPOSE 8000
 
 # Start the application
 CMD ["uv", "run", "python", "main.py"]

README-EN.md

@@ -147,7 +147,7 @@ KiloStar/
 │   │   ├── individual/              #   Agent node implementations
 │   │   │   ├── consciousness_node/  #     Task planning
 │   │   │   ├── regulatory_node/     #     Quality oversight
-│   │   │   ├── control_node/        #     Routing & dispatch
+│   │   │   ├── control_node/        #     Deprecated (name reserved for future remote-probe node)
 │   │   │   └── growth_node/         #     Capability expansion
 │   │   ├── work/                    #   Work execution layer
 │   │   │   ├── workflow/            #     Workflow engine (pydantic-graph)

alembic/versions/2026_06_17_0000-0010_provider_model_settings.py

@@ -0,0 +1,26 @@
+"""add model_settings JSONB column to provider table
+
+Revision ID: 0010
+Revises: 0009
+Create Date: 2026-06-17
+"""
+
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects.postgresql import JSONB
+
+revision = "0010"
+down_revision = "0009"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "provider",
+        sa.Column("model_settings", JSONB, nullable=True),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("provider", "model_settings")

alembic/versions/2026_06_17_0001-0011_task_table.py

@@ -0,0 +1,46 @@
+"""add task table for regulatory_node short tasks
+
+Revision ID: 0011
+Revises: 0010
+Create Date: 2026-06-17
+"""
+
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects.postgresql import JSONB
+
+revision = "0011"
+down_revision = "0010"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "task",
+        sa.Column("task_id", sa.String(64), primary_key=True),
+        sa.Column("user_id", sa.String(64), index=True, nullable=False),
+        sa.Column("chat_id", sa.String(64), index=True, nullable=True),
+        sa.Column("command", sa.Text(), nullable=False),
+        sa.Column("title", sa.String(255), nullable=False),
+        sa.Column(
+            "status", sa.String(20), index=True, server_default="completed"
+        ),
+        sa.Column("result_summary", sa.Text(), nullable=True),
+        sa.Column("artifact_refs", JSONB, nullable=True),
+        sa.Column(
+            "created_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.func.now(),
+            index=True,
+        ),
+        sa.Column(
+            "updated_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.func.now(),
+        ),
+    )
+
+
+def downgrade() -> None:
+    op.drop_table("task")

alembic/versions/2026_06_17_0002-0012_plugin_owned_agents.py

@@ -0,0 +1,32 @@
+"""add plugin_owned column to base_individual
+
+Revision ID: 0012
+Revises: 0011
+Create Date: 2026-06-17
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+
+revision = "0012"
+down_revision = "0011"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "base_individual",
+        sa.Column("plugin_owned", sa.String(64), nullable=True),
+    )
+    op.create_index(
+        "ix_base_individual_plugin_owned",
+        "base_individual",
+        ["plugin_owned"],
+    )
+
+
+def downgrade() -> None:
+    op.drop_index("ix_base_individual_plugin_owned", table_name="base_individual")
+    op.drop_column("base_individual", "plugin_owned")

data/plugin/data_analytics/__init__.py

@@ -0,0 +1 @@
+"""data_analytics 重型插件包。"""

data/plugin/data_analytics/agents.json

@@ -0,0 +1,16 @@
+{
+  "agents": [
+    {
+      "name": "analyst",
+      "role": "数据分析师",
+      "system_prompt": "你是一位严谨、克制的数据分析师。任务进来后：1) 先用 s3_list_objects/s3_peek 看几行了解结构；2) 决定用 python_executor（小数据，单机 pandas）或 ray_submit（大数据，分布式）；3) 执行分析、得出明确结论，必要时给出图表链接或样例数据。注意：你只能读取 S3，**不能写入**。如果用户让你上传/删除/修改对象，请明确告知做不到。",
+      "tools": ["s3_list_objects", "s3_peek", "s3_get_object", "ray_submit", "python_executor"],
+      "skills": [],
+      "peers": []
+    }
+  ],
+  "orchestration": {
+    "type": "react",
+    "entry": "analyst"
+  }
+}

data/plugin/data_analytics/api.py

@@ -0,0 +1,150 @@
+"""data_analytics 插件 API：凭证 CRUD + 分析任务提交/查询/事件流。
+
+挂载后路径前缀为 /api/v1/plugin/data_analytics/...，跟核心 API 完全独立。
+所有数据库读写都走 organization actor 的代理方法（确保分布式模式下不跨 actor
+共享 SQLAlchemy session）。
+"""
+
+from __future__ import annotations
+
+from typing import Optional
+
+from fastapi import APIRouter, Depends, HTTPException
+from fastapi.responses import StreamingResponse
+from pydantic import BaseModel, Field
+
+from kilostar.utils.access import Accessor, TokenData
+from kilostar.utils.actor import get_actor
+
+router = APIRouter(tags=["data_analytics"])
+
+
+# ─── Schemas ────────────────────────────────────────────────────────────────
+
+
+class CredentialCreate(BaseModel):
+    display_name: str = Field(..., max_length=100)
+    endpoint_url: Optional[str] = None
+    region: str = "us-east-1"
+    access_key: str = Field(..., min_length=1)
+    secret_key: str = Field(..., min_length=1)
+
+
+class JobCreate(BaseModel):
+    cred_id: str
+    description: str = Field(..., min_length=1, max_length=2000)
+
+
+# ─── Helpers ────────────────────────────────────────────────────────────────
+
+
+def _get_org():
+    try:
+        return get_actor("org_data_analytics")
+    except Exception as e:
+        raise HTTPException(503, f"data_analytics 插件未就绪：{e}")
+
+
+# ─── Credentials ────────────────────────────────────────────────────────────
+
+
+@router.get("/credentials")
+async def list_credentials(
+    token_data: TokenData = Depends(Accessor.get_current_user),
+):
+    org = _get_org()
+    rows = await org.cred_list(token_data.username)
+    return {"credentials": rows}
+
+
+@router.post("/credentials")
+async def create_credential(
+    payload: CredentialCreate,
+    token_data: TokenData = Depends(Accessor.get_current_user),
+):
+    org = _get_org()
+    row = await org.cred_create(
+        user_id=token_data.username,
+        display_name=payload.display_name,
+        access_key=payload.access_key,
+        secret_key=payload.secret_key,
+        endpoint_url=payload.endpoint_url,
+        region=payload.region,
+    )
+    return row
+
+
+@router.delete("/credentials/{cred_id}")
+async def delete_credential(
+    cred_id: str,
+    token_data: TokenData = Depends(Accessor.get_current_user),
+):
+    org = _get_org()
+    ok = await org.cred_delete(cred_id, token_data.username)
+    if not ok:
+        raise HTTPException(404, "凭证不存在或不属于当前用户")
+    return {"status": "ok"}
+
+
+# ─── Jobs ───────────────────────────────────────────────────────────────────
+
+
+@router.post("/jobs")
+async def create_job(
+    payload: JobCreate,
+    token_data: TokenData = Depends(Accessor.get_current_user),
+):
+    org = _get_org()
+    try:
+        return await org.job_create(
+            user_id=token_data.username,
+            cred_id=payload.cred_id,
+            description=payload.description,
+        )
+    except ValueError as e:
+        raise HTTPException(400, str(e))
+
+
+@router.get("/jobs")
+async def list_jobs(
+    token_data: TokenData = Depends(Accessor.get_current_user),
+):
+    org = _get_org()
+    rows = await org.job_list(token_data.username)
+    return {"jobs": rows}
+
+
+@router.get("/jobs/{job_id}")
+async def get_job(
+    job_id: str,
+    token_data: TokenData = Depends(Accessor.get_current_user),
+):
+    org = _get_org()
+    row = await org.job_get(job_id, token_data.username)
+    if row is None:
+        raise HTTPException(404, "任务不存在")
+    return row
+
+
+@router.get("/jobs/{job_id}/stream")
+async def stream_job(
+    job_id: str,
+    token_data: TokenData = Depends(Accessor.get_current_user),
+):
+    """转发 organization 事件流为 SSE。"""
+    import json
+
+    org = _get_org()
+    row = await org.job_get(job_id, token_data.username)
+    if row is None:
+        raise HTTPException(404, "任务不存在")
+    org_task_id = row.get("org_task_id")
+    if not org_task_id:
+        raise HTTPException(409, "任务尚未投递到 organization")
+
+    async def _generate():
+        async for event in await org.stream(org_task_id):
+            payload = event if isinstance(event, str) else json.dumps(event, ensure_ascii=False)
+            yield f"data: {payload}\n\n"
+
+    return StreamingResponse(_generate(), media_type="text/event-stream")

data/plugin/data_analytics/core/__init__.py

@@ -0,0 +1 @@
+"""data_analytics organization 实现。"""

data/plugin/data_analytics/core/db.py

@@ -0,0 +1,235 @@
+"""data_analytics 插件本地 SQLite 表与 DAO。
+
+注意：本插件用的 ``DeclarativeBase`` 跟核心 PG 完全独立，避免元数据空间串场。
+所有数据落到 ``data/plugin/data_analytics/_data/data_analytics.db``。
+"""
+
+from __future__ import annotations
+
+from datetime import datetime
+from typing import List, Optional
+
+from sqlalchemy import DateTime, String, Text, select
+from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker
+from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column
+
+from kilostar.utils.crypto import decrypt_dict_secrets, encrypt_dict_secrets
+
+
+class Base(DeclarativeBase):
+    """data_analytics 插件私有的元数据空间，跟核心 PG 隔离。"""
+
+    pass
+
+
+class S3Credential(Base):
+    __tablename__ = "s3_credential"
+
+    cred_id: Mapped[str] = mapped_column(String(64), primary_key=True)
+    user_id: Mapped[str] = mapped_column(String(64), index=True, nullable=False)
+    display_name: Mapped[str] = mapped_column(String(100), nullable=False)
+    endpoint_url: Mapped[Optional[str]] = mapped_column(String(255), nullable=True)
+    region: Mapped[str] = mapped_column(String(50), default="us-east-1")
+    access_key: Mapped[str] = mapped_column(String(255), nullable=False)
+    secret_key: Mapped[str] = mapped_column(String(255), nullable=False)
+    created_at: Mapped[datetime] = mapped_column(
+        DateTime, default=datetime.utcnow, nullable=False
+    )
+    updated_at: Mapped[datetime] = mapped_column(
+        DateTime, default=datetime.utcnow, onupdate=datetime.utcnow, nullable=False
+    )
+
+
+class AnalysisJob(Base):
+    __tablename__ = "analysis_job"
+
+    job_id: Mapped[str] = mapped_column(String(64), primary_key=True)
+    user_id: Mapped[str] = mapped_column(String(64), index=True, nullable=False)
+    cred_id: Mapped[Optional[str]] = mapped_column(String(64), nullable=True)
+    description: Mapped[str] = mapped_column(Text, nullable=False)
+    status: Mapped[str] = mapped_column(String(20), default="pending", index=True)
+    org_task_id: Mapped[Optional[str]] = mapped_column(String(64), nullable=True)
+    result: Mapped[Optional[str]] = mapped_column(Text, nullable=True)
+    created_at: Mapped[datetime] = mapped_column(
+        DateTime, default=datetime.utcnow, nullable=False, index=True
+    )
+    updated_at: Mapped[datetime] = mapped_column(
+        DateTime, default=datetime.utcnow, onupdate=datetime.utcnow, nullable=False
+    )
+
+
+class CredentialDAO:
+    """S3 凭证 DAO：写入时自动加密，读取时自动解密。"""
+
+    SENSITIVE_KEYS = ("access_key", "secret_key")
+
+    def __init__(self, sm: async_sessionmaker[AsyncSession]):
+        self._sm = sm
+
+    @staticmethod
+    def _row_to_dict(row: S3Credential, *, include_secrets: bool) -> dict:
+        d = {
+            "cred_id": row.cred_id,
+            "user_id": row.user_id,
+            "display_name": row.display_name,
+            "endpoint_url": row.endpoint_url,
+            "region": row.region,
+            "access_key": row.access_key,
+            "secret_key": row.secret_key,
+            "created_at": row.created_at.isoformat() if row.created_at else None,
+            "updated_at": row.updated_at.isoformat() if row.updated_at else None,
+        }
+        if not include_secrets:
+            ak = decrypt_dict_secrets({"access_key": d["access_key"]}).get("access_key", "")
+            d["access_key"] = (ak[:4] + "***" + ak[-2:]) if len(ak) > 6 else "***"
+            d.pop("secret_key", None)
+            return d
+        # include_secrets=True 用于工具内部，返回明文给 boto3
+        return decrypt_dict_secrets(d)
+
+    async def list_by_user(self, user_id: str) -> List[dict]:
+        async with self._sm() as s:
+            stmt = select(S3Credential).where(S3Credential.user_id == user_id)
+            rows = (await s.execute(stmt)).scalars().all()
+            return [self._row_to_dict(r, include_secrets=False) for r in rows]
+
+    async def get(self, cred_id: str, *, include_secrets: bool = False) -> Optional[dict]:
+        async with self._sm() as s:
+            stmt = select(S3Credential).where(S3Credential.cred_id == cred_id)
+            row = (await s.execute(stmt)).scalar_one_or_none()
+            if row is None:
+                return None
+            return self._row_to_dict(row, include_secrets=include_secrets)
+
+    async def upsert(
+        self,
+        cred_id: str,
+        user_id: str,
+        display_name: str,
+        access_key: str,
+        secret_key: str,
+        endpoint_url: Optional[str] = None,
+        region: str = "us-east-1",
+    ) -> dict:
+        encrypted = encrypt_dict_secrets(
+            {"access_key": access_key, "secret_key": secret_key}
+        )
+        async with self._sm() as s:
+            stmt = select(S3Credential).where(S3Credential.cred_id == cred_id)
+            existing = (await s.execute(stmt)).scalar_one_or_none()
+            if existing is not None:
+                existing.display_name = display_name
+                existing.endpoint_url = endpoint_url
+                existing.region = region
+                existing.access_key = encrypted["access_key"]
+                existing.secret_key = encrypted["secret_key"]
+                s.add(existing)
+                await s.commit()
+                await s.refresh(existing)
+                return self._row_to_dict(existing, include_secrets=False)
+            row = S3Credential(
+                cred_id=cred_id,
+                user_id=user_id,
+                display_name=display_name,
+                endpoint_url=endpoint_url,
+                region=region,
+                access_key=encrypted["access_key"],
+                secret_key=encrypted["secret_key"],
+            )
+            s.add(row)
+            await s.commit()
+            await s.refresh(row)
+            return self._row_to_dict(row, include_secrets=False)
+
+    async def delete(self, cred_id: str, user_id: str) -> bool:
+        async with self._sm() as s:
+            stmt = select(S3Credential).where(
+                S3Credential.cred_id == cred_id, S3Credential.user_id == user_id
+            )
+            row = (await s.execute(stmt)).scalar_one_or_none()
+            if row is None:
+                return False
+            await s.delete(row)
+            await s.commit()
+            return True
+
+
+class JobDAO:
+    """分析任务记录 DAO。"""
+
+    def __init__(self, sm: async_sessionmaker[AsyncSession]):
+        self._sm = sm
+
+    @staticmethod
+    def _row_to_dict(row: AnalysisJob) -> dict:
+        return {
+            "job_id": row.job_id,
+            "user_id": row.user_id,
+            "cred_id": row.cred_id,
+            "description": row.description,
+            "status": row.status,
+            "org_task_id": row.org_task_id,
+            "result": row.result,
+            "created_at": row.created_at.isoformat() if row.created_at else None,
+            "updated_at": row.updated_at.isoformat() if row.updated_at else None,
+        }
+
+    async def create(
+        self,
+        job_id: str,
+        user_id: str,
+        description: str,
+        cred_id: Optional[str] = None,
+    ) -> dict:
+        async with self._sm() as s:
+            row = AnalysisJob(
+                job_id=job_id,
+                user_id=user_id,
+                description=description,
+                cred_id=cred_id,
+            )
+            s.add(row)
+            await s.commit()
+            await s.refresh(row)
+            return self._row_to_dict(row)
+
+    async def update(
+        self,
+        job_id: str,
+        *,
+        status: Optional[str] = None,
+        result: Optional[str] = None,
+        org_task_id: Optional[str] = None,
+    ) -> Optional[dict]:
+        async with self._sm() as s:
+            stmt = select(AnalysisJob).where(AnalysisJob.job_id == job_id)
+            row = (await s.execute(stmt)).scalar_one_or_none()
+            if row is None:
+                return None
+            if status is not None:
+                row.status = status
+            if result is not None:
+                row.result = result
+            if org_task_id is not None:
+                row.org_task_id = org_task_id
+            s.add(row)
+            await s.commit()
+            await s.refresh(row)
+            return self._row_to_dict(row)
+
+    async def list_by_user(self, user_id: str, limit: int = 50) -> List[dict]:
+        async with self._sm() as s:
+            stmt = (
+                select(AnalysisJob)
+                .where(AnalysisJob.user_id == user_id)
+                .order_by(AnalysisJob.created_at.desc())
+                .limit(limit)
+            )
+            rows = (await s.execute(stmt)).scalars().all()
+            return [self._row_to_dict(r) for r in rows]
+
+    async def get(self, job_id: str) -> Optional[dict]:
+        async with self._sm() as s:
+            stmt = select(AnalysisJob).where(AnalysisJob.job_id == job_id)
+            row = (await s.execute(stmt)).scalar_one_or_none()
+            return self._row_to_dict(row) if row else None

data/plugin/data_analytics/core/organization.py

@@ -0,0 +1,135 @@
+"""data_analytics organization：管理本插件的 SQLite 元数据 + 注入凭证 ctx。
+
+凭证经由 ``S3_CREDS_VAR`` ContextVar 传给工具，避免污染 agent tool signature
+（agent 看到的工具不带 cred 参数，模型不会误传）。
+
+API 层通过本类暴露的 ``cred_*`` / ``job_*`` 代理方法跨 actor 调 DAO，
+保证分布式模式下 actor 之间不直接共享 SQLAlchemy session。
+"""
+
+from __future__ import annotations
+
+import contextvars
+import uuid
+from typing import Any, Callable, Dict, List, Optional
+
+from kilostar.plugin_runtime.base_organization import BaseOrganization
+from kilostar.plugin_runtime.event import OrgEvent
+
+from .db import Base, CredentialDAO, JobDAO
+
+# 当前任务的 S3 凭证（明文）：工具内部读 .get() 拿
+S3_CREDS_VAR: contextvars.ContextVar[Optional[Dict[str, Any]]] = contextvars.ContextVar(
+    "data_analytics_s3_creds", default=None
+)
+
+
+class DataAnalyticsOrganization(BaseOrganization):
+    """对接 S3 的数据分析组织。"""
+
+    async def setup(self) -> None:
+        await super().setup()
+        await self.init_local_db([Base])
+        # 跨工具/跨 API 共享的 DAO 实例
+        self.cred_dao = CredentialDAO(self._session_maker)
+        self.job_dao = JobDAO(self._session_maker)
+
+    async def on_first_install(self) -> None:
+        self.logger.info(
+            "data_analytics installed; configure S3 credentials in dashboard."
+        )
+
+    async def react(
+        self,
+        task_description: str,
+        ctx: Dict[str, Any],
+        emit: Callable[[OrgEvent], Any],
+    ) -> Any:
+        cred_id = ctx.get("cred_id")
+        if cred_id and getattr(self, "cred_dao", None) is not None:
+            cred = await self.cred_dao.get(cred_id, include_secrets=True)
+            if cred is None:
+                raise RuntimeError(f"S3 凭证 {cred_id} 不存在")
+            S3_CREDS_VAR.set(cred)
+            ctx["s3_cred_display"] = cred.get("display_name")
+        else:
+            S3_CREDS_VAR.set(None)
+        return await super().react(task_description, ctx, emit)
+
+    # ─── 凭证代理（API 层调用） ─────────────────────────────────────
+
+    async def cred_list(self, user_id: str) -> List[dict]:
+        return await self.cred_dao.list_by_user(user_id)
+
+    async def cred_create(
+        self,
+        user_id: str,
+        display_name: str,
+        access_key: str,
+        secret_key: str,
+        endpoint_url: Optional[str] = None,
+        region: str = "us-east-1",
+    ) -> dict:
+        cred_id = uuid.uuid4().hex
+        return await self.cred_dao.upsert(
+            cred_id=cred_id,
+            user_id=user_id,
+            display_name=display_name,
+            access_key=access_key,
+            secret_key=secret_key,
+            endpoint_url=endpoint_url,
+            region=region,
+        )
+
+    async def cred_delete(self, cred_id: str, user_id: str) -> bool:
+        return await self.cred_dao.delete(cred_id, user_id)
+
+    # ─── 任务代理 ──────────────────────────────────────────────────
+
+    async def job_create(
+        self, user_id: str, cred_id: str, description: str
+    ) -> dict:
+        # 校验凭证归属
+        cred = await self.cred_dao.get(cred_id, include_secrets=False)
+        if cred is None or cred.get("user_id") != user_id:
+            raise ValueError("凭证不存在或不属于当前用户")
+
+        job_id = uuid.uuid4().hex
+        await self.job_dao.create(
+            job_id=job_id,
+            user_id=user_id,
+            description=description,
+            cred_id=cred_id,
+        )
+        # 投递 organization 任务（拿 task_id 回填，便于前端拉事件流）
+        task_id = await self.submit(
+            description, {"user_id": user_id, "cred_id": cred_id, "job_id": job_id}
+        )
+        await self.job_dao.update(job_id, status="running", org_task_id=task_id)
+        return {"job_id": job_id, "task_id": task_id, "status": "running"}
+
+    async def job_list(self, user_id: str) -> List[dict]:
+        return await self.job_dao.list_by_user(user_id)
+
+    async def job_get(self, job_id: str, user_id: str) -> Optional[dict]:
+        row = await self.job_dao.get(job_id)
+        if row is None or row.get("user_id") != user_id:
+            return None
+        # 附带最新 organization 状态
+        org_task_id = row.get("org_task_id")
+        if org_task_id:
+            ts = await self.status(org_task_id)
+            if ts is not None:
+                row["task_status"] = ts.get("status")
+                row["task_result"] = ts.get("result")
+                row["task_error"] = ts.get("error")
+                # 任务终态时把结果回写 SQLite，方便重启后查询
+                if ts.get("status") in ("completed", "failed") and row.get("status") != ts.get("status"):
+                    result_payload = ts.get("result") if ts.get("status") == "completed" else ts.get("error")
+                    await self.job_dao.update(
+                        job_id,
+                        status=ts.get("status"),
+                        result=str(result_payload) if result_payload is not None else None,
+                    )
+                    row["status"] = ts.get("status")
+        return row

data/plugin/data_analytics/frontend/CredentialPanel.tsx

@@ -0,0 +1,174 @@
+import { useState } from 'react';
+import { Plus, Trash2, Loader2, Key, Eye, EyeOff } from 'lucide-react';
+import { usePluginContext } from './client';
+import type { S3Credential } from './types';
+
+const API_BASE = '/api/v1/plugin/data_analytics';
+
+interface Props {
+  credentials: S3Credential[];
+  loading: boolean;
+  onChanged: () => void;
+}
+
+export function CredentialPanel({ credentials, loading, onChanged }: Props) {
+  const { client } = usePluginContext();
+  const [showForm, setShowForm] = useState(false);
+  const [showSecret, setShowSecret] = useState(false);
+  const [busy, setBusy] = useState(false);
+  const [error, setError] = useState('');
+  const [form, setForm] = useState({
+    display_name: '',
+    endpoint_url: '',
+    region: 'us-east-1',
+    access_key: '',
+    secret_key: '',
+  });
+
+  const reset = () => {
+    setForm({ display_name: '', endpoint_url: '', region: 'us-east-1', access_key: '', secret_key: '' });
+    setError('');
+    setShowSecret(false);
+  };
+
+  const submit = async () => {
+    if (!form.display_name.trim() || !form.access_key.trim() || !form.secret_key.trim()) {
+      setError('显示名 / Access Key / Secret Key 必填');
+      return;
+    }
+    setBusy(true);
+    setError('');
+    try {
+      await client.post(`${API_BASE}/credentials`, {
+        display_name: form.display_name.trim(),
+        endpoint_url: form.endpoint_url.trim() || null,
+        region: form.region.trim() || 'us-east-1',
+        access_key: form.access_key,
+        secret_key: form.secret_key,
+      });
+      reset();
+      setShowForm(false);
+      onChanged();
+    } catch (e: unknown) {
+      const msg = (e as { response?: { data?: { detail?: string } } }).response?.data?.detail;
+      setError(msg || '保存失败');
+    } finally {
+      setBusy(false);
+    }
+  };
+
+  const remove = async (cred_id: string) => {
+    if (!confirm('确定删除该凭证？删除后该凭证下的任务将无法继续运行。')) return;
+    try {
+      await client.delete(`${API_BASE}/credentials/${cred_id}`);
+      onChanged();
+    } catch (e) {
+      console.error(e);
+    }
+  };
+
+  return (
+    <div className="bg-bg-card rounded-2xl border border-border-primary p-5">
+      <div className="flex items-center justify-between mb-4">
+        <div>
+          <h3 className="font-semibold text-text-primary flex items-center gap-2">
+            <Key size={16} className="text-accent" /> S3 凭证
+          </h3>
+          <p className="text-xs text-text-muted mt-0.5">访问密钥加密存储于本地 SQLite。</p>
+        </div>
+        <button
+          className="px-3 py-1.5 text-xs rounded-lg bg-accent text-white hover:opacity-90 transition flex items-center gap-1.5"
+          onClick={() => { setShowForm((s) => !s); setError(''); }}
+        >
+          <Plus size={14} /> {showForm ? '取消' : '新增'}
+        </button>
+      </div>
+
+      {showForm && (
+        <div className="space-y-3 mb-4 p-4 bg-bg-secondary rounded-xl border border-border-secondary">
+          <input
+            className="w-full px-3 py-2 text-sm rounded-lg bg-bg-base border border-border-primary focus:outline-none focus:border-accent"
+            placeholder="显示名（如 prod-aws）"
+            value={form.display_name}
+            onChange={(e) => setForm({ ...form, display_name: e.target.value })}
+          />
+          <input
+            className="w-full px-3 py-2 text-sm rounded-lg bg-bg-base border border-border-primary focus:outline-none focus:border-accent"
+            placeholder="Endpoint URL（可选，自托管 S3 / MinIO 填写）"
+            value={form.endpoint_url}
+            onChange={(e) => setForm({ ...form, endpoint_url: e.target.value })}
+          />
+          <input
+            className="w-full px-3 py-2 text-sm rounded-lg bg-bg-base border border-border-primary focus:outline-none focus:border-accent"
+            placeholder="Region（默认 us-east-1）"
+            value={form.region}
+            onChange={(e) => setForm({ ...form, region: e.target.value })}
+          />
+          <input
+            className="w-full px-3 py-2 text-sm font-mono rounded-lg bg-bg-base border border-border-primary focus:outline-none focus:border-accent"
+            placeholder="Access Key"
+            value={form.access_key}
+            onChange={(e) => setForm({ ...form, access_key: e.target.value })}
+          />
+          <div className="relative">
+            <input
+              type={showSecret ? 'text' : 'password'}
+              className="w-full px-3 py-2 pr-10 text-sm font-mono rounded-lg bg-bg-base border border-border-primary focus:outline-none focus:border-accent"
+              placeholder="Secret Key"
+              value={form.secret_key}
+              onChange={(e) => setForm({ ...form, secret_key: e.target.value })}
+            />
+            <button
+              type="button"
+              className="absolute right-2 top-1/2 -translate-y-1/2 text-text-muted hover:text-text-primary"
+              onClick={() => setShowSecret((s) => !s)}
+            >
+              {showSecret ? <EyeOff size={14} /> : <Eye size={14} />}
+            </button>
+          </div>
+          {error && <div className="text-xs text-danger">{error}</div>}
+          <button
+            className="w-full px-3 py-2 text-sm rounded-lg bg-accent text-white hover:opacity-90 disabled:opacity-50 transition flex items-center justify-center gap-2"
+            onClick={submit}
+            disabled={busy}
+          >
+            {busy && <Loader2 size={14} className="animate-spin" />} 保存
+          </button>
+        </div>
+      )}
+
+      {loading ? (
+        <div className="flex items-center justify-center py-8 text-text-muted">
+          <Loader2 size={20} className="animate-spin" />
+        </div>
+      ) : credentials.length === 0 ? (
+        <div className="text-sm text-text-muted text-center py-8 border border-dashed border-border-primary rounded-xl">
+          还没有凭证，点右上角「新增」开始。
+        </div>
+      ) : (
+        <div className="space-y-2">
+          {credentials.map((c) => (
+            <div
+              key={c.cred_id}
+              className="flex items-center justify-between p-3 bg-bg-secondary rounded-xl border border-border-secondary"
+            >
+              <div className="min-w-0 flex-1">
+                <div className="text-sm font-medium text-text-primary truncate">{c.display_name}</div>
+                <div className="text-[11px] text-text-muted font-mono mt-0.5">
+                  {c.endpoint_url || 'aws-s3'} · {c.region} · {c.access_key}
+                </div>
+              </div>
+              <button
+                className="p-1.5 text-text-muted hover:text-danger transition"
+                onClick={() => remove(c.cred_id)}
+                title="删除"
+              >
+                <Trash2 size={14} />
+              </button>
+            </div>
+          ))}
+        </div>
+      )}
+    </div>
+  );
+}

data/plugin/data_analytics/frontend/Dashboard.tsx

@@ -0,0 +1,157 @@
+import { useCallback, useEffect, useState } from 'react';
+import { BarChart3, Plus, Loader2, ListChecks } from 'lucide-react';
+import { usePluginContext } from './client';
+import { CredentialPanel } from './CredentialPanel';
+import { NewJobDialog } from './NewJobDialog';
+import { JobDetail } from './JobDetail';
+import type { S3Credential, AnalysisJob } from './types';
+
+const API_BASE = '/api/v1/plugin/data_analytics';
+
+interface Props {
+  pluginName: string;
+}
+
+export function Dashboard({ pluginName }: Props) {
+  const { client } = usePluginContext();
+  const [credentials, setCredentials] = useState<S3Credential[]>([]);
+  const [credLoading, setCredLoading] = useState(true);
+  const [jobs, setJobs] = useState<AnalysisJob[]>([]);
+  const [jobLoading, setJobLoading] = useState(true);
+  const [showNewJob, setShowNewJob] = useState(false);
+  const [openJobId, setOpenJobId] = useState<string | null>(null);
+  const [error, setError] = useState('');
+
+  const loadCredentials = useCallback(async () => {
+    setCredLoading(true);
+    try {
+      const resp = await client.get<{ credentials: S3Credential[] }>(`${API_BASE}/credentials`);
+      setCredentials(resp.data.credentials || []);
+    } catch (e: unknown) {
+      const msg = (e as { response?: { data?: { detail?: string } } }).response?.data?.detail;
+      setError(msg || '加载凭证失败');
+    } finally {
+      setCredLoading(false);
+    }
+  }, [client]);
+
+  const loadJobs = useCallback(async () => {
+    setJobLoading(true);
+    try {
+      const resp = await client.get<{ jobs: AnalysisJob[] }>(`${API_BASE}/jobs`);
+      setJobs(resp.data.jobs || []);
+    } catch (e: unknown) {
+      const msg = (e as { response?: { data?: { detail?: string } } }).response?.data?.detail;
+      setError(msg || '加载任务失败');
+    } finally {
+      setJobLoading(false);
+    }
+  }, [client]);
+
+  useEffect(() => {
+    loadCredentials();
+    loadJobs();
+  }, [loadCredentials, loadJobs]);
+
+  // 轮询任务列表，方便看状态变化
+  useEffect(() => {
+    const t = setInterval(loadJobs, 5000);
+    return () => clearInterval(t);
+  }, [loadJobs]);
+
+  return (
+    <div className="h-full overflow-y-auto p-6 bg-bg-base">
+      <div className="max-w-5xl mx-auto space-y-6">
+        <div className="flex items-center gap-3 pb-2">
+          <div className="w-10 h-10 rounded-xl bg-accent-light text-accent flex items-center justify-center">
+            <BarChart3 size={20} />
+          </div>
+          <div>
+            <h2 className="text-lg font-bold text-text-primary">数据分析</h2>
+            <p className="text-xs text-text-muted mt-0.5">
+              对接 S3，让 agent 自主决定分析路径（python_executor / ray_submit）。{' '}
+              <span className="font-mono text-[10px] opacity-70">{pluginName}</span>
+            </p>
+          </div>
+        </div>
+
+        {error && (
+          <div className="p-3 bg-danger-bg text-danger text-sm rounded-xl border border-danger/20">{error}</div>
+        )}
+
+        <CredentialPanel credentials={credentials} loading={credLoading} onChanged={loadCredentials} />
+
+        <div className="bg-bg-card rounded-2xl border border-border-primary p-5">
+          <div className="flex items-center justify-between mb-4">
+            <div>
+              <h3 className="font-semibold text-text-primary flex items-center gap-2">
+                <ListChecks size={16} className="text-accent" /> 分析任务
+              </h3>
+              <p className="text-xs text-text-muted mt-0.5">点行查看详情和事件流。每 5 秒自动刷新。</p>
+            </div>
+            <button
+              className="px-3 py-1.5 text-xs rounded-lg bg-accent text-white hover:opacity-90 transition flex items-center gap-1.5 disabled:opacity-50"
+              onClick={() => setShowNewJob(true)}
+              disabled={credentials.length === 0}
+              title={credentials.length === 0 ? '请先添加凭证' : ''}
+            >
+              <Plus size={14} /> 新建任务
+            </button>
+          </div>
+
+          {jobLoading && jobs.length === 0 ? (
+            <div className="flex items-center justify-center py-8 text-text-muted">
+              <Loader2 size={20} className="animate-spin" />
+            </div>
+          ) : jobs.length === 0 ? (
+            <div className="text-sm text-text-muted text-center py-8 border border-dashed border-border-primary rounded-xl">
+              还没有分析任务。点右上角「新建任务」开始。
+            </div>
+          ) : (
+            <div className="space-y-2">
+              {jobs.map((j) => (
+                <button
+                  key={j.job_id}
+                  className="w-full text-left p-3 bg-bg-secondary rounded-xl border border-border-secondary hover:border-accent transition flex items-center justify-between gap-3"
+                  onClick={() => setOpenJobId(j.job_id)}
+                >
+                  <div className="min-w-0 flex-1">
+                    <div className="text-sm text-text-primary truncate">{j.description}</div>
+                    <div className="text-[11px] text-text-muted mt-0.5 font-mono">
+                      {j.job_id.slice(0, 8)} · {j.created_at?.slice(0, 19).replace('T', ' ')}
+                    </div>
+                  </div>
+                  <StatusBadge status={j.status} />
+                </button>
+              ))}
+            </div>
+          )}
+        </div>
+      </div>
+
+      {showNewJob && (
+        <NewJobDialog
+          credentials={credentials}
+          onClose={() => setShowNewJob(false)}
+          onCreated={loadJobs}
+        />
+      )}
+      {openJobId && <JobDetail jobId={openJobId} onClose={() => setOpenJobId(null)} />}
+    </div>
+  );
+}
+
+function StatusBadge({ status }: { status: string }) {
+  const map: Record<string, string> = {
+    pending: 'bg-bg-base text-text-muted border-border-primary',
+    running: 'bg-warning-bg text-warning border-warning/20',
+    completed: 'bg-success-bg text-success border-success/20',
+    failed: 'bg-danger-bg text-danger border-danger/20',
+  };
+  const cls = map[status] || map.pending;
+  return (
+    <span className={`text-[10px] font-medium px-2 py-1 rounded-lg border ${cls} shrink-0`}>
+      {status}
+    </span>
+  );
+}

data/plugin/data_analytics/frontend/JobDetail.tsx

@@ -0,0 +1,174 @@
+import { useEffect, useRef, useState } from 'react';
+import { Loader2, X, Activity } from 'lucide-react';
+import { usePluginContext } from './client';
+import type { AnalysisJob } from './types';
+
+const API_BASE = '/api/v1/plugin/data_analytics';
+
+interface Props {
+  jobId: string;
+  onClose: () => void;
+}
+
+interface StreamEvent {
+  type?: string;
+  ts?: number;
+  payload?: unknown;
+  raw?: string;
+}
+
+export function JobDetail({ jobId, onClose }: Props) {
+  const { client, token, apiBase } = usePluginContext();
+  const [job, setJob] = useState<AnalysisJob | null>(null);
+  const [events, setEvents] = useState<StreamEvent[]>([]);
+  const [loading, setLoading] = useState(true);
+  const eventBoxRef = useRef<HTMLDivElement | null>(null);
+
+  // 初次加载 + 后台轮询
+  useEffect(() => {
+    let cancelled = false;
+    const fetchJob = async () => {
+      try {
+        const resp = await client.get<AnalysisJob>(`${API_BASE}/jobs/${jobId}`);
+        if (!cancelled) setJob(resp.data);
+      } catch (e) {
+        console.error('fetch job failed', e);
+      } finally {
+        if (!cancelled) setLoading(false);
+      }
+    };
+    fetchJob();
+    const t = setInterval(fetchJob, 4000);
+    return () => {
+      cancelled = true;
+      clearInterval(t);
+    };
+  }, [client, jobId]);
+
+  // SSE 事件流（用 fetch + ReadableStream，因为 EventSource 不支持自定义 header）
+  useEffect(() => {
+    const controller = new AbortController();
+    const run = async () => {
+      try {
+        const url = `${apiBase || ''}${API_BASE}/jobs/${jobId}/stream`;
+        const resp = await fetch(url, {
+          headers: { Authorization: `Bearer ${token}` },
+          signal: controller.signal,
+        });
+        if (!resp.body) return;
+        const reader = resp.body.getReader();
+        const decoder = new TextDecoder('utf-8');
+        let buf = '';
+        // eslint-disable-next-line no-constant-condition
+        while (true) {
+          const { value, done } = await reader.read();
+          if (done) break;
+          buf += decoder.decode(value, { stream: true });
+          const parts = buf.split('\n\n');
+          buf = parts.pop() || '';
+          for (const part of parts) {
+            const line = part.split('\n').find((l) => l.startsWith('data:'));
+            if (!line) continue;
+            const payload = line.slice(5).trim();
+            try {
+              setEvents((prev) => [...prev, JSON.parse(payload)]);
+            } catch {
+              setEvents((prev) => [...prev, { raw: payload }]);
+            }
+          }
+        }
+      } catch (e) {
+        if ((e as Error).name !== 'AbortError') console.error('SSE error', e);
+      }
+    };
+    run();
+    return () => controller.abort();
+  }, [apiBase, jobId, token]);
+
+  // 自动滚动到底部
+  useEffect(() => {
+    if (eventBoxRef.current) {
+      eventBoxRef.current.scrollTop = eventBoxRef.current.scrollHeight;
+    }
+  }, [events]);
+
+  return (
+    <div className="fixed inset-0 z-50 flex items-center justify-center bg-black/40 p-4">
+      <div className="w-full max-w-3xl max-h-[85vh] bg-bg-card rounded-2xl border border-border-primary shadow-xl flex flex-col">
+        <div className="flex items-center justify-between p-4 border-b border-border-primary">
+          <div className="min-w-0">
+            <h3 className="font-semibold text-text-primary flex items-center gap-2">
+              <Activity size={16} className="text-accent" />
+              任务详情
+            </h3>
+            <span className="text-[11px] font-mono text-text-muted">{jobId}</span>
+          </div>
+          <button className="p-1 text-text-muted hover:text-text-primary" onClick={onClose}>
+            <X size={16} />
+          </button>
+        </div>
+
+        <div className="flex-1 min-h-0 overflow-y-auto p-5 space-y-4">
+          {loading ? (
+            <div className="flex items-center justify-center py-12 text-text-muted">
+              <Loader2 size={20} className="animate-spin" />
+            </div>
+          ) : job ? (
+            <>
+              <div className="space-y-2">
+                <Field label="状态" value={job.task_status || job.status} />
+                <Field label="描述" value={job.description} />
+                {job.task_error && <Field label="错误" value={job.task_error} danger />}
+              </div>
+
+              {job.task_result !== undefined && job.task_result !== null && (
+                <div>
+                  <div className="text-xs text-text-secondary mb-1.5">执行结果</div>
+                  <pre className="text-xs font-mono whitespace-pre-wrap break-words bg-bg-secondary border border-border-secondary rounded-lg p-3 max-h-64 overflow-auto">
+                    {typeof job.task_result === 'string'
+                      ? job.task_result
+                      : JSON.stringify(job.task_result, null, 2)}
+                  </pre>
+                </div>
+              )}
+
+              <div>
+                <div className="text-xs text-text-secondary mb-1.5">事件流（SSE）</div>
+                <div
+                  ref={eventBoxRef}
+                  className="text-[11px] font-mono bg-bg-secondary border border-border-secondary rounded-lg p-3 max-h-72 overflow-auto space-y-1"
+                >
+                  {events.length === 0 ? (
+                    <span className="text-text-muted">(等待事件…)</span>
+                  ) : (
+                    events.map((e, i) => (
+                      <div key={i} className="text-text-secondary">
+                        <span className="text-accent">{e.type || 'event'}</span>{' '}
+                        {e.payload !== undefined ? (
+                          <span>{JSON.stringify(e.payload)}</span>
+                        ) : e.raw ? (
+                          <span>{e.raw}</span>
+                        ) : null}
+                      </div>
+                    ))
+                  )}
+                </div>
+              </div>
+            </>
+          ) : (
+            <div className="text-sm text-text-muted text-center py-8">任务不存在或已被删除</div>
+          )}
+        </div>
+      </div>
+    </div>
+  );
+}
+
+function Field({ label, value, danger }: { label: string; value: string; danger?: boolean }) {
+  return (
+    <div className="flex gap-3">
+      <div className="text-xs text-text-muted w-16 shrink-0 pt-0.5">{label}</div>
+      <div className={`text-sm flex-1 break-words ${danger ? 'text-danger' : 'text-text-primary'}`}>{value}</div>
+    </div>
+  );
+}

data/plugin/data_analytics/frontend/NewJobDialog.tsx

@@ -0,0 +1,110 @@
+import { useState } from 'react';
+import { Loader2, Send, X } from 'lucide-react';
+import { usePluginContext } from './client';
+import type { S3Credential } from './types';
+
+const API_BASE = '/api/v1/plugin/data_analytics';
+
+interface Props {
+  credentials: S3Credential[];
+  onClose: () => void;
+  onCreated: () => void;
+}
+
+export function NewJobDialog({ credentials, onClose, onCreated }: Props) {
+  const { client } = usePluginContext();
+  const [credId, setCredId] = useState(credentials[0]?.cred_id || '');
+  const [description, setDescription] = useState('');
+  const [busy, setBusy] = useState(false);
+  const [error, setError] = useState('');
+
+  const submit = async () => {
+    if (!credId) {
+      setError('请选择 S3 凭证');
+      return;
+    }
+    if (!description.trim()) {
+      setError('请描述要做的分析');
+      return;
+    }
+    setBusy(true);
+    setError('');
+    try {
+      await client.post(`${API_BASE}/jobs`, {
+        cred_id: credId,
+        description: description.trim(),
+      });
+      onCreated();
+      onClose();
+    } catch (e: unknown) {
+      const msg = (e as { response?: { data?: { detail?: string } } }).response?.data?.detail;
+      setError(msg || '提交失败');
+    } finally {
+      setBusy(false);
+    }
+  };
+
+  return (
+    <div className="fixed inset-0 z-50 flex items-center justify-center bg-black/40 p-4">
+      <div className="w-full max-w-lg bg-bg-card rounded-2xl border border-border-primary shadow-xl">
+        <div className="flex items-center justify-between p-4 border-b border-border-primary">
+          <h3 className="font-semibold text-text-primary">新建分析任务</h3>
+          <button className="p-1 text-text-muted hover:text-text-primary" onClick={onClose}>
+            <X size={16} />
+          </button>
+        </div>
+        <div className="p-5 space-y-4">
+          <div>
+            <label className="text-xs text-text-secondary block mb-1.5">S3 凭证</label>
+            {credentials.length === 0 ? (
+              <div className="text-xs text-warning bg-warning-bg/50 border border-warning/20 rounded-lg p-2">
+                请先在上方添加 S3 凭证。
+              </div>
+            ) : (
+              <select
+                className="w-full px-3 py-2 text-sm rounded-lg bg-bg-base border border-border-primary focus:outline-none focus:border-accent"
+                value={credId}
+                onChange={(e) => setCredId(e.target.value)}
+              >
+                {credentials.map((c) => (
+                  <option key={c.cred_id} value={c.cred_id}>
+                    {c.display_name} · {c.region}
+                  </option>
+                ))}
+              </select>
+            )}
+          </div>
+          <div>
+            <label className="text-xs text-text-secondary block mb-1.5">任务描述</label>
+            <textarea
+              className="w-full px-3 py-2 text-sm rounded-lg bg-bg-base border border-border-primary focus:outline-none focus:border-accent min-h-[120px] resize-y"
+              placeholder="例如：分析 s3://my-bucket/sales/2026-q1/ 的销售趋势，输出按月汇总"
+              value={description}
+              onChange={(e) => setDescription(e.target.value)}
+            />
+            <p className="text-[11px] text-text-muted mt-1">
+              Agent 会先用 s3_peek/s3_list_objects 探查数据，然后选择 python_executor 或 ray_submit 执行分析。
+            </p>
+          </div>
+          {error && <div className="text-xs text-danger">{error}</div>}
+        </div>
+        <div className="flex items-center justify-end gap-2 p-4 border-t border-border-primary">
+          <button
+            className="px-3 py-1.5 text-xs rounded-lg border border-border-primary text-text-secondary hover:text-text-primary"
+            onClick={onClose}
+          >
+            取消
+          </button>
+          <button
+            className="px-3 py-1.5 text-xs rounded-lg bg-accent text-white hover:opacity-90 disabled:opacity-50 transition flex items-center gap-1.5"
+            onClick={submit}
+            disabled={busy || credentials.length === 0}
+          >
+            {busy ? <Loader2 size={14} className="animate-spin" /> : <Send size={14} />}
+            提交
+          </button>
+        </div>
+      </div>
+    </div>
+  );
+}

data/plugin/data_analytics/frontend/build-manifest.mjs

@@ -0,0 +1,22 @@
+import { readdirSync, readFileSync, writeFileSync, existsSync } from 'node:fs';
+import { join } from 'node:path';
+
+// vite lib 模式 build 完后写一份 wc-manifest.json，给后端 /ui-manifest 端点读
+const distDir = 'dist';
+if (!existsSync(distDir)) {
+  console.error('dist/ not found; run vite build first');
+  process.exit(1);
+}
+
+const files = readdirSync(distDir);
+const js = files.find((f) => f.endsWith('.js')) || 'plugin-element.js';
+const css = files.filter((f) => f.endsWith('.css'));
+
+const manifest = {
+  tag: 'plugin-data-analytics',
+  js,
+  css,
+};
+
+writeFileSync(join(distDir, 'wc-manifest.json'), JSON.stringify(manifest, null, 2) + '\n', 'utf-8');
+console.log(`wrote dist/wc-manifest.json: ${JSON.stringify(manifest)}`);

data/plugin/data_analytics/frontend/client.ts

@@ -0,0 +1,30 @@
+import axios, { type AxiosInstance } from 'axios';
+import { createContext, useContext } from 'react';
+
+export interface PluginContextValue {
+  client: AxiosInstance;
+  token: string;
+  apiBase: string;
+}
+
+export const PluginContext = createContext<PluginContextValue | null>(null);
+
+export function usePluginContext(): PluginContextValue {
+  const ctx = useContext(PluginContext);
+  if (!ctx) throw new Error('PluginContext missing — Web Component not initialized');
+  return ctx;
+}
+
+export function makeClient(token: string, apiBase: string): AxiosInstance {
+  const c = axios.create({
+    baseURL: apiBase || undefined,
+  });
+  c.interceptors.request.use((cfg) => {
+    if (token) {
+      cfg.headers = cfg.headers || {};
+      (cfg.headers as Record<string, string>).Authorization = `Bearer ${token}`;
+    }
+    return cfg;
+  });
+  return c;
+}

data/plugin/data_analytics/frontend/index.tsx

@@ -0,0 +1,61 @@
+import React from 'react';
+import { createRoot, type Root } from 'react-dom/client';
+import { Dashboard } from './Dashboard';
+import { PluginContext, makeClient } from './client';
+
+// 把 build 出来的 CSS 当字符串收入，作为 ConstructableStyleSheet 注入到 shadow root，
+// 既能享受 shadow DOM 的样式隔离，也不需要额外的 fetch 步骤。
+import css from './styles.css?inline';
+
+const TAG = 'plugin-data-analytics';
+
+class DataAnalyticsElement extends HTMLElement {
+  private root?: Root;
+  private mount?: HTMLDivElement;
+
+  static get observedAttributes() {
+    return ['token', 'api-base'];
+  }
+
+  connectedCallback() {
+    const shadow = this.attachShadow({ mode: 'open' });
+
+    // 用 <style> 注入 CSS（ConstructableStyleSheet 兼容性更好但 vite 注入字符串更直接）
+    const style = document.createElement('style');
+    style.textContent = css;
+    shadow.appendChild(style);
+
+    this.mount = document.createElement('div');
+    this.mount.style.cssText = 'height:100%;width:100%';
+    shadow.appendChild(this.mount);
+
+    this.root = createRoot(this.mount);
+    this.render();
+  }
+
+  attributeChangedCallback() {
+    if (this.root) this.render();
+  }
+
+  disconnectedCallback() {
+    this.root?.unmount();
+    this.root = undefined;
+  }
+
+  private render() {
+    const token = this.getAttribute('token') ?? '';
+    const apiBase = this.getAttribute('api-base') ?? '';
+    const client = makeClient(token, apiBase);
+    this.root!.render(
+      <React.StrictMode>
+        <PluginContext.Provider value={{ client, token, apiBase }}>
+          <Dashboard pluginName="data_analytics" />
+        </PluginContext.Provider>
+      </React.StrictMode>,
+    );
+  }
+}
+
+if (!customElements.get(TAG)) {
+  customElements.define(TAG, DataAnalyticsElement);
+}

data/plugin/data_analytics/frontend/package.json

@@ -0,0 +1,25 @@
+{
+  "name": "plugin-data-analytics",
+  "private": true,
+  "version": "0.1.0",
+  "type": "module",
+  "scripts": {
+    "build": "vite build && node build-manifest.mjs",
+    "dev": "vite build --watch"
+  },
+  "dependencies": {
+    "axios": "^1.15.1",
+    "lucide-react": "^1.8.0",
+    "react": "^19.2.4",
+    "react-dom": "^19.2.4"
+  },
+  "devDependencies": {
+    "@tailwindcss/vite": "^4.2.2",
+    "@types/react": "^19.1.0",
+    "@types/react-dom": "^19.1.0",
+    "@vitejs/plugin-react": "^6.0.1",
+    "tailwindcss": "^4.2.2",
+    "typescript": "^5.8.0",
+    "vite": "^8.0.4"
+  }
+}

data/plugin/data_analytics/frontend/styles.css

@@ -0,0 +1,91 @@
+@import "tailwindcss";
+
+/* 在 shadow DOM 内 :root 不匹配，用 :host 给 Web Component 自身定义主题 token。
+   颜色名称跟主前端 frontend/src/index.css 保持一致——这样组件里的 bg-bg-card / text-accent
+   等类名在插件 build 时也能解析到对应的 var()。 */
+@theme {
+  --color-bg-primary: var(--bg-primary);
+  --color-bg-secondary: var(--bg-secondary);
+  --color-bg-tertiary: var(--bg-tertiary);
+  --color-bg-card: var(--bg-card);
+  --color-bg-sidebar: var(--bg-sidebar);
+  --color-bg-input: var(--bg-input);
+  --color-bg-hover: var(--bg-hover);
+  --color-bg-active: var(--bg-active);
+  --color-bg-base: var(--bg-base);
+  --color-border-primary: var(--border-primary);
+  --color-border-secondary: var(--border-secondary);
+  --color-text-primary: var(--text-primary);
+  --color-text-secondary: var(--text-secondary);
+  --color-text-tertiary: var(--text-tertiary);
+  --color-text-muted: var(--text-muted);
+  --color-accent: var(--accent);
+  --color-accent-hover: var(--accent-hover);
+  --color-accent-light: var(--accent-light);
+  --color-danger: var(--danger);
+  --color-danger-bg: var(--danger-bg);
+  --color-success: var(--success);
+  --color-success-bg: var(--success-bg);
+  --color-warning: var(--warning);
+  --color-warning-bg: var(--warning-bg);
+}
+
+:host {
+  /* light theme defaults — 跟主前端保持一致 */
+  --bg-primary: #f2f0ed;
+  --bg-secondary: #eae8e4;
+  --bg-tertiary: #e0ddd8;
+  --bg-card: #faf9f7;
+  --bg-sidebar: #eae8e4;
+  --bg-input: #f2f0ed;
+  --bg-hover: rgba(255, 255, 255, 0.4);
+  --bg-active: rgba(156, 175, 136, 0.08);
+  --bg-base: #f2f0ed;
+  --border-primary: #e0ddd8;
+  --border-secondary: #eae8e4;
+  --text-primary: #3d3d3d;
+  --text-secondary: #5a5a5a;
+  --text-tertiary: #8c8680;
+  --text-muted: #b5afa8;
+  --accent: #9caf88;
+  --accent-hover: #8a9e78;
+  --accent-light: rgba(156, 175, 136, 0.12);
+  --danger: #c4917a;
+  --danger-bg: rgba(196, 145, 122, 0.08);
+  --success: #7a8e6a;
+  --success-bg: rgba(122, 142, 106, 0.08);
+  --warning: #c4a882;
+  --warning-bg: rgba(196, 168, 130, 0.08);
+
+  display: block;
+  height: 100%;
+  font-family: "Inter", -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, sans-serif;
+}
+
+/* 跟随系统/主前端的暗色主题：宿主元素加 [data-theme="dark"] 时切换 */
+:host([data-theme="dark"]) {
+  --bg-primary: #1c1b19;
+  --bg-secondary: #232220;
+  --bg-tertiary: #2d2b28;
+  --bg-card: #252421;
+  --bg-sidebar: #1e1d1b;
+  --bg-input: #2d2b28;
+  --bg-hover: rgba(255, 255, 255, 0.04);
+  --bg-active: rgba(156, 175, 136, 0.1);
+  --bg-base: #1c1b19;
+  --border-primary: rgba(255, 255, 255, 0.06);
+  --border-secondary: rgba(255, 255, 255, 0.03);
+  --text-primary: #e8e6e3;
+  --text-secondary: #c8c5c0;
+  --text-tertiary: #a09c96;
+  --text-muted: #7a7772;
+  --accent: #a8bc94;
+  --accent-hover: #b8caa6;
+  --accent-light: rgba(156, 175, 136, 0.15);
+  --danger: #d4a894;
+  --danger-bg: rgba(196, 145, 122, 0.1);
+  --success: #9caf88;
+  --success-bg: rgba(156, 175, 136, 0.1);
+  --warning: #c4a882;
+  --warning-bg: rgba(196, 168, 130, 0.1);
+}

data/plugin/data_analytics/frontend/tsconfig.json

@@ -0,0 +1,13 @@
+{
+  "compilerOptions": {
+    "target": "ES2020",
+    "module": "ESNext",
+    "moduleResolution": "bundler",
+    "jsx": "react-jsx",
+    "strict": true,
+    "esModuleInterop": true,
+    "skipLibCheck": true,
+    "forceConsistentCasingInFileNames": true
+  },
+  "include": ["*.ts", "*.tsx"]
+}

data/plugin/data_analytics/frontend/types.ts

@@ -0,0 +1,25 @@
+export interface S3Credential {
+  cred_id: string;
+  user_id: string;
+  display_name: string;
+  endpoint_url: string | null;
+  region: string;
+  access_key: string;
+  created_at: string | null;
+  updated_at: string | null;
+}
+
+export interface AnalysisJob {
+  job_id: string;
+  user_id: string;
+  cred_id: string | null;
+  description: string;
+  status: string;
+  org_task_id: string | null;
+  result: string | null;
+  created_at: string | null;
+  updated_at: string | null;
+  task_status?: string;
+  task_result?: unknown;
+  task_error?: string | null;
+}

data/plugin/data_analytics/frontend/vite.config.ts

@@ -0,0 +1,25 @@
+import { defineConfig } from 'vite';
+import react from '@vitejs/plugin-react';
+import tailwindcss from '@tailwindcss/vite';
+
+export default defineConfig({
+  plugins: [react(), tailwindcss()],
+  build: {
+    lib: {
+      entry: 'index.tsx',
+      formats: ['es'],
+      fileName: () => 'plugin-element.js',
+    },
+    outDir: 'dist',
+    emptyOutDir: true,
+    cssCodeSplit: false,
+    rollupOptions: {
+      output: {
+        assetFileNames: (info) => {
+          if (info.name && info.name.endsWith('.css')) return 'plugin-element.css';
+          return 'assets/[name]-[hash][extname]';
+        },
+      },
+    },
+  },
+});

data/plugin/data_analytics/manifest.json

@@ -0,0 +1,19 @@
+{
+  "name": "data_analytics",
+  "version": "0.1.0",
+  "display_name": "数据分析",
+  "description": "对接 S3 对象存储，由 agent 自主决定使用 python_executor 或 ray_submit 跑分析。仅读不写。",
+  "entry": "core.organization:DataAnalyticsOrganization",
+  "concurrency": "queue",
+  "node_affinity": "cpu",
+  "api_prefix": "/api/v1/plugin/data_analytics",
+  "capabilities": ["data_analysis", "s3_readonly"],
+  "dependencies": {
+    "python": [],
+    "plugins": []
+  },
+  "ui": {
+    "entry": "frontend/index.tsx",
+    "icon": "BarChart3"
+  }
+}

data/plugin/data_analytics/toolset/__init__.py

@@ -0,0 +1,16 @@
+"""data_analytics 插件本地工具集。
+
+agent 看到这些工具时不带凭证参数，凭证由 organization 通过 ContextVar 注入。
+"""
+
+from .s3_list_objects import s3_list_objects
+from .s3_peek import s3_peek
+from .s3_get_object import s3_get_object
+from .ray_submit import ray_submit
+
+__all__ = [
+    "s3_list_objects",
+    "s3_peek",
+    "s3_get_object",
+    "ray_submit",
+]

data/plugin/data_analytics/toolset/_s3_common.py

@@ -0,0 +1,43 @@
+"""S3 工具共用辅助：从 ContextVar 拿凭证 + 解析 URI。
+
+所有 s3_* 工具都依赖这个模块，把"明文凭证"的取用集中在一处。
+"""
+
+from __future__ import annotations
+
+import re
+from typing import Any, Dict, Tuple
+
+
+def get_s3_creds_or_raise() -> Dict[str, Any]:
+    """从 organization 注入的 ContextVar 中取出明文凭证；未注入则抛错。"""
+    # 延迟 import 避免循环；这里走 organization 子类被加载时注入的虚拟包路径
+    from ..core.organization import S3_CREDS_VAR
+
+    creds = S3_CREDS_VAR.get()
+    if not creds:
+        raise RuntimeError(
+            "未提供 S3 凭证：本任务上下文中没有 cred_id，请在创建 job 时选择凭证。"
+        )
+    return creds
+
+
+def parse_s3_uri(uri: str) -> Tuple[str, str]:
+    """解析 ``s3://bucket/key`` → ``(bucket, key)``；非法格式抛 ValueError。"""
+    m = re.match(r"^s3://([^/]+)/(.+)$", uri.strip())
+    if not m:
+        raise ValueError(f"非法 S3 URI：{uri!r}（期待 s3://bucket/key 形式）")
+    return m.group(1), m.group(2)
+
+
+def make_session_kwargs(creds: Dict[str, Any]) -> Dict[str, Any]:
+    """转 boto3/aiobotocore client 调用所需的 kwargs。"""
+    kw: Dict[str, Any] = {
+        "aws_access_key_id": creds["access_key"],
+        "aws_secret_access_key": creds["secret_key"],
+        "region_name": creds.get("region") or "us-east-1",
+    }
+    endpoint = creds.get("endpoint_url")
+    if endpoint:
+        kw["endpoint_url"] = endpoint
+    return kw

data/plugin/data_analytics/toolset/manifest.json

@@ -0,0 +1,39 @@
+{
+  "name": "data_analytics_internal",
+  "version": "0.1.0",
+  "description": "data_analytics 插件内部工具：S3 只读 + Ray 提交。仅限本插件内部 agent 调用。",
+  "tools": [
+    {
+      "name": "s3_list_objects",
+      "file": "s3_list_objects.py",
+      "is_system": true,
+      "action_scope": ["data_analytics_internal"],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "s3_peek",
+      "file": "s3_peek.py",
+      "is_system": true,
+      "action_scope": ["data_analytics_internal"],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "s3_get_object",
+      "file": "s3_get_object.py",
+      "is_system": true,
+      "action_scope": ["data_analytics_internal"],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "ray_submit",
+      "file": "ray_submit.py",
+      "is_system": true,
+      "action_scope": ["data_analytics_internal"],
+      "config_args": {},
+      "category": "system"
+    }
+  ]
+}

data/plugin/data_analytics/toolset/ray_submit.py

@@ -0,0 +1,93 @@
+"""ray_submit：把分析脚本提交到子进程执行。
+
+名字沿用历史（原设计走 Ray）；当前后端为本地子进程，未来要接远程算力只需替换
+执行器，工具签名不变。凭证以 ``AWS_*`` 环境变量注入子进程，让 boto3/pandas-s3
+自然读到。脚本走 ``kilostar.utils.sandbox.validate_python_code`` 的静态屏蔽兜底。
+"""
+
+from __future__ import annotations
+
+import asyncio
+import os
+import sys
+import tempfile
+
+from kilostar.utils.sandbox import (
+    CodeViolation,
+    get_python_timeout,
+    validate_python_code,
+)
+
+from ._s3_common import get_s3_creds_or_raise
+
+
+def _build_env(creds) -> dict:
+    env = os.environ.copy()
+    env["AWS_ACCESS_KEY_ID"] = creds["access_key"]
+    env["AWS_SECRET_ACCESS_KEY"] = creds["secret_key"]
+    env["AWS_DEFAULT_REGION"] = creds.get("region") or "us-east-1"
+    if creds.get("endpoint_url"):
+        env["AWS_ENDPOINT_URL_S3"] = creds["endpoint_url"]
+        env["AWS_ENDPOINT_URL"] = creds["endpoint_url"]
+    return env
+
+
+async def ray_submit(script: str, timeout: int = 300) -> str:
+    """提交 Python 脚本到子进程执行。
+
+    脚本中可直接 ``import boto3`` 读 S3（凭证已通过环境变量注入）；可用
+    pandas / polars / numpy 等已安装的依赖。**只读**——不要尝试 put/delete。
+
+    Args:
+        script: Python 源码
+        timeout: 超时秒数（默认 300）
+
+    Returns:
+        stdout（必要时尾部追加 stderr 与 exit code）
+    """
+    try:
+        script = validate_python_code(script)
+    except CodeViolation as e:
+        return f"[Sandbox] {e}"
+
+    creds = get_s3_creds_or_raise()
+    env = _build_env(creds)
+    timeout = get_python_timeout(timeout)
+
+    # standalone 与 distributed 第一版都走 subprocess，保证环境变量传递可控
+    # （ray.remote 跑函数时 env vars 需另装 runtime_env，复杂度跟 subprocess 持平
+    #  但前者透明可控，先这样落地）
+    tmp_file = None
+    try:
+        with tempfile.NamedTemporaryFile(
+            mode="w", suffix=".py", delete=False, encoding="utf-8"
+        ) as f:
+            f.write(script)
+            tmp_file = f.name
+
+        proc = await asyncio.create_subprocess_exec(
+            sys.executable,
+            tmp_file,
+            stdout=asyncio.subprocess.PIPE,
+            stderr=asyncio.subprocess.PIPE,
+            env=env,
+        )
+        stdout, stderr = await asyncio.wait_for(proc.communicate(), timeout=timeout)
+        out = stdout.decode("utf-8", errors="replace")
+        err = stderr.decode("utf-8", errors="replace")
+        result = ""
+        if out:
+            result += out
+        if err:
+            result += f"\n[stderr]\n{err}"
+        if proc.returncode != 0:
+            result += f"\n[exit code: {proc.returncode}]"
+        result = result.strip() or "(no output)"
+        return result
+    except asyncio.TimeoutError:
+        return f"[Error] ray_submit 执行超时（{timeout}s）"
+    except Exception as e:
+        return f"[Error] ray_submit 失败：{e}"
+    finally:
+        if tmp_file and os.path.exists(tmp_file):
+            os.unlink(tmp_file)

data/plugin/data_analytics/toolset/s3_get_object.py

@@ -0,0 +1,46 @@
+"""s3_get_object：下载到 artifact 目录（路径强校验防穿越）。"""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+from kilostar.utils.settings import get_artifact_dir
+
+from ._s3_common import get_s3_creds_or_raise, make_session_kwargs, parse_s3_uri
+
+
+async def s3_get_object(uri: str, save_as: str) -> str:
+    """把 S3 对象下载到本进程的 artifact 工作区，返回本地绝对路径。
+
+    ``save_as`` 必须是相对路径，落到 ``data/artifact/data_analytics_downloads/``
+    下面（防越权写入任意目录）。下载后供 python_executor / ray_submit 中以
+    pandas/polars 读取。
+
+    Args:
+        uri: 形如 ``s3://bucket/key`` 的对象路径
+        save_as: 保存的相对文件名（不能含 ``..`` 或绝对路径）
+
+    Returns:
+        本地保存的绝对路径
+    """
+    from aiobotocore.session import get_session
+
+    creds = get_s3_creds_or_raise()
+    bucket, key = parse_s3_uri(uri)
+
+    save_path = Path(save_as).as_posix()
+    if save_path.startswith("/") or ".." in save_path.split("/"):
+        raise ValueError(f"save_as 必须是相对、不含 .. 的路径，收到 {save_as!r}")
+
+    base = get_artifact_dir() / "data_analytics_downloads"
+    base.mkdir(parents=True, exist_ok=True)
+    target = base / save_path
+    target.parent.mkdir(parents=True, exist_ok=True)
+
+    session = get_session()
+    async with session.create_client("s3", **make_session_kwargs(creds)) as client:
+        resp = await client.get_object(Bucket=bucket, Key=key)
+        body = await resp["Body"].read()
+    target.write_bytes(body)
+    return str(target.resolve())

data/plugin/data_analytics/toolset/s3_list_objects.py

@@ -0,0 +1,47 @@
+"""s3_list_objects：列出 bucket+prefix 下的对象列表（key/size/last_modified）。"""
+
+from __future__ import annotations
+
+from typing import Any, Dict, List
+
+from ._s3_common import get_s3_creds_or_raise, make_session_kwargs
+
+
+async def s3_list_objects(
+    bucket: str,
+    prefix: str = "",
+    limit: int = 50,
+) -> List[Dict[str, Any]]:
+    """列出 S3 bucket 下指定 prefix 的对象（最多 limit 条）。
+
+    Args:
+        bucket: S3 bucket 名
+        prefix: 对象 key 前缀，留空表示根路径
+        limit: 最多返回条数（1-1000），默认 50
+
+    Returns:
+        对象信息列表，每项含 key / size / last_modified（ISO 字符串）
+    """
+    from aiobotocore.session import get_session
+
+    creds = get_s3_creds_or_raise()
+    limit = max(1, min(int(limit), 1000))
+
+    session = get_session()
+    out: List[Dict[str, Any]] = []
+    async with session.create_client("s3", **make_session_kwargs(creds)) as client:
+        paginator = client.get_paginator("list_objects_v2")
+        async for page in paginator.paginate(
+            Bucket=bucket, Prefix=prefix, PaginationConfig={"MaxItems": limit}
+        ):
+            for item in page.get("Contents", []) or []:
+                out.append({
+                    "key": item.get("Key"),
+                    "size": item.get("Size"),
+                    "last_modified": (
+                        item["LastModified"].isoformat() if item.get("LastModified") else None
+                    ),
+                })
+                if len(out) >= limit:
+                    return out
+    return out

data/plugin/data_analytics/toolset/s3_peek.py

@@ -0,0 +1,35 @@
+"""s3_peek：读取对象的头若干字节并尝试 UTF-8 解码（看几行用）。"""
+
+from __future__ import annotations
+
+from ._s3_common import get_s3_creds_or_raise, make_session_kwargs, parse_s3_uri
+
+
+async def s3_peek(uri: str, n_bytes: int = 4096) -> str:
+    """读取 S3 对象的头 ``n_bytes`` 字节，UTF-8 解码后返回。
+
+    适合快速预览 csv/json/log 等文本类对象的开头几行。二进制内容会以
+    ``[binary, ...]`` 占位说明返回。
+
+    Args:
+        uri: 形如 ``s3://bucket/key`` 的对象路径
+        n_bytes: 读取字节数，默认 4096，上限 1MB
+
+    Returns:
+        对象内容片段（解码后的字符串或占位说明）
+    """
+    from aiobotocore.session import get_session
+
+    creds = get_s3_creds_or_raise()
+    bucket, key = parse_s3_uri(uri)
+    n = max(1, min(int(n_bytes), 1024 * 1024))
+
+    session = get_session()
+    async with session.create_client("s3", **make_session_kwargs(creds)) as client:
+        resp = await client.get_object(Bucket=bucket, Key=key, Range=f"bytes=0-{n-1}")
+        body = await resp["Body"].read()
+    try:
+        text = body.decode("utf-8")
+        return text
+    except UnicodeDecodeError:
+        return f"[binary, {len(body)} bytes; first 64 hex] {body[:64].hex()}"

data/plugin/example_dept/README.md

@@ -0,0 +1,16 @@
+# 示例部门 (Example Dept)
+
+演示用的重型插件骨架。包含两个平级 agent（analyst + executor），
+可作为开发新组织插件的模板。
+
+## 目录结构
+
+```
+example_dept/
+├── manifest.json    # 插件元数据
+├── agents.json      # agent 定义
+├── core/            # 业务逻辑
+├── toolset/         # 本地工具
+├── skills/          # 本地技能
+└── dashboard/       # 前端面板（占位）
+```

data/plugin/example_dept/agents.json

@@ -0,0 +1,32 @@
+{
+  "agents": [
+    {
+      "name": "analyst",
+      "role": "数据分析专家",
+      "system_prompt": "你是一位数据分析专家，负责理解用户需求并给出分析方案。",
+      "model": {
+        "provider_title": "",
+        "model_id": ""
+      },
+      "tools": [],
+      "skills": [],
+      "peers": ["executor"]
+    },
+    {
+      "name": "executor",
+      "role": "执行专家",
+      "system_prompt": "你是一位执行专家，负责将分析方案转化为具体操作。",
+      "model": {
+        "provider_title": "",
+        "model_id": ""
+      },
+      "tools": ["shell_executor", "python_executor"],
+      "skills": [],
+      "peers": ["analyst"]
+    }
+  ],
+  "orchestration": {
+    "type": "react",
+    "entry": "analyst"
+  }
+}

data/plugin/example_dept/core/organization.py

@@ -0,0 +1,6 @@
+from kilostar.plugin_runtime.base_organization import BaseOrganization
+
+
+class ExampleOrganization(BaseOrganization):
+    """示例组织 — 直接使用基类的 react 逻辑。"""
+    pass

data/plugin/example_dept/manifest.json

@@ -0,0 +1,19 @@
+{
+  "name": "example_dept",
+  "version": "0.1.0",
+  "display_name": "示例部门",
+  "description": "演示用的重型插件骨架，可作为开发模板。",
+  "entry": "core.organization:ExampleOrganization",
+  "concurrency": "queue",
+  "node_affinity": "cpu",
+  "api_prefix": "/plugin/example_dept",
+  "capabilities": ["text_processing"],
+  "dependencies": {
+    "python": [],
+    "plugins": []
+  },
+  "ui": {
+    "entry": "dashboard/index.html",
+    "icon": null
+  }
+}

data/toolset/base_toolset/README.md

@@ -0,0 +1,30 @@
+# base_toolset
+
+KiloStar 内置基础工具集。提供文件操作、命令执行、搜索等通用能力，所有 Agent 默认可用。
+
+## 工具列表
+
+| 工具 | 说明 |
+|------|------|
+| `shell_executor` | 执行 shell 命令，返回 stdout/stderr |
+| `file_reader` | 读取文件内容（支持按行偏移和行数限制） |
+| `edit_file` | 按 old_string → new_string 的方式精确替换文件内容 |
+| `write_file` | 整体写入或覆盖文件 |
+| `search_file` | 在目录树内按 glob/正则搜索文件名或内容 |
+| `python_executor` | 在沙箱中运行 Python 代码片段 |
+| `tavily_search` | 调用 Tavily API 进行联网搜索（需配置 `api_key`） |
+
+## 配置说明
+
+`tavily_search` 需要在工具配置中填入 `api_key`，可选参数：
+
+- `max_results`：返回结果条数，默认 `5`
+- `search_depth`：`basic` 或 `advanced`
+- `include_answer`：是否带 LLM 摘要，默认 `true`
+
+其他工具开箱即用，无需配置。
+
+## 安全提示
+
+- `shell_executor` / `python_executor` 会在受限沙箱内执行，但仍建议在受信环境下使用
+- `edit_file` / `write_file` 会修改本地文件系统，注意权限范围

data/toolset/base_toolset/__init__.py

@@ -0,0 +1,17 @@
+from .shell_executor import shell_executor
+from .file_reader import file_reader
+from .edit_file import edit_file
+from .write_file import write_file
+from .search_file import search_file
+from .python_executor import python_executor
+from .tavily_search import tavily_search
+
+__all__ = [
+    "shell_executor",
+    "file_reader",
+    "edit_file",
+    "write_file",
+    "search_file",
+    "python_executor",
+    "tavily_search",
+]

data/toolset/base_toolset/edit_file.py

@@ -0,0 +1,43 @@
+import os
+
+
+async def edit_file(
+    file_path: str,
+    old_content: str,
+    new_content: str,
+) -> str:
+    """通过查找替换的方式编辑文件内容。
+
+    Args:
+        file_path: 文件的路径
+        old_content: 要被替换的原始内容片段
+        new_content: 替换后的新内容
+
+    Returns:
+        操作结果描述
+    """
+    from kilostar.utils.sandbox import validate_path, PathViolation
+
+    try:
+        file_path = validate_path(file_path, write=True)
+    except PathViolation as e:
+        return f"[Sandbox] {e}"
+
+    try:
+        if not os.path.exists(file_path):
+            return f"[Error] 文件不存在: {file_path}"
+
+        with open(file_path, "r", encoding="utf-8") as f:
+            content = f.read()
+
+        if old_content not in content:
+            return f"[Error] 未在文件中找到要替换的内容片段"
+
+        new_file_content = content.replace(old_content, new_content, 1)
+
+        with open(file_path, "w", encoding="utf-8") as f:
+            f.write(new_file_content)
+
+        return f"已成功编辑文件: {file_path}"
+    except Exception as e:
+        return f"[Error] 编辑文件失败: {e}"

data/toolset/base_toolset/file_reader.py

@@ -0,0 +1,23 @@
+async def file_reader(file_path: str) -> str:
+    """读取本地文件的内容。
+
+    Args:
+        file_path: 文件的绝对路径或相对路径
+
+    Returns:
+        文件内容文本，若文件不存在则返回错误信息
+    """
+    from kilostar.utils.sandbox import validate_path, PathViolation
+
+    try:
+        file_path = validate_path(file_path, write=False)
+    except PathViolation as e:
+        return f"[Sandbox] {e}"
+
+    try:
+        with open(file_path, "r", encoding="utf-8") as f:
+            return f.read()
+    except FileNotFoundError:
+        return f"[Error] File not found: {file_path}"
+    except Exception as e:
+        return f"[Error] Failed to read file: {str(e)}"

data/toolset/base_toolset/manifest.json

@@ -0,0 +1,68 @@
+{
+  "name": "基础工具集",
+  "version": "0.1.0",
+  "description": "文件读写、命令执行、Python/搜索等通用能力",
+  "tools": [
+    {
+      "name": "shell_executor",
+      "file": "shell_executor.py",
+      "is_system": true,
+      "action_scope": [],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "file_reader",
+      "file": "file_reader.py",
+      "is_system": true,
+      "action_scope": [],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "edit_file",
+      "file": "edit_file.py",
+      "is_system": true,
+      "action_scope": [],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "write_file",
+      "file": "write_file.py",
+      "is_system": true,
+      "action_scope": [],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "search_file",
+      "file": "search_file.py",
+      "is_system": true,
+      "action_scope": [],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "python_executor",
+      "file": "python_executor.py",
+      "is_system": true,
+      "action_scope": [],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "tavily_search",
+      "file": "tavily_search.py",
+      "is_system": false,
+      "action_scope": ["consciousness_node", "regulatory_node"],
+      "config_args": {
+        "api_key": "",
+        "max_results": "5",
+        "search_depth": "basic",
+        "include_answer": "true"
+      },
+      "category": "search"
+    }
+  ]
+}

data/toolset/base_toolset/python_executor.py

@@ -0,0 +1,59 @@
+import asyncio
+import sys
+import tempfile
+import os
+
+
+async def python_executor(code: str, timeout: int = 30) -> str:
+    """执行 Python 代码片段并返回输出。
+
+    Args:
+        code: 要执行的 Python 代码
+        timeout: 超时秒数，默认 30 秒
+
+    Returns:
+        代码的标准输出 + 标准错误
+    """
+    from kilostar.utils.sandbox import (
+        validate_python_code, CodeViolation, get_python_timeout,
+    )
+
+    try:
+        code = validate_python_code(code)
+    except CodeViolation as e:
+        return f"[Sandbox] {e}"
+    timeout = get_python_timeout(timeout)
+
+    tmp_file = None
+    try:
+        with tempfile.NamedTemporaryFile(
+            mode="w", suffix=".py", delete=False, encoding="utf-8"
+        ) as f:
+            f.write(code)
+            tmp_file = f.name
+
+        proc = await asyncio.create_subprocess_exec(
+            sys.executable, tmp_file,
+            stdout=asyncio.subprocess.PIPE,
+            stderr=asyncio.subprocess.PIPE,
+        )
+        stdout, stderr = await asyncio.wait_for(
+            proc.communicate(), timeout=timeout
+        )
+        output = stdout.decode("utf-8", errors="replace")
+        err_output = stderr.decode("utf-8", errors="replace")
+        result = ""
+        if output:
+            result += output
+        if err_output:
+            result += f"\n[stderr]\n{err_output}"
+        if proc.returncode != 0:
+            result += f"\n[exit code: {proc.returncode}]"
+        return result.strip() or "(no output)"
+    except asyncio.TimeoutError:
+        return f"[Error] Python 代码执行超时（{timeout}s）"
+    except Exception as e:
+        return f"[Error] 执行失败: {e}"
+    finally:
+        if tmp_file and os.path.exists(tmp_file):
+            os.unlink(tmp_file)

data/toolset/base_toolset/search_file.py

@@ -0,0 +1,55 @@
+import asyncio
+
+
+async def search_file(
+    keyword: str,
+    directory: str = ".",
+    file_pattern: str = "*",
+    max_results: int = 20,
+) -> str:
+    """在指定目录下递归搜索包含关键字的文件内容。
+
+    Args:
+        keyword: 要搜索的关键字或正则表达式
+        directory: 搜索的根目录，默认当前目录
+        file_pattern: 文件名匹配模式，如 "*.py"
+        max_results: 最大返回结果数
+
+    Returns:
+        匹配的文件名和行内容
+    """
+    from kilostar.utils.sandbox import validate_path, PathViolation
+
+    try:
+        directory = validate_path(directory, write=False)
+    except PathViolation as e:
+        return f"[Sandbox] {e}"
+
+    max_results = min(max_results, 100)
+
+    try:
+        grep_args = [
+            "grep", "-rn",
+            f"--include={file_pattern}",
+            "-m", str(max_results),
+            "--", keyword, directory,
+        ]
+        proc = await asyncio.create_subprocess_exec(
+            *grep_args,
+            stdout=asyncio.subprocess.PIPE,
+            stderr=asyncio.subprocess.PIPE,
+        )
+        stdout, _ = await asyncio.wait_for(
+            proc.communicate(), timeout=30
+        )
+        output = stdout.decode("utf-8", errors="replace").strip()
+        if not output:
+            return f"未找到包含 '{keyword}' 的匹配项"
+        lines = output.split("\n")
+        if len(lines) > max_results:
+            output = "\n".join(lines[:max_results])
+        return output
+    except asyncio.TimeoutError:
+        return "[Error] 搜索超时"
+    except Exception as e:
+        return f"[Error] 搜索失败: {e}"

data/toolset/base_toolset/shell_executor.py

@@ -0,0 +1,46 @@
+import asyncio
+
+
+async def shell_executor(command: str, timeout: int = 30) -> str:
+    """在服务器上执行 shell 命令并返回输出。
+
+    Args:
+        command: 要执行的 shell 命令
+        timeout: 超时秒数，默认 30 秒
+
+    Returns:
+        命令的 stdout + stderr 输出
+    """
+    from kilostar.utils.sandbox import (
+        validate_shell_command, CommandViolation, get_shell_timeout,
+    )
+
+    try:
+        command = validate_shell_command(command)
+    except CommandViolation as e:
+        return f"[Sandbox] {e}"
+    timeout = get_shell_timeout(timeout)
+
+    try:
+        proc = await asyncio.create_subprocess_shell(
+            command,
+            stdout=asyncio.subprocess.PIPE,
+            stderr=asyncio.subprocess.PIPE,
+        )
+        stdout, stderr = await asyncio.wait_for(
+            proc.communicate(), timeout=timeout
+        )
+        output = stdout.decode("utf-8", errors="replace")
+        err_output = stderr.decode("utf-8", errors="replace")
+        result = ""
+        if output:
+            result += output
+        if err_output:
+            result += f"\n[stderr]\n{err_output}"
+        if proc.returncode != 0:
+            result += f"\n[exit code: {proc.returncode}]"
+        return result.strip() or "(no output)"
+    except asyncio.TimeoutError:
+        return f"[Error] 命令执行超时（{timeout}s）"
+    except Exception as e:
+        return f"[Error] 执行失败: {e}"

data/toolset/base_toolset/tavily_search.py

@@ -0,0 +1,78 @@
+import os
+from typing import Optional
+
+from tavily import AsyncTavilyClient
+
+
+async def _resolve_api_key(explicit: Optional[str]) -> Optional[str]:
+    """按优先级解析 Tavily API key：显式参数 > GSM 配置 > 环境变量。"""
+    if explicit:
+        return explicit
+    try:
+        from kilostar.core.global_state_machine.gsm_snapshot import fetch_snapshot
+
+        snapshot = await fetch_snapshot()
+        cfg = snapshot.tool_configs.get("tavily_search") or {}
+        if isinstance(cfg, dict) and cfg.get("api_key"):
+            return cfg["api_key"]
+    except Exception:
+        pass
+    return os.environ.get("TAVILY_API_KEY")
+
+
+async def tavily_search(
+    query: str,
+    max_results: int = 5,
+    search_depth: str = "basic",
+    include_answer: bool = True,
+    api_key: Optional[str] = None,
+) -> str:
+    """使用 Tavily 进行网络搜索，获取高质量的网络搜索结果。
+
+    Args:
+        query: 搜索查询内容
+        max_results: 返回的最大结果数量（1-10）
+        search_depth: 搜索深度，"basic" 或 "advanced"
+        include_answer: 是否包含 AI 生成的答案摘要
+        api_key: 可选；不传则按 GSM 配置 → 环境变量顺序解析
+
+    Returns:
+        格式化的搜索结果文本，包含标题、URL、摘要和可选的 AI 答案
+    """
+    resolved_key = await _resolve_api_key(api_key)
+    if not resolved_key:
+        return (
+            "[Error] Tavily API key 未配置。"
+            "请在 ``/api/v1/resource/tool/config`` 写入或设置环境变量 ``TAVILY_API_KEY``。"
+        )
+
+    try:
+        client = AsyncTavilyClient(api_key=resolved_key)
+        result = await client.search(
+            query=query,
+            max_results=min(max_results, 10),
+            search_depth=search_depth,
+            include_answer=include_answer,
+        )
+
+        lines = []
+        if include_answer and result.get("answer"):
+            lines.append(f"【AI 摘要】{result['answer']}\n")
+
+        results = result.get("results", [])
+        if not results:
+            return "No results found for the query."
+
+        lines.append("【搜索结果】")
+        for i, item in enumerate(results, 1):
+            title = item.get("title", "Untitled")
+            url = item.get("url", "")
+            content = item.get("content", "").strip()
+            lines.append(f"\n{i}. {title}")
+            lines.append(f"   URL: {url}")
+            if content:
+                lines.append(f"   {content[:300]}{'...' if len(content) > 300 else ''}")
+
+        return "\n".join(lines)
+    except Exception as e:
+        return f"[Error] Tavily search failed: {str(e)}"

data/toolset/base_toolset/write_file.py

@@ -0,0 +1,31 @@
+import os
+
+
+async def write_file(file_path: str, content: str) -> str:
+    """将内容写入指定文件（会覆盖已有内容，自动创建目录）。
+
+    Args:
+        file_path: 文件的路径
+        content: 要写入的内容
+
+    Returns:
+        操作结果描述
+    """
+    from kilostar.utils.sandbox import validate_path, PathViolation
+
+    try:
+        file_path = validate_path(file_path, write=True)
+    except PathViolation as e:
+        return f"[Sandbox] {e}"
+
+    try:
+        dir_path = os.path.dirname(file_path)
+        if dir_path:
+            os.makedirs(dir_path, exist_ok=True)
+
+        with open(file_path, "w", encoding="utf-8") as f:
+            f.write(content)
+
+        return f"已成功写入文件: {file_path}（{len(content)} 字符）"
+    except Exception as e:
+        return f"[Error] 写入文件失败: {e}"

data/toolset/interactive_toolset/README.md

@@ -0,0 +1,26 @@
+# interactive__toolset
+
+KiloStar 工作流交互工具集。这些工具用于 Agent 与用户/前端之间的实时交互，依赖 `global_workflow_manager` 的消息通道。
+
+## 工具列表
+
+| 工具 | 说明 |
+|------|------|
+| `approval` | 在执行高风险操作前向用户发送审批请求，阻塞等待用户回复 |
+| `send_file` | 把 Agent 生成的文件作为附件推送到当前对话窗口，前端渲染为可下载卡片 |
+
+## 使用前提
+
+这两个工具需要工作流上下文：调用方必须在 deps 中传入 `trace_id`，工具会通过 `global_workflow_manager` 的 pending 队列与前端通信。
+
+- 在普通聊天场景下，`send_file` 在 `trace_id` 为空时会退化为直接返回文件内容字符串
+- `approval` 在没有合法 `trace_id` 时会一直阻塞，建议仅在工作流节点中绑定
+
+## 配置说明
+
+无需任何配置，开箱即用。
+
+## 适用场景
+
+- Agent 计划执行删除、转账等高风险操作前的人工确认
+- 让 Agent 把生成的报告、代码片段、图表数据以文件形式投递给用户

data/toolset/interactive_toolset/__init__.py

@@ -0,0 +1,5 @@
+from .approval import approval
+
+__all__ = [
+    "approval",
+]

data/toolset/interactive_toolset/approval.py

@@ -0,0 +1,17 @@
+from kilostar.utils.actor import get_actor
+
+
+async def approval(message: str, trace_id: str) -> str:
+    """当任务存在某些高风险操作或者计划需要让用户审批，发送请求给用户等待用户审批。
+
+    Args:
+        message: 发送给用户的请求
+        trace_id: 当前工作流的 trace_id
+
+    Returns:
+        用户的审批结果
+    """
+    gwm = get_actor("global_workflow_manager")
+    await gwm.put_pending(trace_id, message)
+    reply = await gwm.get_received(trace_id)
+    return reply

data/toolset/interactive_toolset/manifest.json

@@ -0,0 +1,15 @@
+{
+  "name": "交互工具集",
+  "version": "0.1.0",
+  "description": "工作流场景下与用户/前端交互的工具（HITL 审批）",
+  "tools": [
+    {
+      "name": "approval",
+      "file": "approval.py",
+      "is_system": true,
+      "action_scope": [],
+      "config_args": {},
+      "category": "system"
+    }
+  ]
+}

data/toolset/regulatory_toolset/__init__.py

@@ -0,0 +1,9 @@
+from .query_workflow_status import query_workflow_status
+from .query_task_list import query_task_list
+from .send_file import send_file
+
+__all__ = [
+    "query_workflow_status",
+    "query_task_list",
+    "send_file",
+]

data/toolset/regulatory_toolset/manifest.json

@@ -0,0 +1,31 @@
+{
+  "name": "监管节点工具集",
+  "version": "0.1.0",
+  "description": "监管节点（regulatory_node）专属能力：查询工作流、查询任务列表、发送文件",
+  "tools": [
+    {
+      "name": "query_workflow_status",
+      "file": "query_workflow_status.py",
+      "is_system": true,
+      "action_scope": ["regulatory_node"],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "query_task_list",
+      "file": "query_task_list.py",
+      "is_system": true,
+      "action_scope": ["regulatory_node"],
+      "config_args": {},
+      "category": "system"
+    },
+    {
+      "name": "send_file",
+      "file": "send_file.py",
+      "is_system": true,
+      "action_scope": ["regulatory_node"],
+      "config_args": {},
+      "category": "system"
+    }
+  ]
+}

data/toolset/regulatory_toolset/query_task_list.py

@@ -0,0 +1,59 @@
+"""query_task_list：列出当前用户的短任务记录。
+
+regulatory_node 用以回答"我之前那份报告呢""昨天那个查询结果是什么"。
+返回 task 表中的精简元数据列表（不含工作流的 graph state、context 等）。
+
+注：此处的 "task" 是 regulatory_node 完成的轻量短任务（出报告/写文件/查询整理等），
+与 workflow（多步骤复杂任务）是两套独立体系。如需查工作流进度，使用 query_workflow_status。
+"""
+
+from typing import Any, Dict, List, Optional
+
+from kilostar.utils.actor import get_actor
+
+
+async def query_task_list(
+    user_id: str,
+    status_filter: Optional[str] = None,
+    limit: int = 20,
+) -> Dict[str, Any]:
+    """列出当前用户的短任务记录，按时间倒序。
+
+    Args:
+        user_id: 用户 ID（通常由调用方从对话上下文中带入）
+        status_filter: 可选，按状态过滤（running/completed/failed）
+        limit: 最多返回条数，默认 20
+
+    Returns:
+        {
+            "user_id": str,
+            "tasks": [
+                {"task_id": ..., "title": ..., "status": ...,
+                 "result_summary": ..., "created_at": ...}
+            ],
+            "total": int
+        }
+    """
+    pg = get_actor("postgres_database")
+    rows: List[Dict[str, Any]] = await pg.list_tasks_by_user(
+        user_id=user_id,
+        status=status_filter,
+        limit=limit,
+    ) or []
+
+    tasks = [
+        {
+            "task_id": r.get("task_id"),
+            "title": r.get("title"),
+            "status": r.get("status"),
+            "result_summary": r.get("result_summary"),
+            "created_at": r.get("created_at"),
+        }
+        for r in rows
+    ]
+
+    return {
+        "user_id": user_id,
+        "tasks": tasks,
+        "total": len(tasks),
+    }

data/toolset/regulatory_toolset/query_workflow_status.py

@@ -0,0 +1,51 @@
+"""query_workflow_status：查询某个 trace_id 对应工作流的最近事件。
+
+regulatory_node 在与用户对话时，可以借此工具回答"我那个任务跑到哪一步了"
+之类的问题。返回最近 N 条事件 + 当前工作流 status。
+"""
+
+from typing import Any, Dict, List
+
+from kilostar.utils.actor import get_actor
+
+
+async def query_workflow_status(trace_id: str, limit: int = 10) -> Dict[str, Any]:
+    """查询指定工作流 trace_id 的状态与最近事件。
+
+    Args:
+        trace_id: 工作流追踪 ID
+        limit: 返回最近多少条事件，默认 10
+
+    Returns:
+        {
+            "trace_id": str,
+            "status": str | None,        # 工作流当前状态（pending/running/completed/failed）
+            "title": str | None,
+            "recent_events": [           # 最近事件，按时间倒序
+                {"event_type": ..., "level": ..., "message": ..., "node_name": ..., "created_at": ...}
+            ]
+        }
+    """
+    pg = get_actor("postgres_database")
+
+    workflow = await pg.get_workflow(trace_id)
+    events = await pg.query_event_logs(trace_id=trace_id, limit=limit)
+
+    recent: List[Dict[str, Any]] = []
+    for e in events or []:
+        recent.append(
+            {
+                "event_type": getattr(e, "event_type", None),
+                "level": getattr(e, "level", None),
+                "message": getattr(e, "message", None),
+                "node_name": getattr(e, "node_name", None),
+                "created_at": str(getattr(e, "created_at", "")),
+            }
+        )
+
+    return {
+        "trace_id": trace_id,
+        "status": getattr(workflow, "status", None) if workflow else None,
+        "title": getattr(workflow, "title", None) if workflow else None,
+        "recent_events": recent,
+    }

data/toolset/regulatory_toolset/send_file.py

@@ -0,0 +1,63 @@
+"""send_file：在对话/工作流场景下投递一份文件给用户。
+
+regulatory_node 直接对话场景下用此工具把生成的文件发给用户：
+- 工作流场景（带 trace_id）：写入 data/artifact/<trace_id>/，前端通过 SSE
+  收到带下载链接的卡片。
+- 直接对话场景（无 trace_id）：退化为把文件内容拼回字符串返回给 agent，
+  让 agent 再以代码块形式吐给用户。
+"""
+
+import json
+import re
+import uuid
+from pathlib import Path
+
+from kilostar.utils.actor import get_actor
+from kilostar.utils.settings import get_artifact_dir
+
+
+_SAFE_NAME_RE = re.compile(r"[^A-Za-z0-9._-]+")
+
+
+def _sanitize_filename(name: str) -> str:
+    name = name.strip().replace("\\", "/").split("/")[-1]
+    name = _SAFE_NAME_RE.sub("_", name)
+    return name or "file"
+
+
+async def send_file(filename: str, content: str, trace_id: str = "") -> str:
+    """把 agent 生成的文件作为附件投递给用户。
+
+    Args:
+        filename: 文件名（含扩展名），如 "report.md" / "main.py"
+        content: 文件内容（UTF-8 文本）
+        trace_id: 当前会话/工作流的 trace_id；为空时退化为直接返回内容
+
+    Returns:
+        发送结果说明或文件内容
+    """
+    if not trace_id:
+        return f"文件 {filename} 内容如下:\n\n```\n{content}\n```"
+
+    safe_name = _sanitize_filename(filename)
+    artifact_id = uuid.uuid4().hex[:12]
+    trace_dir: Path = get_artifact_dir() / trace_id
+    trace_dir.mkdir(parents=True, exist_ok=True)
+    file_path = trace_dir / f"{artifact_id}_{safe_name}"
+    file_path.write_text(content, encoding="utf-8")
+
+    payload = json.dumps(
+        {
+            "type": "file",
+            "filename": safe_name,
+            "artifact_id": artifact_id,
+            "url": f"/api/v1/resource/artifact/{trace_id}/{artifact_id}",
+            "size": len(content.encode("utf-8")),
+        },
+        ensure_ascii=False,
+    )
+    gwm = get_actor("global_workflow_manager")
+    await gwm.put_pending(
+        trace_id, f"__FILE__{payload}"
+    )
+    return f"已发送文件: {safe_name}"

docker-compose.yml

@@ -19,7 +19,6 @@ services:
     container_name: kilostar_test
     ports:
       - "8000:8000"
-      - "8265:8265"
     depends_on:
       db:
         condition: service_healthy
@@ -31,5 +30,4 @@ services:
       POSTGRES_DB: kilostar
       SECRET_KEY: test-secret-key-not-for-production
       KILOSTAR_SECRET_KEY: test-secret-key-not-for-production
-      KILOSTAR_MODE: standalone
       KILOSTAR_ENV: dev

docs/STRUCTURE.md

@@ -38,7 +38,7 @@ KiloStar/
 │   │   │   ├── regulatory_node/     #     监管节点：直面用户对话、质量把关
 │   │   │   │   ├── regulatory_node.py
 │   │   │   │   └── template.py
-│   │   │   ├── control_node/        #     控制节点：工作流节点内路由调度
+│   │   │   ├── control_node/        #     控制节点（已废弃，名字保留给未来远程探针节点）
 │   │   │   │   ├── control_node.py
 │   │   │   │   └── template.py
 │   │   │   └── growth_node/         #     生长节点：能力自扩展（占位）

frontend/CLAUDE.md

@@ -64,7 +64,7 @@ Mode is set via `KILOSTAR_MODE` env var. Entry point `main.py` branches into `st
 ### Backend Layout (`kilostar/`)
 
 - `api/` — FastAPI routers (auth, chat, agent, workflow, system, resource, platform)
-- `core/individual/` — 4 node types: RegulatoryNode (user-facing QA), ConsciousnessNode (planning), ControlNode (routing), GrowthNode (capability expansion)
+- `core/individual/` — 4 node types: RegulatoryNode (user-facing QA + short tasks), ConsciousnessNode (workflow planning), ControlNode (deprecated; name reserved for future remote-probe node), GrowthNode (capability expansion, not yet implemented)
 - `core/global_state_machine/` — Provider registry, model config state
 - `core/global_workflow_manager/` — Workflow queue & recovery
 - `core/postgres_database/` — DAO layer: `model/` (SQLAlchemy models), `module/` (CRUD methods), `postgres.py` (facade)

frontend/src/App.tsx

@@ -15,6 +15,7 @@ import { RightPanel } from './components/Chat/RightPanel';
 import { WorkflowListView } from './components/Chat/WorkflowListView';
 import { NewWorkflowDialog } from './components/Chat/NewWorkflowDialog';
 import { AuthPage } from './components/Auth/AuthPage';
+import { HeavyPluginShell } from './plugins/HeavyPluginShell';
 import { useAppStore } from './store/useAppStore';
 import { useChatStore } from './store/useChatStore';
 
@@ -29,10 +30,12 @@ function App() {
     agentTab,
     applyTheme,
     locale,
+    loadInstalledPlugins,
   } = useAppStore();
 
   const { loadSessions } = useChatStore();
   const [showSetupGuide, setShowSetupGuide] = useState(false);
+  const activeHeavyPlugin = useAppStore((s) => s.activeHeavyPlugin);
 
   useEffect(() => {
     applyTheme();
@@ -58,9 +61,10 @@ function App() {
   useEffect(() => {
     if (isAuthenticated) {
       loadSessions();
+      loadInstalledPlugins();
       setShowSetupGuide(true);
     }
-  }, [isAuthenticated, loadSessions]);
+  }, [isAuthenticated, loadSessions, loadInstalledPlugins]);
 
   if (!isAuthenticated) {
     return <AuthPage onLoginSuccess={() => setIsAuthenticated(true)} />;
@@ -90,11 +94,17 @@ function App() {
               {mode === 'work' && workTab === 'chat' && (
                 <div className="flex-1 flex overflow-hidden">
                   <LeftPanel activeTab="chats" />
+                  {activeHeavyPlugin ? (
+                    <HeavyPluginShell name={activeHeavyPlugin} />
+                  ) : (
                     <ChatPanel />
+                  )}
                 </div>
               )}
 
-              {mode === 'work' && workTab === 'workflow' && <WorkflowShell />}
+              {mode === 'work' && workTab === 'workflow' && (
+                <WorkflowShell />
+              )}
 
               {mode === 'agent' && agentTab === 'agents' && <AgentLayout />}
 
@@ -149,7 +159,12 @@ function WorkflowShell() {
     );
   }
 
-  return <WorkflowListView onSelectWorkflow={setSelectedWorkflow} />;
+  return (
+    <div className="flex-1 flex overflow-hidden">
+      <LeftPanel activeTab="workflows" />
+      <WorkflowListView onSelectWorkflow={setSelectedWorkflow} />
+    </div>
+  );
 }
 
 export default App;

frontend/src/components/Agent/ProvidersSettings.tsx

@@ -37,13 +37,14 @@ export function ProvidersSettings() {
   const [isModalOpen, setIsModalOpen] = useState(false);
   const [editingProvider, setEditingProvider] = useState<string | null>(null);
   const [selectedTypeId, setSelectedTypeId] = useState<string>('openai');
-  const [formData, setFormData] = useState({ provider_title: '', provider_url: '', provider_apikey: '', custom_models: '' });
+  const [formData, setFormData] = useState({ provider_title: '', provider_url: '', provider_apikey: '', custom_models: '', model_settings: '' });
   const [showAdvanced, setShowAdvanced] = useState(false);
   const [submitLoading, setSubmitLoading] = useState(false);
   const [testLoading, setTestLoading] = useState(false);
   const [testResult, setTestResult] = useState<{ success: boolean; error?: string; model_count?: number } | null>(null);
   const [error, setError] = useState('');
   const [expandedProvider, setExpandedProvider] = useState<string | null>(null);
+  const [modelSettingsError, setModelSettingsError] = useState('');
 
   const selectedType = PROVIDER_TYPES.find((p) => p.id === selectedTypeId) || PROVIDER_TYPES[0];
 
@@ -65,8 +66,9 @@ export function ProvidersSettings() {
   const openAddModal = () => {
     setEditingProvider(null);
     setSelectedTypeId('openai');
-    setFormData({ provider_title: '', provider_url: PROVIDER_TYPES[0].defaultUrl, provider_apikey: '', custom_models: '' });
+    setFormData({ provider_title: '', provider_url: PROVIDER_TYPES[0].defaultUrl, provider_apikey: '', custom_models: '', model_settings: '' });
     setError('');
+    setModelSettingsError('');
     setTestResult(null);
     setShowAdvanced(false);
     setIsModalOpen(true);
@@ -81,8 +83,12 @@ export function ProvidersSettings() {
       provider_url: provider.provider_url || '',
       provider_apikey: '',
       custom_models: '',
+      model_settings: provider.model_settings && Object.keys(provider.model_settings).length > 0
+        ? JSON.stringify(provider.model_settings, null, 2)
+        : '',
     });
     setError('');
+    setModelSettingsError('');
     setTestResult(null);
     setShowAdvanced(false);
     setIsModalOpen(true);
@@ -97,6 +103,29 @@ export function ProvidersSettings() {
     setTestResult(null);
   };
 
+  const parseModelSettings = (): { ok: true; value: Record<string, Record<string, unknown>> | undefined } | { ok: false } => {
+    const raw = formData.model_settings.trim();
+    if (!raw) return { ok: true, value: undefined };
+    try {
+      const parsed = JSON.parse(raw);
+      if (typeof parsed !== 'object' || parsed === null || Array.isArray(parsed)) {
+        setModelSettingsError(t('agent.providerModelSettingsInvalid'));
+        return { ok: false };
+      }
+      for (const v of Object.values(parsed)) {
+        if (typeof v !== 'object' || v === null || Array.isArray(v)) {
+          setModelSettingsError(t('agent.providerModelSettingsInvalid'));
+          return { ok: false };
+        }
+      }
+      setModelSettingsError('');
+      return { ok: true, value: parsed as Record<string, Record<string, unknown>> };
+    } catch {
+      setModelSettingsError(t('agent.providerModelSettingsInvalid'));
+      return { ok: false };
+    }
+  };
+
   const buildPayload = () => {
     const customModels = formData.custom_models
       .split(',').map((s) => s.trim()).filter(Boolean);
@@ -107,6 +136,8 @@ export function ProvidersSettings() {
       provider_apikey: formData.provider_apikey,
     };
     if (customModels.length > 0) payload.custom_models = customModels;
+    const ms = parseModelSettings();
+    if (ms.ok && ms.value) payload.model_settings = ms.value;
     return payload;
   };
 
@@ -134,6 +165,8 @@ export function ProvidersSettings() {
       setError(t('agent.providerFillAll'));
       return;
     }
+    const ms = parseModelSettings();
+    if (!ms.ok) return;
     setSubmitLoading(true);
     setError('');
     try {
@@ -361,6 +394,31 @@ export function ProvidersSettings() {
                         />
                         <p className="text-[10px] text-text-muted mt-1">{t('agent.providerCustomModelsHint')}</p>
                       </div>
+
+                      <div>
+                        <label className="block text-xs font-semibold text-text-secondary mb-1.5 uppercase tracking-wider">
+                          {t('agent.providerModelSettings')}
+                        </label>
+                        <textarea
+                          value={formData.model_settings}
+                          onChange={(e) => {
+                            setFormData({ ...formData, model_settings: e.target.value });
+                            if (modelSettingsError) setModelSettingsError('');
+                          }}
+                          placeholder={t('agent.providerModelSettingsPlaceholder')}
+                          rows={8}
+                          className={`w-full bg-bg-input border text-sm rounded-xl px-3.5 py-2.5 focus:outline-none focus:ring-2 text-text-primary placeholder:text-text-muted/50 font-mono resize-none ${
+                            modelSettingsError
+                              ? 'border-danger/50 focus:ring-danger/20 focus:border-danger'
+                              : 'border-border-primary focus:ring-accent/20 focus:border-accent'
+                          }`}
+                        />
+                        {modelSettingsError ? (
+                          <p className="text-[10px] text-danger mt-1">{modelSettingsError}</p>
+                        ) : (
+                          <p className="text-[10px] text-text-muted mt-1">{t('agent.providerModelSettingsHint')}</p>
+                        )}
+                      </div>
                     </div>
                   )}
                 </div>

frontend/src/components/Chat/LeftPanel.tsx

@@ -1,9 +1,10 @@
 import { useState, useEffect } from 'react';
 import { useTranslation } from 'react-i18next';
-import { Plus, Trash2, MessageSquare, Workflow as WorkflowIcon, Pencil, Check } from 'lucide-react';
+import { Plus, Trash2, MessageSquare, Workflow as WorkflowIcon, Pencil, Check, Box } from 'lucide-react';
 import apiClient from '../../api/client';
 import type { Workflow } from '../../types';
 import { useChatStore } from '../../store/useChatStore';
+import { useAppStore } from '../../store/useAppStore';
 
 interface LeftPanelProps {
   activeTab: string;
@@ -26,6 +27,12 @@ export function LeftPanel({ activeTab }: LeftPanelProps) {
     updateSessionTitle,
   } = useChatStore();
 
+  const installedHeavyPlugins = useAppStore((s) => s.installedHeavyPlugins);
+  const activeHeavyPlugin = useAppStore((s) => s.activeHeavyPlugin);
+  const setActiveHeavyPlugin = useAppStore((s) => s.setActiveHeavyPlugin);
+  const heavyPluginsWithUi = installedHeavyPlugins.filter((p) => p.has_ui);
+
+
   useEffect(() => {
     let intervalId: ReturnType<typeof setInterval>;
 
@@ -61,6 +68,7 @@ export function LeftPanel({ activeTab }: LeftPanelProps) {
 
   const handleNewChat = () => {
     setActiveSessionId(null);
+    setActiveHeavyPlugin(null);
   };
 
   const handleDeleteChat = (e: React.MouseEvent, id: string) => {
@@ -115,7 +123,10 @@ export function LeftPanel({ activeTab }: LeftPanelProps) {
               return (
                 <div
                   key={session.id}
-                  onClick={() => setActiveSessionId(session.id)}
+                  onClick={() => {
+                    setActiveSessionId(session.id);
+                    setActiveHeavyPlugin(null);
+                  }}
                   className={`group flex items-center gap-2.5 px-2.5 py-2 rounded-lg cursor-pointer transition-all mb-px ${
                     isActive
                       ? 'bg-bg-card shadow-[0_1px_3px_rgba(0,0,0,0.04)]'
@@ -160,7 +171,8 @@ export function LeftPanel({ activeTab }: LeftPanelProps) {
             })
           )
         ) : (
-          loadingWorkflows ? (
+          <>
+            {loadingWorkflows ? (
               <div className="px-3 py-8 text-center text-text-muted text-xs">{t('workflow.loading')}</div>
             ) : workflows.length === 0 ? (
               <div className="px-3 py-8 text-center text-text-muted text-xs">
@@ -190,9 +202,42 @@ export function LeftPanel({ activeTab }: LeftPanelProps) {
                   </div>
                 );
               })
-          )
+            )}
+          </>
         )}
       </div>
+
+      {isChats && heavyPluginsWithUi.length > 0 && (
+        <div className="border-t border-border-primary px-2 py-3 shrink-0">
+          <div className="px-2.5 text-[10px] font-semibold text-text-muted uppercase tracking-[1.2px] mb-1.5">
+            {t('chat.plugins.title')}
+          </div>
+          {heavyPluginsWithUi.map((p) => {
+            const isActive = activeHeavyPlugin === p.name;
+            return (
+              <button
+                key={p.name}
+                onClick={() => {
+                  setActiveHeavyPlugin(p.name);
+                  setActiveSessionId(null);
+                }}
+                className={`w-full flex items-center gap-2.5 px-2.5 py-2 rounded-lg transition-all mb-px ${
+                  isActive
+                    ? 'bg-bg-card shadow-[0_1px_3px_rgba(0,0,0,0.04)]'
+                    : 'hover:bg-white/60 dark:hover:bg-white/[0.04]'
+                }`}
+              >
+                <div className={`w-7 h-7 rounded-[7px] flex items-center justify-center flex-shrink-0 ${isActive ? 'bg-accent-light' : 'bg-bg-primary'}`}>
+                  <Box size={12} className={isActive ? 'text-accent' : 'text-text-muted'} />
+                </div>
+                <span className={`text-xs truncate ${isActive ? 'text-text-primary font-medium' : 'text-text-secondary'}`}>
+                  {p.display_name}
+                </span>
+              </button>
+            );
+          })}
+        </div>
+      )}
     </div>
   );
 }

frontend/src/components/Plugin/HeavyPluginList.tsx

@@ -0,0 +1,87 @@
+import { useEffect, useState } from 'react';
+import { useTranslation } from 'react-i18next';
+import { Boxes, Loader2, Package } from 'lucide-react';
+import apiClient from '../../api/client';
+
+interface HeavyPlugin {
+  name: string;
+  display_name: string;
+  description: string;
+  status: string;
+}
+
+export function HeavyPluginList() {
+  const { t } = useTranslation();
+  const [plugins, setPlugins] = useState<HeavyPlugin[]>([]);
+  const [loading, setLoading] = useState(true);
+  const [error, setError] = useState('');
+
+  useEffect(() => {
+    const fetchPlugins = async () => {
+      setLoading(true);
+      try {
+        const resp = await apiClient.get('/api/v1/plugin/list');
+        setPlugins(resp.data.plugins || []);
+        setError('');
+      } catch (err) {
+        console.error('Failed to fetch heavy plugins:', err);
+        setError(t('plugin.heavyPluginLoadFailed'));
+        setPlugins([]);
+      } finally {
+        setLoading(false);
+      }
+    };
+    fetchPlugins();
+  }, [t]);
+
+  return (
+    <div className="max-w-4xl mx-auto space-y-6">
+      <div>
+        <h3 className="text-lg font-bold text-text-primary">{t('plugin.heavyPluginManagement')}</h3>
+        <p className="text-sm text-text-muted mt-0.5">{t('plugin.heavyPluginDesc')}</p>
+      </div>
+
+      {loading ? (
+        <div className="flex flex-col items-center justify-center py-12 text-text-muted">
+          <Loader2 size={24} className="animate-spin mb-3" />
+          <span className="text-sm">{t('common.loading')}</span>
+        </div>
+      ) : error ? (
+        <div className="p-3 bg-danger-bg text-danger text-sm rounded-xl border border-danger/20">{error}</div>
+      ) : plugins.length === 0 ? (
+        <div className="flex flex-col items-center justify-center py-12 bg-bg-card rounded-2xl border border-border-primary border-dashed text-text-muted">
+          <Boxes size={32} className="mb-3 opacity-40" />
+          <span className="text-sm">{t('plugin.heavyPluginEmpty')}</span>
+          <span className="text-[11px] mt-1.5 opacity-70">{t('plugin.heavyPluginEmptyHint')}</span>
+        </div>
+      ) : (
+        <div className="grid grid-cols-1 md:grid-cols-2 gap-4">
+          {plugins.map((p) => (
+            <div key={p.name} className="bg-bg-card border border-border-primary rounded-2xl p-5 card-hover">
+              <div className="flex items-start gap-3 mb-3">
+                <div className="w-9 h-9 rounded-xl bg-bg-secondary border border-border-secondary flex items-center justify-center text-accent shrink-0">
+                  <Package size={18} />
+                </div>
+                <div className="flex-1 min-w-0">
+                  <h4 className="font-semibold text-sm text-text-primary truncate">{p.display_name || p.name}</h4>
+                  <span className="text-[10px] text-text-muted font-mono">{p.name}</span>
+                </div>
+                <span className={`flex items-center gap-1 text-[10px] font-medium px-2 py-1 rounded-lg border ${
+                  p.status === 'running'
+                    ? 'bg-success-bg text-success border-success/20'
+                    : 'bg-bg-secondary text-text-muted border-border-primary'
+                }`}>
+                  {p.status === 'running' && <span className="w-1 h-1 rounded-full bg-success" />}
+                  {p.status}
+                </span>
+              </div>
+              {p.description && (
+                <p className="text-xs text-text-secondary leading-relaxed">{p.description}</p>
+              )}
+            </div>
+          ))}
+        </div>
+      )}
+    </div>
+  );
+}

frontend/src/components/Plugin/PluginLayout.tsx

@@ -1,10 +1,11 @@
 import { useState } from 'react';
 import { useTranslation } from 'react-i18next';
-import { Sparkles, Plug } from 'lucide-react';
+import { Sparkles, Plug, Boxes } from 'lucide-react';
 import { SkillSettings } from './SkillSettings';
 import { MCPSettings } from './MCPSettings';
+import { HeavyPluginList } from './HeavyPluginList';
 
-type PluginTab = 'skill' | 'mcp';
+type PluginTab = 'skill' | 'mcp' | 'heavy';
 
 export function PluginLayout() {
   const { t } = useTranslation();
@@ -13,6 +14,7 @@ export function PluginLayout() {
   const tabs: { key: PluginTab; label: string; icon: typeof Sparkles }[] = [
     { key: 'skill', label: t('plugin.skillTab'), icon: Sparkles },
     { key: 'mcp', label: t('plugin.mcpTab'), icon: Plug },
+    { key: 'heavy', label: t('plugin.heavyTab'), icon: Boxes },
   ];
 
   return (
@@ -39,6 +41,7 @@ export function PluginLayout() {
       <div className="flex-1 overflow-y-auto p-8">
         {tab === 'skill' && <SkillSettings />}
         {tab === 'mcp' && <MCPSettings />}
+        {tab === 'heavy' && <HeavyPluginList />}
       </div>
     </div>
   );

frontend/src/i18n/locales/en.json

@@ -9,6 +9,7 @@
     "chat": "Chat",
     "workflow": "Workflow",
     "plugin": "Plugin",
+    "heavyPlugin": "Plugins",
     "agents": "Agents",
     "toolsets": "Toolsets",
     "config": "Config",
@@ -50,6 +51,9 @@
       "writeCode": "Write code",
       "summarize": "Summarize doc",
       "search": "Search info"
+    },
+    "plugins": {
+      "title": "Heavy Plugins"
     }
   },
   "workflow": {
@@ -251,7 +255,11 @@
     "providerAdvanced": "Parameter Settings",
     "providerCustomModels": "Custom Model List",
     "providerCustomModelsPlaceholder": "Comma separated, e.g. gpt-4o, gpt-4o-mini",
-    "providerCustomModelsHint": "Optional. Leave empty to auto-fetch model list from provider."
+    "providerCustomModelsHint": "Optional. Leave empty to auto-fetch model list from provider.",
+    "providerModelSettings": "Model Call Parameters",
+    "providerModelSettingsHint": "JSON object. Keys are model_id or __default__, values are ModelSettings (temperature/max_tokens/thinking etc.). __default__ is the fallback; specific model_id overrides it.",
+    "providerModelSettingsPlaceholder": "{\n  \"__default__\": {\"temperature\": 0.7, \"max_tokens\": 4096},\n  \"o1\": {\"thinking\": \"high\"}\n}",
+    "providerModelSettingsInvalid": "Invalid JSON. Please check the syntax."
   },
   "plugin": {
     "toolManagement": "Toolset Center",
@@ -284,6 +292,12 @@
     "install": "Install",
     "skillTab": "Skills",
     "mcpTab": "MCP Servers",
+    "heavyTab": "Heavy Plugins",
+    "heavyPluginManagement": "Heavy Plugin Management",
+    "heavyPluginDesc": "Heavy plugins are extension modules with UI and multi-agent collaboration. Loaded from data/plugin/ directory.",
+    "heavyPluginEmpty": "No heavy plugins loaded",
+    "heavyPluginEmptyHint": "Place plugin directories in data/plugin/ and they will load on startup",
+    "heavyPluginLoadFailed": "Failed to load heavy plugin list",
     "mcpManagement": "MCP Server Management",
     "mcpDesc": "Manage Model Context Protocol servers to extend agent tools",
     "mcpAdd": "Add MCP Server",

frontend/src/i18n/locales/zh.json

@@ -9,6 +9,7 @@
     "chat": "对话",
     "workflow": "工作流",
     "plugin": "插件",
+    "heavyPlugin": "重型插件",
     "agents": "智能体",
     "toolsets": "工具集",
     "config": "配置",
@@ -50,6 +51,9 @@
       "writeCode": "写代码",
       "summarize": "总结文档",
       "search": "查找资料"
+    },
+    "plugins": {
+      "title": "重型插件"
     }
   },
   "workflow": {
@@ -251,7 +255,11 @@
     "providerAdvanced": "参数设置",
     "providerCustomModels": "自定义模型列表",
     "providerCustomModelsPlaceholder": "用逗号分隔，如：gpt-4o, gpt-4o-mini",
-    "providerCustomModelsHint": "可选，留空则自动从供应商拉取模型清单"
+    "providerCustomModelsHint": "可选，留空则自动从供应商拉取模型清单",
+    "providerModelSettings": "模型调用参数",
+    "providerModelSettingsHint": "JSON 对象，键是 model_id 或 __default__，值是 ModelSettings（temperature/max_tokens/thinking 等）。__default__ 是兜底参数，具体 model_id 会覆盖之。",
+    "providerModelSettingsPlaceholder": "{\n  \"__default__\": {\"temperature\": 0.7, \"max_tokens\": 4096},\n  \"o1\": {\"thinking\": \"high\"}\n}",
+    "providerModelSettingsInvalid": "JSON 格式错误，请检查"
   },
   "plugin": {
     "toolManagement": "工具集中心",
@@ -284,6 +292,12 @@
     "install": "安装",
     "skillTab": "技能",
     "mcpTab": "MCP 服务",
+    "heavyTab": "重型插件",
+    "heavyPluginManagement": "重型插件管理",
+    "heavyPluginDesc": "重型插件是带 UI 与多 Agent 协作能力的扩展模块，从 data/plugin/ 目录加载",
+    "heavyPluginEmpty": "暂无已加载的重型插件",
+    "heavyPluginEmptyHint": "把插件目录放进 data/plugin/，启动时会自动加载",
+    "heavyPluginLoadFailed": "加载重型插件列表失败",
     "mcpManagement": "MCP 服务管理",
     "mcpDesc": "管理 Model Context Protocol 服务器，扩展 agent 工具能力",
     "mcpAdd": "添加 MCP 服务",

frontend/src/plugins/HeavyPluginShell.tsx

@@ -0,0 +1,109 @@
+import { useEffect, useRef, useState } from 'react';
+import { ChevronLeft, Loader2 } from 'lucide-react';
+import { useAppStore } from '../store/useAppStore';
+import apiClient from '../api/client';
+
+interface UiManifest {
+  name: string;
+  tag: string;
+  js: string;
+  css?: string[];
+  display_name?: string;
+  icon?: string | null;
+}
+
+interface Props {
+  name: string;
+}
+
+// 已加载过的插件 JS：避免切换插件时重复 import 同一个 module（customElements 也不允许重复 define）
+const loadedJs = new Set<string>();
+
+export function HeavyPluginShell({ name }: Props) {
+  const setActiveHeavyPlugin = useAppStore((s) => s.setActiveHeavyPlugin);
+  const meta = useAppStore((s) => s.installedHeavyPlugins.find((p) => p.name === name));
+  const resolvedTheme = useAppStore((s) => s.resolvedTheme);
+  const [manifest, setManifest] = useState<UiManifest | null>(null);
+  const [error, setError] = useState<string | null>(null);
+  const hostRef = useRef<HTMLDivElement | null>(null);
+
+  // 取 token：和 apiClient 拦截器同源（localStorage 'token'，详见 frontend/src/api/client.ts）
+  const token =
+    typeof window !== 'undefined' ? localStorage.getItem('token') ?? '' : '';
+
+  // 1. 拉取 ui-manifest 并按需注入 JS（CSS 在 Web Component 内部 inline 注入，这里不管）
+  useEffect(() => {
+    let canceled = false;
+    setManifest(null);
+    setError(null);
+    (async () => {
+      try {
+        const resp = await apiClient.get<UiManifest>(`/api/v1/plugin/${name}/ui-manifest`);
+        const m = resp.data;
+        if (!loadedJs.has(m.js)) {
+          // 动态 import 插件 ESM bundle —— @vite-ignore 避免 vite 把这个 URL 当成构建期资源
+          await import(/* @vite-ignore */ m.js);
+          loadedJs.add(m.js);
+        }
+        if (!canceled) setManifest(m);
+      } catch (e: unknown) {
+        if (canceled) return;
+        const msg =
+          (e as { response?: { data?: { detail?: string } } }).response?.data?.detail ||
+          (e as Error).message ||
+          'failed to load plugin UI';
+        setError(msg);
+      }
+    })();
+    return () => {
+      canceled = true;
+    };
+  }, [name]);
+
+  // 2. 把 token / api-base / data-theme 三个属性写到自定义元素上
+  // 创建/复用一个元素 placeholder：每次 manifest 切换时清空 host 重新挂载
+  useEffect(() => {
+    const host = hostRef.current;
+    if (!host || !manifest) return;
+    host.innerHTML = '';
+    const el = document.createElement(manifest.tag);
+    el.setAttribute('token', token);
+    el.setAttribute('api-base', '');
+    el.setAttribute('data-theme', resolvedTheme);
+    (el as HTMLElement).style.cssText = 'display:block;width:100%;height:100%';
+    host.appendChild(el);
+    return () => {
+      host.contains(el) && host.removeChild(el);
+    };
+  }, [manifest, token, resolvedTheme]);
+
+  return (
+    <div className="flex-1 flex flex-col overflow-hidden">
+      <div className="flex items-center gap-2 px-4 py-2 border-b border-border-primary bg-bg-card">
+        <button
+          onClick={() => setActiveHeavyPlugin(null)}
+          className="flex items-center gap-1 text-xs text-text-muted hover:text-accent transition-all"
+        >
+          <ChevronLeft size={14} />
+          返回会话
+        </button>
+        <span className="ml-2 text-sm font-medium text-text-primary">
+          {meta?.display_name || manifest?.display_name || name}
+        </span>
+      </div>
+      <div className="flex-1 overflow-hidden relative">
+        {error ? (
+          <div className="p-8 text-danger text-sm">
+            插件 <code className="text-accent">{name}</code> 加载失败：{error}
+          </div>
+        ) : !manifest ? (
+          <div className="absolute inset-0 flex items-center justify-center text-text-muted">
+            <Loader2 size={20} className="animate-spin" />
+          </div>
+        ) : (
+          <div ref={hostRef} className="w-full h-full" />
+        )}
+      </div>
+    </div>
+  );
+}

frontend/src/store/useAppStore.ts

@@ -6,6 +6,15 @@ type WorkTab = 'chat' | 'workflow';
 type AgentTab = 'plugin' | 'agents' | 'toolsets' | 'config' | 'logs';
 type ThemeMode = 'light' | 'dark' | 'system';
 
+export interface InstalledPlugin {
+  name: string;
+  display_name: string;
+  description: string;
+  status: string;
+  has_ui?: boolean;
+  icon?: string | null;
+}
+
 interface AppState {
   // Auth
   isAuthenticated: boolean;
@@ -44,6 +53,12 @@ interface AppState {
   // Locale
   locale: string;
   setLocale: (locale: string) => void;
+
+  // Heavy plugin (workspace dashboards)
+  activeHeavyPlugin: string | null;
+  setActiveHeavyPlugin: (name: string | null) => void;
+  installedHeavyPlugins: InstalledPlugin[];
+  loadInstalledPlugins: () => Promise<void>;
 }
 
 function resolveTheme(theme: ThemeMode): 'light' | 'dark' {
@@ -107,6 +122,21 @@ export const useAppStore = create<AppState>()(
         // 同步到 i18next，确保刷新后语言一致
         import('i18next').then((i18n) => i18n.default.changeLanguage(locale));
       },
+
+      activeHeavyPlugin: null,
+      setActiveHeavyPlugin: (activeHeavyPlugin) => set({ activeHeavyPlugin }),
+      installedHeavyPlugins: [],
+      loadInstalledPlugins: async () => {
+        try {
+          const apiClient = (await import('../api/client')).default;
+          const resp = await apiClient.get('/api/v1/plugin/list');
+          const list: InstalledPlugin[] = resp.data?.plugins || [];
+          set({ installedHeavyPlugins: list.filter((p) => p.has_ui) });
+        } catch (e) {
+          console.warn('Failed to fetch installed plugins', e);
+          set({ installedHeavyPlugins: [] });
+        }
+      },
     }),
     {
       name: 'kilostar-app-storage',

frontend/src/types/index.ts

@@ -22,6 +22,7 @@ export interface Provider {
   provider_status?: string;
   status?: string;
   model?: string;
+  model_settings?: Record<string, Record<string, unknown>>;
 }
 
 export interface ProviderRegisterRequest {
@@ -29,6 +30,7 @@ export interface ProviderRegisterRequest {
   provider_title: string;
   provider_url: string;
   provider_apikey: string;
+  model_settings?: Record<string, Record<string, unknown>>;
 }
 
 export interface ProviderListResponse {

frontend/tsconfig.app.json

@@ -19,7 +19,13 @@
     "noUnusedLocals": true,
     "noUnusedParameters": true,
     "erasableSyntaxOnly": true,
-    "noFallthroughCasesInSwitch": true
+    "noFallthroughCasesInSwitch": true,
+
+    "baseUrl": ".",
+    "ignoreDeprecations": "6.0",
+    "paths": {
+      "@app/*": ["src/*"]
+    }
   },
   "include": ["src"]
 }

frontend/vite.config.ts

@@ -1,8 +1,14 @@
 import { defineConfig } from 'vite'
 import react from '@vitejs/plugin-react'
 import tailwindcss from '@tailwindcss/vite'
+import path from 'node:path'
 
 // https://vite.dev/config/
 export default defineConfig({
   plugins: [react(), tailwindcss()],
+  resolve: {
+    alias: {
+      '@app': path.resolve(__dirname, 'src'),
+    },
+  },
 })

kilostar/adapter/model_adapter/agent_factory.py

@@ -12,7 +12,7 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 
-from typing import Sequence, Any
+from typing import Sequence, Any, Dict
 
 from pydantic_ai import Agent
 from pydantic_ai.models.openai import OpenAIChatModel
@@ -24,8 +24,9 @@ from pydantic_ai.providers.deepseek import DeepSeekProvider
 from pydantic_ai.providers.google import GoogleProvider
 from pydantic_ai.toolsets import AbstractToolset
 
+from pydantic import BaseModel
+
 from kilostar.core.global_state_machine.model_provider import Provider
-from kilostar.utils.agent_model import ResponseModel, DepsModel
 from kilostar.utils.error import ModelNotExistError
 
 
@@ -65,9 +66,9 @@ class AgentFactory:
         self,
         provider: Provider,
         model_id: str,
-        output_type: ResponseModel,
+        output_type: type[BaseModel],
         system_prompt: str,
-        deps_type: DepsModel,
+        deps_type: type[BaseModel],
         agent_name: str,
         tools: list = None,
         toolsets: Sequence[AbstractToolset[Any]] = None,
@@ -126,3 +127,19 @@ class AgentFactory:
             toolsets=toolsets or [],
         )
         return agent
+
+    @staticmethod
+    def resolve_model_settings(provider: Provider, model_id: str) -> Dict[str, Any]:
+        """合并 provider.model_settings 中 ``__default__`` 与具体 model_id 的参数。
+
+        - ``__default__`` 是全 Provider 的兜底参数
+        - ``model_id`` 键覆盖 default 中相同 key
+        - 都缺省时返回空 dict（``agent.run(model_settings={})`` 等效于不传）
+        """
+        settings = getattr(provider, "model_settings", None) or {}
+        if not isinstance(settings, dict):
+            return {}
+        default = settings.get("__default__", {}) or {}
+        specific = settings.get(model_id, {}) or {}
+        merged: Dict[str, Any] = {**default, **specific}
+        return merged

kilostar/api/__init__.py

@@ -13,19 +13,14 @@
 #  limitations under the License.
 
 import os
-from typing import Dict
 
-from fastapi import FastAPI, WebSocket, Request
+from fastapi import FastAPI, Request
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import FileResponse, JSONResponse
 from fastapi.staticfiles import StaticFiles
 
-from kilostar.utils.ray_compat import _STANDALONE
 from kilostar.utils.settings import get_settings
 
-if not _STANDALONE:
-    from ray import serve
-
 from .agent import agent_router
 from .auth import auth_router
 from .system import system_router, system_api_router
@@ -36,6 +31,7 @@ from .resource import resource_router
 from .workflow import workflow_router
 from .chat import chat_router
 from .plugin import plugin_router
+from .task import task_router
 from kilostar.utils.error import (
     KiloStarError,
     BusinessError,
@@ -105,6 +101,18 @@ app.include_router(agent_router)  # agent路径
 app.include_router(workflow_router)  # workflow路径
 app.include_router(chat_router)  # chat路径
 app.include_router(plugin_router)  # plugin路径
+app.include_router(task_router)  # 短任务路径
+
+
+# 重型插件：纯文件扫描挂载各插件自带的 ``api.py`` router（不依赖 actor 启动顺序）
+try:
+    from kilostar.plugin_runtime.loader import collect_plugin_routers
+    from kilostar.utils.settings import get_plugin_dir as _get_plugin_dir
+
+    for _prefix, _plugin_router in collect_plugin_routers(_get_plugin_dir()):
+        app.include_router(_plugin_router, prefix=_prefix)
+except Exception as _e:
+    _api_logger.warning(f"failed to mount plugin routers: {_e}")
 
 
 @app.exception_handler(BusinessError)
@@ -147,6 +155,30 @@ base_dir = os.path.dirname(
 )
 frontend_dir = os.path.join(base_dir, "frontend", "dist")
 
+# 重型插件 UI：扫描 ``data/plugin/*/frontend/dist/``，按目录名挂成静态资源
+# （/plugin-ui/<name>/ → <plugin>/frontend/dist/）。html=True 让目录访问回退到 index.html
+# （单页应用风格），但实际我们靠 wc-manifest.json + import() 加载，不依赖该回退。
+try:
+    from kilostar.utils.settings import get_plugin_dir as _get_plugin_dir_for_ui
+
+    _plugin_root_for_ui = _get_plugin_dir_for_ui()
+    if _plugin_root_for_ui.exists():
+        for _p in _plugin_root_for_ui.iterdir():
+            if not _p.is_dir():
+                continue
+            _dist = _p / "frontend" / "dist"
+            if not _dist.is_dir():
+                continue
+            app.mount(
+                f"/plugin-ui/{_p.name}",
+                StaticFiles(directory=str(_dist), html=True),
+                name=f"plugin_ui_{_p.name}",
+            )
+            _api_logger.info(f"mounted plugin UI: /plugin-ui/{_p.name} → {_dist}")
+except Exception as _e:
+    _api_logger.warning(f"failed to mount plugin UIs: {_e}")
+
+
 if os.path.exists(frontend_dir):
     app.mount(
         "/assets",
@@ -184,11 +216,3 @@ else:
     )
 
 
-if not _STANDALONE:
-    @serve.deployment
-    @serve.ingress(app)
-    class KiloStarGateway:
-        gateway: Dict[str, WebSocket]
-
-        def __init__(self):
-            self.gateway = {}

kilostar/api/agent.py

@@ -14,7 +14,7 @@
 
 
 from typing import Union
-from kilostar.utils.ray_hook import ray_actor_hook
+from kilostar.utils.actor import get_actor
 from fastapi import APIRouter, Depends, Request
 from pydantic import BaseModel, field_validator
 from kilostar.utils.access import Accessor, TokenData, RoleChecker
@@ -52,8 +52,8 @@ async def get_system_nodes(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
     """返回两大系统节点（regulatory/consciousness）当前的持久化配置。"""
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    configs = await postgres_database.get_all_system_node_configs.remote()
+    postgres_database = get_actor("postgres_database")
+    configs = await postgres_database.get_all_system_node_configs()
     return {"system_nodes": configs}
 
 
@@ -64,8 +64,8 @@ async def load_agent(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.SUPER_ADMINISTRATOR)),
 ):
     """加载/重载某个系统节点的 Agent：先持久化配置，再调用对应节点 Actor 的 ``create_agent``。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
+    global_state_machine = get_actor("global_state_machine")
+    postgres_database = get_actor("postgres_database")
     accept_lang = request.headers.get("accept-language", "")
 
     if isinstance(agent_register, AgentLocalRegister):
@@ -73,7 +73,7 @@ async def load_agent(
 
     elif isinstance(agent_register, AgentRegister):
         try:
-            await postgres_database.upsert_system_node_config.remote(
+            await postgres_database.upsert_system_node_config(
                 agent_register.individual_name,
                 agent_register.provider_title,
                 agent_register.model_id,
@@ -90,14 +90,14 @@ async def load_agent(
             # Resolve persona system_prompt from DB
             persona_prompt = None
             if agent_register.persona_id:
-                tpl = await postgres_database.get_template.remote(agent_register.persona_id)
+                tpl = await postgres_database.get_template(agent_register.persona_id)
                 if tpl:
                     persona_prompt = tpl.system_prompt
 
             match scope:
                 case "regulatory_node":
-                    node = ray_actor_hook("regulatory_node").regulatory_node
-                    await node.create_agent.remote(
+                    node = get_actor("regulatory_node")
+                    await node.create_agent(
                         global_state_machine,
                         agent_register.provider_title,
                         agent_register.model_id,
@@ -107,8 +107,8 @@ async def load_agent(
                         persona_prompt,
                     )
                 case "consciousness_node":
-                    node = ray_actor_hook("consciousness_node").consciousness_node
-                    await node.create_agent.remote(
+                    node = get_actor("consciousness_node")
+                    await node.create_agent(
                         global_state_machine,
                         agent_register.provider_title,
                         agent_register.model_id,
@@ -181,10 +181,10 @@ async def create_worker_individual(
     token_data: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
     """创建一个 Worker Agent，``owner_id`` 自动绑定为当前登录用户。"""
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
+    postgres_database = get_actor("postgres_database")
     data_dict = worker_data.model_dump()
     data_dict["owner_id"] = token_data.user_id
-    worker = await postgres_database.add_worker_individual.remote(**data_dict)
+    worker = await postgres_database.add_worker_individual(**data_dict)
     return {"message": "success", "agent_id": worker.agent_id}
 
 
@@ -192,24 +192,34 @@ async def create_worker_individual(
 async def get_worker_individual_list(
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    """列出当前登录用户名下的全部 Worker Agent。"""
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    workers = await postgres_database.get_worker_individual_list.remote(
+    """列出当前登录用户名下的全部 Worker Agent，并附加所有 plugin_owned slot。
+
+    plugin_owned slot 是插件登记的"占位 agent"，所有用户共享同一份配置，
+    在前端展示时会用徽标标记，并允许任何登录用户装配 provider/model。
+    """
+    postgres_database = get_actor("postgres_database")
+    workers = await postgres_database.get_worker_individual_list(
         owner_id=token_data.user_id
-    )
-    return {"workers": workers}
+    ) or []
+    all_workers = await postgres_database.get_all_worker_individual() or []
+    seen_ids = {w.agent_id for w in workers}
+    plugin_slots = [
+        w for w in all_workers
+        if getattr(w, "plugin_owned", None) and w.agent_id not in seen_ids
+    ]
+    return {"workers": list(workers) + plugin_slots}
 
 
 @agent_router.get("/worker/{agent_id}")
 async def get_worker_individual(
     agent_id: str, token_data: TokenData = Depends(Accessor.get_current_user)
 ):
-    """按 ``agent_id`` 查询 Worker Agent；非本人的 Agent 返回 403。"""
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    worker = await postgres_database.get_worker_individual.remote(agent_id=agent_id)
+    """按 ``agent_id`` 查询 Worker Agent；非本人的 Agent 返回 403（plugin_owned slot 例外）。"""
+    postgres_database = get_actor("postgres_database")
+    worker = await postgres_database.get_worker_individual(agent_id=agent_id)
     if not worker:
         raise HTTPException(status_code=404, detail="Agent not found")
-    if worker.owner_id != token_data.user_id:
+    if not getattr(worker, "plugin_owned", None) and worker.owner_id != token_data.user_id:
         raise HTTPException(
             status_code=403, detail="Forbidden: You do not own this agent"
         )
@@ -223,23 +233,32 @@ async def update_worker_individual(
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
     """局部更新 Worker Agent 配置；同时把状态机里的旧实例移除等待懒加载。"""
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    worker = await postgres_database.get_worker_individual.remote(agent_id=agent_id)
+    postgres_database = get_actor("postgres_database")
+    worker = await postgres_database.get_worker_individual(agent_id=agent_id)
     if not worker:
         raise HTTPException(status_code=404, detail="Agent not found")
-    if worker.owner_id != token_data.user_id:
+    # plugin_owned slot：任何登录用户都能装配 provider/model；普通 worker 仅 owner 可改
+    if not getattr(worker, "plugin_owned", None) and worker.owner_id != token_data.user_id:
         raise HTTPException(
             status_code=403, detail="Forbidden: You do not own this agent"
         )
 
     update_data = worker_data.model_dump(exclude_unset=True)
-    updated_worker = await postgres_database.update_worker_individual.remote(
+    updated_worker = await postgres_database.update_worker_individual(
         agent_id=agent_id, **update_data
     )
 
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
+    global_state_machine = get_actor("global_state_machine")
     try:
-        await global_state_machine.remove_individual.remote(agent_id)
+        await global_state_machine.remove_individual(agent_id)
+    except Exception:
+        pass
+
+    # plugin_owned 时顺带触发对应插件的 reload，让新 provider/model 立刻生效
+    if getattr(worker, "plugin_owned", None):
+        try:
+            pm = get_actor("global_plugin_manager")
+            await pm.reload(worker.plugin_owned)
         except Exception:
             pass
 
@@ -251,8 +270,8 @@ async def reload_worker_individual(
     agent_id: str, token_data: TokenData = Depends(Accessor.get_current_user)
 ):
     """强制把 Worker 从内存池中卸载，下次调用时按最新配置重新加载。"""
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    worker = await postgres_database.get_worker_individual.remote(agent_id=agent_id)
+    postgres_database = get_actor("postgres_database")
+    worker = await postgres_database.get_worker_individual(agent_id=agent_id)
     if not worker:
         raise HTTPException(status_code=404, detail="Agent not found")
     if worker.owner_id != token_data.user_id:
@@ -260,8 +279,8 @@ async def reload_worker_individual(
             status_code=403, detail="Forbidden: You do not own this agent"
         )
 
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    await global_state_machine.remove_individual.remote(agent_id)
+    global_state_machine = get_actor("global_state_machine")
+    await global_state_machine.remove_individual(agent_id)
 
     return {"message": "Worker will be reloaded on next use"}
 
@@ -271,15 +290,15 @@ async def delete_worker_individual(
     agent_id: str, token_data: TokenData = Depends(Accessor.get_current_user)
 ):
     """删除 Worker Agent；非本人 Agent 返回 403。"""
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    worker = await postgres_database.get_worker_individual.remote(agent_id=agent_id)
+    postgres_database = get_actor("postgres_database")
+    worker = await postgres_database.get_worker_individual(agent_id=agent_id)
     if not worker:
         raise HTTPException(status_code=404, detail="Agent not found")
     if worker.owner_id != token_data.user_id:
         raise HTTPException(
             status_code=403, detail="Forbidden: You do not own this agent"
         )
-    await postgres_database.delete_worker_individual.remote(agent_id=agent_id)
+    await postgres_database.delete_worker_individual(agent_id=agent_id)
     return {"message": "success"}
 
 
@@ -299,8 +318,8 @@ class PersonaTemplateUpdate(BaseModel):
 async def list_templates(
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    templates = await postgres_database.list_templates.remote(
+    postgres_database = get_actor("postgres_database")
+    templates = await postgres_database.list_templates(
         owner_id=token_data.user_id
     )
     return {"templates": templates}
@@ -311,8 +330,8 @@ async def create_template(
     data: PersonaTemplateCreate,
     token_data: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    tpl = await postgres_database.add_template.remote(
+    postgres_database = get_actor("postgres_database")
+    tpl = await postgres_database.add_template(
         name=data.name,
         system_prompt=data.system_prompt,
         owner_id=token_data.user_id,
@@ -326,13 +345,13 @@ async def update_template(
     data: PersonaTemplateUpdate,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    tpl = await postgres_database.get_template.remote(template_id)
+    postgres_database = get_actor("postgres_database")
+    tpl = await postgres_database.get_template(template_id)
     if not tpl:
         raise HTTPException(status_code=404, detail="Template not found")
     if tpl.owner_id != token_data.user_id:
         raise HTTPException(status_code=403, detail="Forbidden")
-    updated = await postgres_database.update_template.remote(
+    updated = await postgres_database.update_template(
         template_id, **data.model_dump(exclude_unset=True)
     )
     return {"message": "success", "template": updated}
@@ -343,11 +362,11 @@ async def delete_template(
     template_id: str,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    tpl = await postgres_database.get_template.remote(template_id)
+    postgres_database = get_actor("postgres_database")
+    tpl = await postgres_database.get_template(template_id)
     if not tpl:
         raise HTTPException(status_code=404, detail="Template not found")
     if tpl.owner_id != token_data.user_id:
         raise HTTPException(status_code=403, detail="Forbidden")
-    await postgres_database.delete_template.remote(template_id)
+    await postgres_database.delete_template(template_id)
     return {"message": "success"}

kilostar/api/auth.py

@@ -17,7 +17,7 @@ from fastapi import Depends
 from pydantic import BaseModel
 from kilostar.utils.access import Accessor, TokenData, RoleChecker
 from fastapi.concurrency import run_in_threadpool
-from kilostar.utils.ray_hook import ray_actor_hook
+from kilostar.utils.actor import get_actor
 from kilostar.core.postgres_database.model import UserAuthority
 from kilostar.utils.error import UserNotExistError
 from kilostar.utils.rate_limit import register_limiter, login_limiter
@@ -36,7 +36,7 @@ class UserRegister(BaseModel):
 async def create_user(user_register: UserRegister, request: Request):
     """注册新用户：异步线程池里做 argon2 哈希，再交由 PostgresDatabase Actor 落库。"""
     register_limiter.check(request)
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
+    postgres_database = get_actor("postgres_database")
     try:
         hashed_password = await run_in_threadpool(
             Accessor.hash_password, user_register.password
@@ -47,7 +47,7 @@ async def create_user(user_register: UserRegister, request: Request):
             status_code=400,
             content={"code": "password_invalid", "message": str(e)},
         )
-    user = await postgres_database.add_user.remote(
+    user = await postgres_database.add_user(
         user_register.user_name, hashed_password
     )
     return {"message": "success", "user_id": user.user_id}
@@ -64,8 +64,8 @@ class UserLogin(BaseModel):
 async def login_user(user_login: UserLogin, request: Request):
     """用户登录：查询用户后在线程池中校验口令，校验成功则签发 JWT。"""
     login_limiter.check(request)
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    user = await postgres_database.login_user.remote(user_login.user_name)
+    postgres_database = get_actor("postgres_database")
+    user = await postgres_database.login_user(user_login.user_name)
     if not user:
         raise UserNotExistError()
     tokens = await run_in_threadpool(
@@ -113,8 +113,8 @@ async def change_authority(
     """
     Update a user's authority level. Only accessible by SUPER_ADMINISTRATOR.
     """
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    user = await postgres_database.change_user_authority.remote(
+    postgres_database = get_actor("postgres_database")
+    user = await postgres_database.change_user_authority(
         user_id=request.user_id, new_authority=request.new_authority
     )
     return {
@@ -133,8 +133,8 @@ async def get_user_list(
     """
     Get a list of all users. Only accessible by SUPER_ADMINISTRATOR.
     """
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    users = await postgres_database.get_all_users.remote()
+    postgres_database = get_actor("postgres_database")
+    users = await postgres_database.get_all_users()
     return {
         "users": [
             {"user_id": u.user_id, "user_name": u.user_name, "role": u.user_authority}
@@ -153,6 +153,6 @@ async def delete_user(
     """
     Delete a user. Only accessible by SUPER_ADMINISTRATOR.
     """
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    await postgres_database.delete_user_by_id.remote(user_id=user_id)
+    postgres_database = get_actor("postgres_database")
+    await postgres_database.delete_user_by_id(user_id=user_id)
     return {"message": "success"}

kilostar/api/chat.py

@@ -17,7 +17,7 @@ import asyncio
 from fastapi import APIRouter, Depends, HTTPException, Request
 from fastapi.responses import StreamingResponse
 from pydantic import BaseModel
-from kilostar.utils.ray_hook import ray_actor_hook
+from kilostar.utils.actor import get_actor
 from kilostar.utils.access import Accessor, TokenData
 from kilostar.core.individual.regulatory_node.template import (
     MessageRequest,
@@ -56,8 +56,8 @@ def _build_message_history(rows) -> list:
 
 
 async def _load_message_history(chat_id: str) -> list:
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    rows = await postgres_database.list_chat_messages.remote(chat_id=chat_id)
+    postgres_database = get_actor("postgres_database")
+    rows = await postgres_database.list_chat_messages(chat_id=chat_id)
     return _build_message_history(rows or [])
 
 
@@ -76,14 +76,14 @@ async def _ask_regulatory(
     *, user_id: str, chat_id: str, message: str, message_history: list | None = None
 ) -> str | None:
     """统一封装 chat 入口对 RegulatoryNode 的调用。"""
-    regulatory_node = ray_actor_hook("regulatory_node").regulatory_node
+    regulatory_node = get_actor("regulatory_node")
     payload = MessageRequest(
         platform="client",
         user_name=user_id,
         platform_id=chat_id,
         message=message,
     )
-    resp: MessageResponse | None = await regulatory_node.working.remote(
+    resp: MessageResponse | None = await regulatory_node.working(
         payload, message_history
     )
     return _extract_reply(resp)
@@ -103,13 +103,13 @@ async def create_chat_session(
     request: CreateChatRequest,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    chat = await postgres_database.create_chat_session.remote(
+    postgres_database = get_actor("postgres_database")
+    chat = await postgres_database.create_chat_session(
         user_id=token_data.user_id, title=request.title
     )
 
     # 存入用户消息
-    await postgres_database.add_chat_message.remote(
+    await postgres_database.add_chat_message(
         chat_id=chat.chat_id, message=request.initial_message, message_owner="user"
     )
 
@@ -122,7 +122,7 @@ async def create_chat_session(
 
     # 存入回复消息
     if response_msg:
-        await postgres_database.add_chat_message.remote(
+        await postgres_database.add_chat_message(
             chat_id=chat.chat_id, message=response_msg, message_owner="regulatory_node"
         )
 
@@ -133,8 +133,8 @@ async def create_chat_session(
 async def list_chat_sessions(
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    sessions = await postgres_database.list_chat_sessions.remote(
+    postgres_database = get_actor("postgres_database")
+    sessions = await postgres_database.list_chat_sessions(
         user_id=token_data.user_id
     )
     return {"sessions": sessions}
@@ -144,8 +144,8 @@ async def list_chat_sessions(
 async def get_chat_history(
     chat_id: str, token_data: TokenData = Depends(Accessor.get_current_user)
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    messages = await postgres_database.list_chat_messages.remote(chat_id=chat_id)
+    postgres_database = get_actor("postgres_database")
+    messages = await postgres_database.list_chat_messages(chat_id=chat_id)
     return {"messages": messages}
 
 
@@ -155,10 +155,10 @@ async def send_chat_message(
     request: SendMessageRequest,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
+    postgres_database = get_actor("postgres_database")
     # 先取历史（不含当前输入），再写入用户消息，避免历史里出现重复
     message_history = await _load_message_history(chat_id)
-    await postgres_database.add_chat_message.remote(
+    await postgres_database.add_chat_message(
         chat_id=chat_id, message=request.message, message_owner="user"
     )
 
@@ -172,7 +172,7 @@ async def send_chat_message(
 
     # 存回复
     if response_msg:
-        await postgres_database.add_chat_message.remote(
+        await postgres_database.add_chat_message(
             chat_id=chat_id, message=response_msg, message_owner="regulatory_node"
         )
 
@@ -184,13 +184,13 @@ async def delete_chat_session(
     chat_id: str,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    session = await postgres_database.get_chat_session.remote(chat_id=chat_id)
+    postgres_database = get_actor("postgres_database")
+    session = await postgres_database.get_chat_session(chat_id=chat_id)
     if not session:
         raise HTTPException(status_code=404, detail="Chat session not found")
     if session.user_id != token_data.user_id:
         raise HTTPException(status_code=403, detail="Forbidden")
-    await postgres_database.delete_chat_session.remote(chat_id=chat_id)
+    await postgres_database.delete_chat_session(chat_id=chat_id)
     return {"message": "success"}
 
 
@@ -201,14 +201,12 @@ async def stream_chat_message(
     request: Request,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    """SSE 流式聊天端点：standalone 模式下逐 token 流式输出；distributed 模式 fallback 到整段回复。"""
-    from kilostar.utils.ray_compat import _STANDALONE
-
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
+    """SSE 流式聊天端点：逐 token 流式输出。"""
+    postgres_database = get_actor("postgres_database")
 
     message_history = await _load_message_history(chat_id)
 
-    await postgres_database.add_chat_message.remote(
+    await postgres_database.add_chat_message(
         chat_id=chat_id, message=request_body.message, message_owner="user"
     )
 
@@ -219,23 +217,12 @@ async def stream_chat_message(
         message=request_body.message,
     )
 
-    regulatory_node = ray_actor_hook("regulatory_node").regulatory_node
-
-    if not _STANDALONE:
-        async def fallback_generator():
-            resp = await regulatory_node.working.remote(payload, message_history)
-            full_response = resp.reply_message if resp else ""
-            if full_response:
-                await postgres_database.add_chat_message.remote(
-                    chat_id=chat_id, message=full_response, message_owner="regulatory_node"
-                )
-            yield f"data: {json.dumps({'token': full_response})}\n\n"
-            yield f"data: {json.dumps({'done': True, 'full_message': full_response})}\n\n"
-
-        return StreamingResponse(fallback_generator(), media_type="text/event-stream")
+    regulatory_node = get_actor("regulatory_node")
 
     token_queue = asyncio.Queue()
-    stream_task = regulatory_node.stream_working.remote(payload, token_queue, message_history)
+    stream_task = asyncio.create_task(
+        regulatory_node.stream_working(payload, token_queue, message_history)
+    )
 
     async def event_generator():
         full_response = ""
@@ -260,7 +247,7 @@ async def stream_chat_message(
                 yield f"data: {json.dumps({'token': full_response})}\n\n"
 
         if full_response:
-            await postgres_database.add_chat_message.remote(
+            await postgres_database.add_chat_message(
                 chat_id=chat_id,
                 message=full_response,
                 message_owner="regulatory_node",

kilostar/api/platform/frontend.py

@@ -15,7 +15,7 @@
 from fastapi import APIRouter, Depends, HTTPException, UploadFile, File
 from pydantic import BaseModel
 from kilostar.utils.access import Accessor, TokenData
-from kilostar.utils.ray_hook import ray_actor_hook
+from kilostar.utils.actor import get_actor
 from kilostar.core.individual.regulatory_node.template import (
     MessageRequest,
     MessageResponse,
@@ -42,14 +42,14 @@ async def create_message(
     """把前端消息转交给 RegulatoryNode 处理，并把回复透传给前端。"""
     logger.info("收到消息，来源：客户端")
     logger.debug(f"消息内容：{message.message}")
-    regulatory_node = ray_actor_hook("regulatory_node").regulatory_node
+    regulatory_node = get_actor("regulatory_node")
     msg_request = MessageRequest(
         platform="client",
         user_name=token_data.username,
         platform_id=token_data.user_id,
         message=message.message,
     )
-    result = await regulatory_node.working.remote(msg_request)
+    result = await regulatory_node.working(msg_request)
     if isinstance(result, MessageResponse):
         return {"message": result.reply_message}
     if isinstance(result, str):

kilostar/api/platform/onebot.py

@@ -39,7 +39,7 @@ from kilostar.core.individual.regulatory_node.template import (
     MessageResponse,
 )
 from kilostar.utils.logger import get_logger
-from kilostar.utils.ray_hook import ray_actor_hook
+from kilostar.utils.actor import get_actor
 
 logger = get_logger("onebot")
 
@@ -129,8 +129,8 @@ async def _dispatch_event(payload: Dict[str, Any]) -> Optional[Dict[str, Any]]:
     )
 
     try:
-        regulatory_node = ray_actor_hook("regulatory_node").regulatory_node
-        result = await regulatory_node.working.remote(msg_request)
+        regulatory_node = get_actor("regulatory_node")
+        result = await regulatory_node.working(msg_request)
     except Exception as e:
         logger.exception(f"[OneBot] RegulatoryNode 调用失败: {e}")
         return None

kilostar/api/plugin.py

@@ -1,12 +1,15 @@
 from __future__ import annotations
 
+import json
 from typing import Optional
+
 from fastapi import APIRouter, Depends, HTTPException
 from fastapi.responses import StreamingResponse
 from pydantic import BaseModel
 
 from kilostar.utils.access import Accessor, TokenData
-from kilostar.utils.ray_hook import ray_actor_hook
+from kilostar.utils.actor import get_actor
+from kilostar.utils.settings import get_plugin_dir
 
 plugin_router = APIRouter(prefix="/api/v1/plugin", tags=["plugin"])
 
@@ -22,15 +25,16 @@ async def submit_task(
     req: SubmitRequest,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    pm = ray_actor_hook("global_plugin_manager").global_plugin_manager
-    plugins = await pm.list_plugins.remote()
-    if req.org_name not in plugins:
+    pm = get_actor("global_plugin_manager")
+    plugins = await pm.list_plugins()
+    # list_plugins 返回 [{"name": ..., ...}]，按 name 判断插件是否已加载
+    if req.org_name not in {p["name"] for p in plugins}:
         raise HTTPException(404, f"Plugin '{req.org_name}' not found")
 
-    org = ray_actor_hook(f"org_{req.org_name}").get(f"org_{req.org_name}")
+    org = get_actor(f"org_{req.org_name}")
     ctx = req.context or {}
     ctx["user"] = token_data.username
-    task_id = await org.submit.remote(req.task_description, ctx)
+    task_id = await org.submit(req.task_description, ctx)
     return {"task_id": task_id}
 
 
@@ -39,8 +43,8 @@ async def get_task_status(
     task_id: str,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    db = ray_actor_hook("postgres_database").postgres_database
-    task = await db.get_org_task.remote(task_id)
+    db = get_actor("postgres_database")
+    task = await db.get_org_task(task_id)
     if not task:
         raise HTTPException(404, "Task not found")
     return task
@@ -51,8 +55,8 @@ async def get_task_events(
     task_id: str,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    db = ray_actor_hook("postgres_database").postgres_database
-    events = await db.query_org_events.remote(task_id)
+    db = get_actor("postgres_database")
+    events = await db.query_org_events(task_id)
     return {"events": events}
 
 
@@ -61,19 +65,16 @@ async def stream_task(
     task_id: str,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    import asyncio
-
-    org_name = None
-    db = ray_actor_hook("postgres_database").postgres_database
-    task = await db.get_org_task.remote(task_id)
+    db = get_actor("postgres_database")
+    task = await db.get_org_task(task_id)
     if not task:
         raise HTTPException(404, "Task not found")
     org_name = task["org_name"]
 
-    org = ray_actor_hook(f"org_{org_name}").get(f"org_{org_name}")
+    org = get_actor(f"org_{org_name}")
 
     async def _generate():
-        async for event in await org.stream.remote(task_id):
+        async for event in await org.stream(task_id):
             yield f"data: {event}\n\n"
 
     return StreamingResponse(_generate(), media_type="text/event-stream")
@@ -83,8 +84,8 @@ async def stream_task(
 async def list_plugins(
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    pm = ray_actor_hook("global_plugin_manager").global_plugin_manager
-    plugins = await pm.list_plugins.remote()
+    pm = get_actor("global_plugin_manager")
+    plugins = await pm.list_plugins()
     return {"plugins": plugins}
 
 
@@ -93,8 +94,8 @@ async def install_plugin(
     name: str,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    pm = ray_actor_hook("global_plugin_manager").global_plugin_manager
-    await pm.install.remote(name)
+    pm = get_actor("global_plugin_manager")
+    await pm.install(name)
     return {"status": "ok", "name": name}
 
 
@@ -103,6 +104,59 @@ async def reload_plugin(
     name: str,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    pm = ray_actor_hook("global_plugin_manager").global_plugin_manager
-    await pm.reload.remote(name)
+    pm = get_actor("global_plugin_manager")
+    await pm.reload(name)
     return {"status": "ok", "name": name}
+
+
+@plugin_router.get("/{name}/ui-manifest")
+async def ui_manifest(
+    name: str,
+    token_data: TokenData = Depends(Accessor.get_current_user),
+):
+    """返回 Web Component 注入所需的元数据。
+
+    读插件 build 产物 ``<plugin>/frontend/dist/wc-manifest.json``，把 ``js`` / ``css`` 路径
+    转成绝对静态路径（与 ``/plugin-ui/<name>/`` 静态挂载对齐）。dist 不存在 → 404。
+    """
+    plugin_dir = get_plugin_dir() / name
+    if not plugin_dir.is_dir():
+        raise HTTPException(404, f"plugin {name!r} not found")
+    wc_path = plugin_dir / "frontend" / "dist" / "wc-manifest.json"
+    if not wc_path.is_file():
+        raise HTTPException(404, f"plugin {name!r} has no built UI")
+    try:
+        wc = json.loads(wc_path.read_text(encoding="utf-8"))
+    except Exception as e:
+        raise HTTPException(500, f"invalid wc-manifest.json: {e}")
+
+    js_rel = wc.get("js")
+    if not js_rel or not isinstance(js_rel, str):
+        raise HTTPException(500, "wc-manifest.json missing 'js'")
+    tag = wc.get("tag")
+    if not tag or not isinstance(tag, str):
+        raise HTTPException(500, "wc-manifest.json missing 'tag'")
+
+    base_url = f"/plugin-ui/{name}"
+    css_rel = wc.get("css") or []
+    if isinstance(css_rel, str):
+        css_rel = [css_rel]
+
+    # 顺便把 manifest.json 的展示元数据带回前端，少一次往返
+    display_name = name
+    icon = None
+    try:
+        manifest_data = json.loads((plugin_dir / "manifest.json").read_text(encoding="utf-8"))
+        display_name = manifest_data.get("display_name") or name
+        icon = (manifest_data.get("ui") or {}).get("icon")
+    except Exception:
+        pass
+
+    return {
+        "name": name,
+        "tag": tag,
+        "js": f"{base_url}/{js_rel.lstrip('/')}",
+        "css": [f"{base_url}/{c.lstrip('/')}" for c in css_rel],
+        "display_name": display_name,
+        "icon": icon,
+    }

kilostar/api/provider.py

@@ -13,12 +13,12 @@
 #  limitations under the License.
 
 from fastapi import APIRouter, Depends
-from pydantic import BaseModel
-from typing import Any, Dict, Literal
+from pydantic import BaseModel, Field
+from typing import Any, Dict, Literal, Optional
 from kilostar.utils.access import TokenData, Accessor, RoleChecker
 from kilostar.core.postgres_database.model import UserAuthority
 from kilostar.core.global_state_machine.model_provider.base_provider import Provider
-from kilostar.utils.ray_hook import ray_actor_hook
+from kilostar.utils.actor import get_actor
 
 provider_router = APIRouter(prefix="/api/v1/provider", tags=["provider"])
 
@@ -30,6 +30,7 @@ class ProviderRegister(BaseModel):
     provider_title: str
     provider_url: str
     provider_apikey: str
+    model_settings: Optional[Dict[str, Dict[str, Any]]] = Field(default=None)
 
 
 @provider_router.post("")
@@ -38,13 +39,14 @@ async def create_provider(
     token_data: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ) -> None:
     """注册一个 Provider；owner 为当前登录用户的 ``user_id``。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    await global_state_machine.add_provider_wrap.remote(
+    global_state_machine = get_actor("global_state_machine")
+    await global_state_machine.add_provider_wrap(
         provider_type=provider_register.provider_type,
         provider_title=provider_register.provider_title,
         provider_url=provider_register.provider_url,
         provider_apikey=provider_register.provider_apikey,
         provider_owner=token_data.user_id,
+        model_settings=provider_register.model_settings or {},
     )
 
 
@@ -59,10 +61,10 @@ async def get_provider_list(
     _: TokenData = Depends(Accessor.get_current_user),
 ) -> Dict[str, Any]:
     """返回当前所有已注册的 Provider，前端用以展示模型清单。apikey 脱敏。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
+    global_state_machine = get_actor("global_state_machine")
     provider_list: Dict[
         str, Provider
-    ] = await global_state_machine.get_provider_list.remote()
+    ] = await global_state_machine.get_provider_list()
     masked = {}
     for title, p in provider_list.items():
         d = p.model_dump() if hasattr(p, "model_dump") else dict(p)
@@ -136,6 +138,6 @@ async def delete_provider(
     ),
 ) -> dict:
     """删除指定 ``provider_title`` 的 Provider；仅超管可调用。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    await global_state_machine.delete_provider.remote(provider_title=provider_title)
+    global_state_machine = get_actor("global_state_machine")
+    await global_state_machine.delete_provider(provider_title=provider_title)
     return {"message": "success"}

kilostar/api/resource.py

@@ -15,7 +15,7 @@
 from typing import Any, Dict, List, Optional
 from pydantic import BaseModel
 import viceroy
-from kilostar.utils.ray_hook import ray_actor_hook
+from kilostar.utils.actor import get_actor
 from fastapi import APIRouter, Depends, HTTPException
 from fastapi.responses import FileResponse
 from kilostar.utils.access import TokenData, RoleChecker, Accessor
@@ -50,7 +50,7 @@ async def install_skill(
     skill: Skill, _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER))
 ):
     """通过 viceroy 把 skill 仓库克隆到 ``data/plugin/skill``，并在状态机中登记。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
+    global_state_machine = get_actor("global_state_machine")
     import os
     from kilostar.utils.settings import get_plugin_dir
 
@@ -63,7 +63,7 @@ async def install_skill(
         skill_name = skill.path.split("/")[-1]
     else:
         skill_name = skill.repo_url.split("/")[-1]
-    await global_state_machine.add_skill.remote(skill_name)
+    await global_state_machine.add_skill(skill_name)
     return {"message": "创建成功"}
 
 
@@ -72,8 +72,8 @@ async def get_skills(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
     """返回当前状态机中已登记的所有 skill 名称列表。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    skills = await global_state_machine.get_skill_list.remote()
+    global_state_machine = get_actor("global_state_machine")
+    skills = await global_state_machine.get_skill_list()
     return {"skills": skills}
 
 
@@ -85,8 +85,8 @@ async def delete_skill(
     ),
 ):
     """从状态机中移除 skill 注册项；不会删除磁盘上的代码文件。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    await global_state_machine.remove_skill.remote(skill_name)
+    global_state_machine = get_actor("global_state_machine")
+    await global_state_machine.remove_skill(skill_name)
     return {"message": "success"}
 
 
@@ -98,12 +98,12 @@ async def add_mcp_server(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.SUPER_ADMINISTRATOR)),
 ):
     """注册一个 MCP 服务器到全局状态机。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
+    global_state_machine = get_actor("global_state_machine")
     import uuid
 
     server_id = str(uuid.uuid4())[:8]
     cfg_dict = config.model_dump(exclude_none=True)
-    await global_state_machine.add_mcp_server.remote(server_id, cfg_dict)
+    await global_state_machine.add_mcp_server(server_id, cfg_dict)
     return {"server_id": server_id, "message": "MCP server registered"}
 
 
@@ -112,8 +112,8 @@ async def list_mcp_servers(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
     """返回已注册的全部 MCP 服务器配置；env 中的敏感字段脱敏。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    servers = await global_state_machine.list_mcp_servers.remote()
+    global_state_machine = get_actor("global_state_machine")
+    servers = await global_state_machine.list_mcp_servers()
     for s in servers:
         if "env" in s and isinstance(s["env"], dict):
             s["env"] = _mask_config(s["env"])
@@ -126,8 +126,8 @@ async def delete_mcp_server(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.SUPER_ADMINISTRATOR)),
 ):
     """从状态机中移除一个 MCP 服务器配置。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    ok = await global_state_machine.delete_mcp_server.remote(server_id)
+    global_state_machine = get_actor("global_state_machine")
+    ok = await global_state_machine.delete_mcp_server(server_id)
     if not ok:
         raise HTTPException(status_code=404, detail="MCP server not found")
     return {"message": "success"}
@@ -152,8 +152,8 @@ async def download_artifact(
     if not artifact_id.isalnum() or len(artifact_id) > 32:
         raise HTTPException(status_code=400, detail="invalid artifact id")
 
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    wf = await postgres_database.get_workflow.remote(trace_id)
+    postgres_database = get_actor("postgres_database")
+    wf = await postgres_database.get_workflow(trace_id)
     if not wf:
         raise HTTPException(status_code=404, detail="Workflow not found")
     if getattr(wf, "user_id", None) != token_data.user_id:
@@ -187,8 +187,8 @@ async def list_toolset_packages(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
     """列出所有磁盘上的工具包（``data/toolset/<name>/`` 单元）。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    packages = await global_state_machine.list_toolset_packages.remote()
+    global_state_machine = get_actor("global_state_machine")
+    packages = await global_state_machine.list_toolset_packages()
     return {"packages": packages}
 
 
@@ -198,8 +198,8 @@ async def get_toolset_package_readme(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
     """返回指定工具包的 README.md 内容（markdown 文本）。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    content = await global_state_machine.get_toolset_package_readme.remote(name)
+    global_state_machine = get_actor("global_state_machine")
+    content = await global_state_machine.get_toolset_package_readme(name)
     if content is None:
         raise HTTPException(status_code=404, detail="README not found")
     return {"name": name, "content": content}
@@ -216,9 +216,9 @@ async def get_tools(
     其中 ``mcp_servers`` 会现场尝试连接每个已注册的 MCP 服务器并列出它们暴露的
     工具名，便于前端展示；任意一台 MCP server 不可达不影响其他工具的返回。
     """
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    tool_mapper = await global_state_machine.get_tool_mapper.remote()
-    categories = await global_state_machine.get_tool_categories.remote()
+    global_state_machine = get_actor("global_state_machine")
+    tool_mapper = await global_state_machine.get_tool_mapper()
+    categories = await global_state_machine.get_tool_categories()
 
     all_tool_names = set()
     for scope_tools in tool_mapper.values():
@@ -266,8 +266,8 @@ async def list_tool_configs(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.SUPER_ADMINISTRATOR)),
 ):
     """列出所有工具运行期配置；敏感字段会被脱敏。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    raw = await global_state_machine.list_tool_configs.remote()
+    global_state_machine = get_actor("global_state_machine")
+    raw = await global_state_machine.list_tool_configs()
     return {
         "configs": {name: _mask_config(cfg) for name, cfg in raw.items()},
     }
@@ -279,8 +279,8 @@ async def get_tool_config(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.SUPER_ADMINISTRATOR)),
 ):
     """按工具名取出脱敏后的配置。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    raw = await global_state_machine.get_tool_config.remote(tool_name)
+    global_state_machine = get_actor("global_state_machine")
+    raw = await global_state_machine.get_tool_config(tool_name)
     return {"tool_name": tool_name, "config": _mask_config(raw)}
 
 
@@ -291,8 +291,8 @@ async def set_tool_config(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.SUPER_ADMINISTRATOR)),
 ):
     """写入/覆盖某工具的运行期配置（如 ``tavily_search`` 的 ``api_key``）。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    await global_state_machine.set_tool_config.remote(tool_name, body.config)
+    global_state_machine = get_actor("global_state_machine")
+    await global_state_machine.set_tool_config(tool_name, body.config)
     return {"message": "success"}
 
 
@@ -302,8 +302,8 @@ async def delete_tool_config(
     _: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.SUPER_ADMINISTRATOR)),
 ):
     """删除某工具的运行期配置。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    ok = await global_state_machine.delete_tool_config.remote(tool_name)
+    global_state_machine = get_actor("global_state_machine")
+    ok = await global_state_machine.delete_tool_config(tool_name)
     if not ok:
         raise HTTPException(status_code=404, detail="Tool config not found")
     return {"message": "success"}
@@ -343,12 +343,12 @@ async def create_custom_toolset(
     body: CustomToolsetCreate,
     token_data: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
+    global_state_machine = get_actor("global_state_machine")
     import uuid
 
     toolset_id = str(uuid.uuid4())[:8]
     try:
-        saved = await global_state_machine.add_custom_toolset.remote(
+        saved = await global_state_machine.add_custom_toolset(
             toolset_id=toolset_id,
             name=body.name,
             tools=body.tools,
@@ -368,8 +368,8 @@ async def list_custom_toolsets(
     """列出工具组：支持按 category 过滤。USER 只能看到自己的+系统的；ADMIN 看全部。"""
     from kilostar.utils.access import get_authority
 
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    toolsets = await global_state_machine.list_custom_toolsets.remote()
+    global_state_machine = get_actor("global_state_machine")
+    toolsets = await global_state_machine.list_custom_toolsets()
     authority = await get_authority(token_data.user_id)
     if authority < UserAuthority.ADMINISTRATOR:
         toolsets = [
@@ -386,8 +386,8 @@ async def get_custom_toolset(
     toolset_id: str,
     token_data: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    ts = await global_state_machine.get_custom_toolset.remote(toolset_id)
+    global_state_machine = get_actor("global_state_machine")
+    ts = await global_state_machine.get_custom_toolset(toolset_id)
     if not ts:
         raise HTTPException(status_code=404, detail="Custom toolset not found")
     await _assert_toolset_owner_or_admin(ts, token_data)
@@ -400,8 +400,8 @@ async def update_custom_toolset(
     body: CustomToolsetUpdate,
     token_data: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    existing = await global_state_machine.get_custom_toolset.remote(toolset_id)
+    global_state_machine = get_actor("global_state_machine")
+    existing = await global_state_machine.get_custom_toolset(toolset_id)
     if not existing:
         raise HTTPException(status_code=404, detail="Custom toolset not found")
     if existing.get("is_system"):
@@ -411,7 +411,7 @@ async def update_custom_toolset(
     tools = body.tools if body.tools is not None else existing["tools"]
     description = body.description if body.description is not None else existing.get("description")
     try:
-        saved = await global_state_machine.add_custom_toolset.remote(
+        saved = await global_state_machine.add_custom_toolset(
             toolset_id=toolset_id,
             name=name,
             tools=tools,
@@ -429,14 +429,14 @@ async def delete_custom_toolset(
     token_data: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER)),
 ):
     """删除工具组：系统预置不可删；USER 只能删自己的；ADMIN 及以上可删任意用户的。"""
-    global_state_machine = ray_actor_hook("global_state_machine").global_state_machine
-    existing = await global_state_machine.get_custom_toolset.remote(toolset_id)
+    global_state_machine = get_actor("global_state_machine")
+    existing = await global_state_machine.get_custom_toolset(toolset_id)
     if not existing:
         raise HTTPException(status_code=404, detail="Custom toolset not found")
     if existing.get("is_system"):
         raise HTTPException(status_code=403, detail="系统预置工具集不可删除")
     await _assert_toolset_owner_or_admin(existing, token_data)
-    ok = await global_state_machine.delete_custom_toolset.remote(toolset_id)
+    ok = await global_state_machine.delete_custom_toolset(toolset_id)
     if not ok:
         raise HTTPException(status_code=404, detail="Custom toolset not found")
     return {"message": "success"}

kilostar/api/system.py

@@ -24,7 +24,7 @@ from __future__ import annotations
 from fastapi import APIRouter, Depends
 from fastapi.responses import JSONResponse
 
-from kilostar.utils.ray_hook import ray_actor_hook
+from kilostar.utils.actor import get_actor
 from kilostar.utils.access import Accessor, TokenData, RoleChecker
 from kilostar.core.postgres_database.model import UserAuthority
 from kilostar.utils.config_loader import (
@@ -49,15 +49,15 @@ async def readiness():
     checks = {"postgres": False, "global_state_machine": False}
 
     try:
-        postgres_database = ray_actor_hook("postgres_database").postgres_database
-        await postgres_database.ping.remote()
+        postgres_database = get_actor("postgres_database")
+        await postgres_database.ping()
         checks["postgres"] = True
     except Exception:
         pass
 
     try:
-        gsm = ray_actor_hook("global_state_machine").global_state_machine
-        await gsm.get_skill_list.remote()
+        gsm = get_actor("global_state_machine")
+        await gsm.get_skill_list()
         checks["global_state_machine"] = True
     except Exception:
         pass
@@ -95,8 +95,8 @@ async def query_system_logs(
     offset: int = 0,
     _: TokenData = Depends(Accessor.get_current_user),
 ):
-    pg = ray_actor_hook("postgres_database").postgres_database
-    logs = await pg.query_event_logs.remote(
+    pg = get_actor("postgres_database")
+    logs = await pg.query_event_logs(
         trace_id=trace_id,
         event_type=event_type,
         level=level,

kilostar/api/task.py

@@ -0,0 +1,55 @@
+#  Copyright 2026 zhaoxi826
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#       http://www.apache.org/licenses/LICENSE-2.0
+
+"""Task API：管控节点短任务的查询接口。
+
+task 由 regulatory_node 在完成短任务时内部建立，因此本路由只暴露读取，
+不开放对外创建。
+"""
+
+from typing import Optional
+
+from fastapi import APIRouter, Depends, HTTPException
+
+from kilostar.utils.access import Accessor, TokenData
+from kilostar.utils.actor import get_actor
+
+task_router = APIRouter(prefix="/api/v1/task", tags=["task"])
+
+
+@task_router.get("/list")
+async def list_tasks(
+    status: Optional[str] = None,
+    limit: int = 20,
+    offset: int = 0,
+    token_data: TokenData = Depends(Accessor.get_current_user),
+):
+    """列出当前用户的所有短任务，按时间倒序。"""
+    postgres_database = get_actor("postgres_database")
+    tasks = await postgres_database.list_tasks_by_user(
+        user_id=token_data.user_id,
+        status=status,
+        limit=limit,
+        offset=offset,
+    )
+    return {"tasks": tasks, "total": len(tasks)}
+
+
+@task_router.get("/{task_id}")
+async def get_task(
+    task_id: str,
+    token_data: TokenData = Depends(Accessor.get_current_user),
+):
+    """按 task_id 读取一条 task 详情。仅 owner 可访问。"""
+    postgres_database = get_actor("postgres_database")
+    task = await postgres_database.get_task(task_id)
+    if not task:
+        raise HTTPException(status_code=404, detail="task not found")
+    if task.get("user_id") != token_data.user_id:
+        raise HTTPException(status_code=403, detail="forbidden")
+    return task

kilostar/api/workflow.py

@@ -12,7 +12,7 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 
-from kilostar.utils.ray_hook import ray_actor_hook
+from kilostar.utils.actor import get_actor, spawn_background
 from fastapi import APIRouter, Request, HTTPException, Depends
 from fastapi.responses import StreamingResponse
 from pydantic import BaseModel
@@ -34,22 +34,23 @@ async def create_workflow(
     request: CreateWorkflowRequest,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
+    postgres_database = get_actor("postgres_database")
     trace_id = str(ULID())
-    await postgres_database.create_workflow.remote(
+    await postgres_database.create_workflow(
         trace_id=trace_id,
         user_id=token_data.user_id,
         title=request.title,
         command=request.command,
     )
 
-    global_workflow_manager = ray_actor_hook(
-        "global_workflow_manager"
-    ).global_workflow_manager
-    await global_workflow_manager.create_trace.remote(trace_id)
+    global_workflow_manager = get_actor("global_workflow_manager")
+    await global_workflow_manager.create_trace(trace_id)
 
-    consciousness_node = ray_actor_hook("consciousness_node").consciousness_node
-    consciousness_node.start_workflow_design.remote(trace_id, request.command)
+    consciousness_node = get_actor("consciousness_node")
+    # fire-and-forget：设计过程可能跑很久，不能阻塞 HTTP 响应
+    spawn_background(
+        consciousness_node.start_workflow_design(trace_id, request.command)
+    )
 
     return {"trace_id": trace_id, "status": "creating"}
 
@@ -57,8 +58,8 @@ async def create_workflow(
 @workflow_router.get("/list")
 async def get_workflow_list(
         token_data: TokenData = Depends(RoleChecker(allowed_roles=UserAuthority.USER))):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    workflows = await postgres_database.list_workflows.remote(
+    postgres_database = get_actor("postgres_database")
+    workflows = await postgres_database.list_workflows(
         user_id=token_data.user_id
     )
     return {"workflows": workflows}
@@ -75,23 +76,21 @@ async def get_workflow_sse(
     鉴权走标准 ``Authorization: Bearer`` 头（前端用 fetch-based SSE，
     token 不进 URL）。校验该 trace_id 属于当前用户。
     """
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    wf = await postgres_database.get_workflow.remote(trace_id)
+    postgres_database = get_actor("postgres_database")
+    wf = await postgres_database.get_workflow(trace_id)
     if not wf:
         raise HTTPException(status_code=404, detail="Workflow not found")
     if getattr(wf, "user_id", None) != token_data.user_id:
         raise HTTPException(status_code=403, detail="Forbidden")
 
-    global_workflow_manager = ray_actor_hook(
-        "global_workflow_manager"
-    ).global_workflow_manager
+    global_workflow_manager = get_actor("global_workflow_manager")
 
     async def event_generator():
         try:
             while True:
                 if await request.is_disconnected():
                     break
-                message = await global_workflow_manager.get_pending.remote(trace_id)
+                message = await global_workflow_manager.get_pending(trace_id)
                 if message:
                     yield f"data: {message}\n\n"
                 else:
@@ -108,8 +107,8 @@ async def post_workflow_reply(
     request: Request,
     token_data: TokenData = Depends(Accessor.get_current_user),
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    wf = await postgres_database.get_workflow.remote(trace_id)
+    postgres_database = get_actor("postgres_database")
+    wf = await postgres_database.get_workflow(trace_id)
     if not wf:
         raise HTTPException(status_code=404, detail="Workflow not found")
     if getattr(wf, "user_id", None) != token_data.user_id:
@@ -117,10 +116,8 @@ async def post_workflow_reply(
 
     data = await request.json()
     reply_msg = data.get("message", "")
-    global_workflow_manager = ray_actor_hook(
-        "global_workflow_manager"
-    ).global_workflow_manager
-    await global_workflow_manager.put_received.remote(trace_id, reply_msg)
+    global_workflow_manager = get_actor("global_workflow_manager")
+    await global_workflow_manager.put_received(trace_id, reply_msg)
     return {"status": "ok"}
 
 
@@ -128,14 +125,14 @@ async def post_workflow_reply(
 async def get_workflow_detail(
     trace_id: str, token_data: TokenData = Depends(Accessor.get_current_user)
 ):
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    wf = await postgres_database.get_workflow.remote(trace_id)
+    postgres_database = get_actor("postgres_database")
+    wf = await postgres_database.get_workflow(trace_id)
     if not wf:
         raise HTTPException(status_code=404, detail="Workflow not found")
     if getattr(wf, "user_id", None) != token_data.user_id:
         raise HTTPException(status_code=403, detail="Forbidden")
 
-    context = await postgres_database.get_workflow_context.remote(trace_id)
+    context = await postgres_database.get_workflow_context(trace_id)
 
     work_link = (
         context.work_link if context and hasattr(context, "work_link") else []
@@ -209,32 +206,30 @@ async def resume_workflow(
 ):
     """从 ``workflow_graph_state`` 持久化恢复一个被中断/挂起的工作流。
 
-    新 fire 一个 ray task，task 入口的 ``hydrate`` 检查会自动走 resume 路径
+    新 fire 一个后台任务，入口的 ``hydrate`` 检查会自动走 resume 路径
     把剩余节点跑完。
     """
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    wf = await postgres_database.get_workflow.remote(trace_id)
+    postgres_database = get_actor("postgres_database")
+    wf = await postgres_database.get_workflow(trace_id)
     if not wf:
         raise HTTPException(status_code=404, detail="Workflow not found")
     if getattr(wf, "user_id", None) != token_data.user_id:
         raise HTTPException(status_code=403, detail="Forbidden")
 
-    record = await postgres_database.get_workflow_graph_state.remote(trace_id)
+    record = await postgres_database.get_workflow_graph_state(trace_id)
     if record is None:
         raise HTTPException(
             status_code=409, detail="该工作流没有可恢复的图持久化记录"
         )
 
-    global_workflow_manager = ray_actor_hook(
-        "global_workflow_manager"
-    ).global_workflow_manager
-    await global_workflow_manager.create_trace.remote(trace_id)
+    global_workflow_manager = get_actor("global_workflow_manager")
+    await global_workflow_manager.create_trace(trace_id)
 
     from kilostar.core.work.workflow.workflow_engine import run_workflow_task
 
     # resume_only=True：task 入口 hydrate 失败会 fail-fast，绝不 fall through
     # 到"全新模式空跑"。workflow_data 在 resume 路径上不会被使用，传空 dict 占位。
-    run_workflow_task.remote({}, trace_id, resume_only=True)
+    spawn_background(run_workflow_task({}, trace_id, resume_only=True))
     return {"trace_id": trace_id, "status": "resuming"}
 
 
@@ -251,15 +246,15 @@ async def get_workflow_graph_mermaid(
     """
     from kilostar.core.work.workflow.workflow_engine import workflow_graph
 
-    postgres_database = ray_actor_hook("postgres_database").postgres_database
-    wf = await postgres_database.get_workflow.remote(trace_id)
+    postgres_database = get_actor("postgres_database")
+    wf = await postgres_database.get_workflow(trace_id)
     if not wf:
         raise HTTPException(status_code=404, detail="Workflow not found")
     if getattr(wf, "user_id", None) != token_data.user_id:
         raise HTTPException(status_code=403, detail="Forbidden")
 
     visited: list[str] = []
-    record = await postgres_database.get_workflow_graph_state.remote(trace_id)
+    record = await postgres_database.get_workflow_graph_state(trace_id)
     if record is not None:
         history = getattr(record, "history", None) or []
         # history 里每条 NodeSnapshot.id 形如 "ClassName:hash"，截前缀作为 NodeIdent

kilostar/core/global_state_machine/global_state_machine.py

@@ -13,10 +13,6 @@
 #  limitations under the License.
 
 from typing import Any, Dict, List, Optional, Tuple
-from kilostar.utils.ray_compat import actor_class, _STANDALONE
-
-if not _STANDALONE:
-    import ray
 
 from kilostar.core.global_state_machine.individual_manager import (
     GlobalIndividualManager,
@@ -28,7 +24,6 @@ from kilostar.core.global_state_machine.gsm_snapshot import GSMSnapshot
 from kilostar.core.postgres_database import PostgresDatabase
 
 
-@actor_class
 class GlobalStateMachine:
     """全局状态机 Actor，统一持有 Provider/Tool/Skill/Individual/MCP/CustomToolset 注册表。
 
@@ -62,13 +57,13 @@ class GlobalStateMachine:
             self.postgres_database
         )
         # MCP servers
-        rows = await self.postgres_database.list_mcp_servers_db.remote()
+        rows = await self.postgres_database.list_mcp_servers_db()
         self._mcp_servers = {row["server_id"]: row for row in rows}
         # Tool configs
-        cfg_rows = await self.postgres_database.list_tool_configs_db.remote()
+        cfg_rows = await self.postgres_database.list_tool_configs_db()
         self._tool_configs = {row["tool_name"]: row["config"] for row in cfg_rows}
         # Custom toolsets（含系统预置）
-        ts_rows = await self.postgres_database.list_custom_toolsets.remote()
+        ts_rows = await self.postgres_database.list_custom_toolsets()
         self._custom_toolsets = {row["toolset_id"]: row for row in ts_rows}
         # 补种系统预置工具集（首次启动或 DB 被 create_all 重建后）
         await self._seed_system_toolsets()
@@ -96,12 +91,12 @@ class GlobalStateMachine:
                 continue
             if tid in wanted_ids:
                 continue
-            await self.postgres_database.delete_custom_toolset.remote(tid)
+            await self.postgres_database.delete_custom_toolset(tid)
             self._custom_toolsets.pop(tid, None)
 
         for name, pkg in packages.items():
             tid = f"system::{name}"
-            saved = await self.postgres_database.upsert_custom_toolset.remote(
+            saved = await self.postgres_database.upsert_custom_toolset(
                 toolset_id=tid,
                 name=pkg.get("display_name") or name,
                 tools=list(pkg.get("tools", [])),
@@ -155,18 +150,10 @@ class GlobalStateMachine:
     def _publish_snapshot(self) -> None:
         """版本号 +1 并发布当前状态快照。"""
         self._config_version += 1
-        snapshot = self._build_snapshot()
-        if _STANDALONE:
-            self._current_ref = snapshot
-        else:
-            self._current_ref = ray.put(snapshot)
+        self._current_ref = self._build_snapshot()
 
     async def current_config_ref(self) -> Tuple[int, Any]:
-        """返回 ``(version, ObjectRef 或 snapshot)``。
-
-        分布式模式返回 ObjectRef，调用方用 ``ray.get`` 自取；
-        单机模式直接返回 snapshot 对象。
-        """
+        """返回 ``(version, snapshot)``。进程内直接持有 snapshot 对象。"""
         if self._current_ref is None:
             self._publish_snapshot()
         return self._config_version, self._current_ref
@@ -184,6 +171,7 @@ class GlobalStateMachine:
         provider_url,
         provider_apikey,
         provider_owner,
+        model_settings=None,
     ):
         """新增一个模型 Provider：内存注册 + 数据库持久化一并完成。"""
         result = await self._global_provider_manager.add_provider(
@@ -193,15 +181,16 @@ class GlobalStateMachine:
             provider_apikey=provider_apikey,
             provider_owner=provider_owner,
             postgres_database=self.postgres_database,
+            model_settings=model_settings or {},
         )
         self._publish_snapshot()
         return result
 
-    def get_provider_list(self):
+    async def get_provider_list(self):
         """返回内存中已登记的全部 Provider。"""
         return self._global_provider_manager.get_provider_list()
 
-    def get_provider(self, provider_title):
+    async def get_provider(self, provider_title):
         """按 provider_title 取出单个 Provider 实例。"""
         return self._global_provider_manager.get_provider(provider_title)
 
@@ -215,17 +204,17 @@ class GlobalStateMachine:
 
     # ─── Tool / Toolset ────────────────────────────────────────
 
-    def get_tool_mapper(self):
+    async def get_tool_mapper(self):
         """返回 agent_name -> {tool_name: callable} 的全量映射。"""
         return self._global_tool_manager.tool_mapper
 
-    def get_tool_list(self, agent_name: str):
+    async def get_tool_list(self, agent_name: str):
         """返回某个 agent 可用的工具集（其专属工具与 default 工具的并集）。"""
         tools = self._global_tool_manager.tool_mapper.get(agent_name, {})
         default_tools = self._global_tool_manager.tool_mapper.get("default", {})
         return {**default_tools, **tools}
 
-    def get_tool_categories(self):
+    async def get_tool_categories(self):
         """返回工具按分类聚合的完整信息。"""
         return {
             "system": self._global_tool_manager.get_system_tools(),
@@ -239,22 +228,22 @@ class GlobalStateMachine:
             "all": self._global_tool_manager.get_all_tools(),
         }
 
-    def get_toolsets_for_scope(self, scope: str) -> List[Any]:
+    async def get_toolsets_for_scope(self, scope: str) -> List[Any]:
         """返回某个 scope 下的"系统 + 自定义工具组"toolset 列表（不含 MCP）。"""
         return self._global_tool_manager.get_toolsets_for_scope(scope)
 
-    def get_retrieval_toolsets_for_scope(self, scope: str) -> List[Any]:
+    async def get_retrieval_toolsets_for_scope(self, scope: str) -> List[Any]:
         """仅返回 retrieval 工具集（system_node 专用，不包含 generation 工具）。"""
         return self._global_tool_manager.get_retrieval_toolsets_for_scope(scope)
 
-    def list_toolset_packages(self) -> List[Dict[str, Any]]:
+    async def list_toolset_packages(self) -> List[Dict[str, Any]]:
         """列出所有磁盘工具包（前端"工具插件"页面卡片即由此渲染）。"""
         return [
             {k: v for k, v in pkg.items() if k != "readme_path"}
             for pkg in self._global_tool_manager.toolset_packages.values()
         ]
 
-    def get_toolset_package_readme(self, name: str) -> Optional[str]:
+    async def get_toolset_package_readme(self, name: str) -> Optional[str]:
         """读取指定工具包的 README.md 内容；不存在返回 None。"""
         pkg = self._global_tool_manager.toolset_packages.get(name)
         if not pkg or not pkg.get("readme_path"):
@@ -269,26 +258,26 @@ class GlobalStateMachine:
 
     async def add_mcp_server(self, server_id: str, config: Dict[str, Any]) -> bool:
         """注册一个 MCP 服务器配置（写库 → 写内存）。"""
-        saved = await self.postgres_database.upsert_mcp_server.remote(server_id, config)
+        saved = await self.postgres_database.upsert_mcp_server(server_id, config)
         self._mcp_servers[server_id] = saved
         self._publish_snapshot()
         return True
 
-    def get_mcp_server(self, server_id: str) -> Optional[Dict[str, Any]]:
+    async def get_mcp_server(self, server_id: str) -> Optional[Dict[str, Any]]:
         return self._mcp_servers.get(server_id)
 
-    def list_mcp_servers(self) -> List[Dict[str, Any]]:
+    async def list_mcp_servers(self) -> List[Dict[str, Any]]:
         return [
             {"server_id": sid, **cfg} for sid, cfg in self._mcp_servers.items()
         ]
 
     async def delete_mcp_server(self, server_id: str) -> bool:
-        ok = await self.postgres_database.delete_mcp_server_db.remote(server_id)
+        ok = await self.postgres_database.delete_mcp_server_db(server_id)
         self._mcp_servers.pop(server_id, None)
         self._publish_snapshot()
         return bool(ok)
 
-    def get_mcp_server_configs(self) -> Dict[str, Dict[str, Any]]:
+    async def get_mcp_server_configs(self) -> Dict[str, Dict[str, Any]]:
         """返回原始 MCP 服务器配置字典（供节点创建 toolsets 时使用）。"""
         return dict(self._mcp_servers)
 
@@ -296,23 +285,23 @@ class GlobalStateMachine:
 
     async def set_tool_config(self, tool_name: str, config: Dict[str, Any]) -> bool:
         """整体覆盖某工具的运行期配置（敏感字段在 DAO 内自动加密）。"""
-        saved = await self.postgres_database.upsert_tool_config.remote(
+        saved = await self.postgres_database.upsert_tool_config(
             tool_name, config
         )
         self._tool_configs[tool_name] = saved["config"]
         self._publish_snapshot()
         return True
 
-    def get_tool_config(self, tool_name: str) -> Dict[str, Any]:
+    async def get_tool_config(self, tool_name: str) -> Dict[str, Any]:
         """按工具名取出配置；不存在则返回空字典。"""
         return dict(self._tool_configs.get(tool_name, {}))
 
-    def list_tool_configs(self) -> Dict[str, Dict[str, Any]]:
+    async def list_tool_configs(self) -> Dict[str, Dict[str, Any]]:
         """返回全部已配置工具的配置（包含敏感字段，调用方需自行脱敏）。"""
         return dict(self._tool_configs)
 
     async def delete_tool_config(self, tool_name: str) -> bool:
-        ok = await self.postgres_database.delete_tool_config_db.remote(tool_name)
+        ok = await self.postgres_database.delete_tool_config_db(tool_name)
         self._tool_configs.pop(tool_name, None)
         self._publish_snapshot()
         return bool(ok)
@@ -343,7 +332,7 @@ class GlobalStateMachine:
                 raise ValueError(
                     f"用户工具集只允许包含第三方工具，以下不合法：{invalid}"
                 )
-        saved = await self.postgres_database.upsert_custom_toolset.remote(
+        saved = await self.postgres_database.upsert_custom_toolset(
             toolset_id=toolset_id,
             name=name,
             tools=list(tools),
@@ -357,14 +346,14 @@ class GlobalStateMachine:
         self._publish_snapshot()
         return saved
 
-    def list_custom_toolsets(self) -> List[Dict[str, Any]]:
+    async def list_custom_toolsets(self) -> List[Dict[str, Any]]:
         return list(self._custom_toolsets.values())
 
-    def get_custom_toolset(self, toolset_id: str) -> Optional[Dict[str, Any]]:
+    async def get_custom_toolset(self, toolset_id: str) -> Optional[Dict[str, Any]]:
         return self._custom_toolsets.get(toolset_id)
 
     async def delete_custom_toolset(self, toolset_id: str) -> bool:
-        ok = await self.postgres_database.delete_custom_toolset.remote(toolset_id)
+        ok = await self.postgres_database.delete_custom_toolset(toolset_id)
         self._custom_toolsets.pop(toolset_id, None)
         self._global_tool_manager.rebuild_custom_toolsets(self._custom_toolsets)
         self._publish_snapshot()
@@ -372,33 +361,33 @@ class GlobalStateMachine:
 
     # ─── Skill ────────────────────────────────────────────────
 
-    def add_skill(self, skill_name: str):
+    async def add_skill(self, skill_name: str):
         result = self._global_skill_manager.add_skill(skill_name)
         self._publish_snapshot()
         return result
 
-    def get_skill_list(self):
+    async def get_skill_list(self):
         return self._global_skill_manager.get_skill_list()
 
-    def remove_skill(self, skill_name: str):
+    async def remove_skill(self, skill_name: str):
         result = self._global_skill_manager.remove_skill(skill_name)
         self._publish_snapshot()
         return result
 
     # ─── Individual ───────────────────────────────────────────
 
-    def add_individual(self, agent_id: str, config):
+    async def add_individual(self, agent_id: str, config):
         result = self._global_individual_manager.add_individual(agent_id, config)
         self._publish_snapshot()
         return result
 
-    def get_individual(self, agent_id: str):
+    async def get_individual(self, agent_id: str):
         return self._global_individual_manager.get_individual(agent_id)
 
-    def remove_individual(self, agent_id: str):
+    async def remove_individual(self, agent_id: str):
         result = self._global_individual_manager.remove_individual(agent_id)
         self._publish_snapshot()
         return result
 
-    def list_individuals(self):
+    async def list_individuals(self):
         return self._global_individual_manager.list_individuals()

kilostar/core/global_state_machine/gsm_snapshot.py

@@ -14,14 +14,14 @@
 
 """GSM 快照对象与客户端拉取工具。
 
-设计动机：把 GSM actor 内存里的"读路径配置"打包成可放进 Ray Object Store
-的不可变快照。读端不再走 actor RPC，而是 ``fetch_snapshot()`` 一次拿到全量
-当前配置，亚毫秒级共享内存读，绕开单 actor 的吞吐瓶颈。
+设计动机：把 GSM 内存里的"读路径配置"打包成一个不可变快照对象。读端不逐字段
+走 GSM 方法，而是 ``fetch_snapshot()`` 一次拿到全量当前配置，进程内直接引用，
+配合版本号做进程内缓存失效。
 
 GSM 仍然是 source of truth + 写入串行化器，但读路径解耦：
 
-- 写入 → GSM 内存更新 → ``ray.put(snapshot)`` 拿到新 ObjectRef → 版本号 +1
-- 读取 → ``current_config_ref()`` 拿 (version, ref) → ``ray.get(ref)`` 直读
+- 写入 → GSM 内存更新 → 重建 snapshot 对象 → 版本号 +1
+- 读取 → ``current_version()`` 校验缓存新鲜度 → ``current_config_ref()`` 拿快照
 
 旧的 ``get_provider / get_individual / ...`` 接口保留不动，是低频路径的兜底；
 新代码（特别是 skill task 这种高并发热路径）应优先走 ``fetch_snapshot``。
@@ -29,15 +29,9 @@ GSM 仍然是 source of truth + 写入串行化器，但读路径解耦：
 
 from __future__ import annotations
 
-import asyncio
 from dataclasses import dataclass, field
 from typing import Any, Callable, Dict, List, Optional, Tuple
 
-from kilostar.utils.ray_compat import _STANDALONE
-
-if not _STANDALONE:
-    import ray
-
 from kilostar.core.global_state_machine.model_provider.base_provider import Provider
 from kilostar.utils.logger import get_logger
 
@@ -73,7 +67,6 @@ class GSMSnapshot:
 
 
 _local_cache: Dict[str, Any] = {"version": -1, "snapshot": None}
-_cache_lock = asyncio.Lock()
 
 
 async def fetch_snapshot(
@@ -83,55 +76,35 @@ async def fetch_snapshot(
 ) -> GSMSnapshot:
     """拉取当前 GSM 快照。
 
-    优先走"版本号检查 + ObjectRef 直读"路径：
+    单进程 asyncio 模型下，写入路径原子发布快照对象，读端不需要加锁：
 
-    1. 调 ``gsm.current_version.remote()`` 看本地缓存是否还新（一次轻量 RPC）
-    2. 若本地缓存版本号一致，直接返回缓存（亚毫秒，零网络）
-    3. 否则调 ``gsm.current_config_ref.remote()`` 拿 ref，``ray.get`` 解出
-    4. 更新本地缓存
-
-    Args:
-        use_cache: 默认开启进程内 LRU 缓存（实际是单槽位，持有当前版本）；
-                   测试或诊断场景可关掉强制重拉。
-        gsm_actor: 可选传入 GSM actor handle；省略时通过 ``ray_actor_hook`` 获取。
-
-    Note:
-        本函数在 task / actor 进程内多次调用是廉价的；建议每次需要 config 时
-        现取，不要把 snapshot 长期持有跨任务边界（避免 ObjectRef 阻碍回收）。
+    1. 先看本地缓存版本号是否与 ``current_version()`` 一致；
+    2. 一致则直接返回缓存对象（少一次全量数据传递）；
+    3. 不一致则调 ``current_config_ref()`` 拿最新快照并更新缓存。
     """
     if gsm_actor is None:
-        from kilostar.utils.ray_hook import ray_actor_hook
+        from kilostar.utils.actor import get_gsm
 
-        gsm_actor = ray_actor_hook("global_state_machine").global_state_machine
+        gsm_actor = get_gsm()
 
     if use_cache:
-        async with _cache_lock:
         try:
-                latest_version = await gsm_actor.current_version.remote()
+            latest_version = await gsm_actor.current_version()
         except Exception:
             latest_version = None
 
         if (
             latest_version is not None
-                and _local_cache.get("version") == latest_version
-                and _local_cache.get("snapshot") is not None
+            and _local_cache["version"] == latest_version
+            and _local_cache["snapshot"] is not None
         ):
             return _local_cache["snapshot"]
 
-            version, ref_or_snapshot = await gsm_actor.current_config_ref.remote()
-            if _STANDALONE:
-                snapshot = ref_or_snapshot
-            else:
-                snapshot = ray.get(ref_or_snapshot)
+    version, snapshot = await gsm_actor.current_config_ref()
     _local_cache["version"] = version
     _local_cache["snapshot"] = snapshot
     return snapshot
 
-    version, ref_or_snapshot = await gsm_actor.current_config_ref.remote()
-    if _STANDALONE:
-        return ref_or_snapshot
-    return ray.get(ref_or_snapshot)
-
 
 def reset_local_cache() -> None:
     """清空本进程内的快照缓存（测试用）。"""

kilostar/core/global_state_machine/individual_manager.py

@@ -32,7 +32,7 @@ class GlobalIndividualManager:
         """
         try:
             try:
-                individuals = await postgres.get_all_worker_individual.remote()
+                individuals = await postgres.get_all_worker_individual()
                 for ind in individuals:
                     agent_id = getattr(ind, "agent_id", None)
                     if agent_id:

kilostar/core/global_state_machine/model_provider/base_provider.py

@@ -13,8 +13,8 @@
 #  limitations under the License.
 
 from abc import ABC, abstractmethod
-from pydantic import BaseModel
-from typing import List
+from pydantic import BaseModel, Field
+from typing import Any, Dict, List
 from enum import Enum
 
 
@@ -35,6 +35,7 @@ class Provider(BaseModel):
     provider_type: str
     provider_owner: str | None = None
     provider_status: ProviderStatus = ProviderStatus.UP
+    model_settings: Dict[str, Dict[str, Any]] = Field(default_factory=dict)
 
 
 class ProviderArgs(BaseModel):
@@ -44,6 +45,7 @@ class ProviderArgs(BaseModel):
     provider_url: str
     provider_apikey: str
     provider_owner: str
+    model_settings: Dict[str, Dict[str, Any]] = Field(default_factory=dict)
 
 
 class BaseProvider(ABC):

kilostar/core/global_state_machine/model_provider/claude_provider.py

@@ -78,4 +78,5 @@ class ClaudeProvider(BaseProvider):
             provider_url=provider_args.provider_url,
             provider_models=provider_models,
             provider_type="claude",
+            model_settings=provider_args.model_settings,
         )

kilostar/core/global_state_machine/model_provider/deepseek_provider.py

@@ -81,4 +81,5 @@ class DeepseekProvider(BaseProvider):
             provider_url=provider_args.provider_url,
             provider_models=provider_models,
             provider_type="deepseek",
+            model_settings=provider_args.model_settings,
         )

kilostar/core/global_state_machine/model_provider/gemini_provider.py

@@ -78,4 +78,5 @@ class GeminiProvider(BaseProvider):
             provider_url=provider_args.provider_url,
             provider_models=provider_models,
             provider_type="gemini",
+            model_settings=provider_args.model_settings,
         )

kilostar/core/global_state_machine/model_provider/openai_provider.py

@@ -81,4 +81,5 @@ class OpenAIProvider(BaseProvider):
             provider_url=provider_args.provider_url,
             provider_models=provider_models,
             provider_type="openai",
+            model_settings=provider_args.model_settings,
         )

kilostar/core/global_state_machine/provider_manager.py

@@ -46,7 +46,7 @@ class ProviderManager:
 
     async def init_provider_register(self, postgres) -> None:
         """从 Postgres 读取已存的 Provider 列表，按 provider_title 装入内存注册表。"""
-        providers = await postgres.get_provider.remote()
+        providers = await postgres.get_provider()
         for provider in providers:
             self.provider_register[provider.provider_title] = provider
 
@@ -58,6 +58,7 @@ class ProviderManager:
         provider_apikey,
         provider_owner,
         postgres_database,
+        model_settings=None,
     ) -> None:
         """新增并落库一个 Provider：
 
@@ -77,6 +78,7 @@ class ProviderManager:
             provider_url=provider_url,
             provider_apikey=provider_apikey,
             provider_owner=provider_owner,
+            model_settings=model_settings or {},
         )
         try:
             import ulid
@@ -88,7 +90,7 @@ class ProviderManager:
             provider: Provider = await provider_class.create_provider(provider_args)
             provider.provider_owner = provider_owner
             self.provider_register[provider_title] = provider
-            await postgres_database.add_provider_db.remote(
+            await postgres_database.add_provider_db(
                 provider_id=str(ulid.ULID()),
                 provider_title=provider.provider_title,
                 provider_url=provider.provider_url,
@@ -96,6 +98,7 @@ class ProviderManager:
                 provider_models=provider.provider_models,
                 provider_type=provider.provider_type,
                 provider_owner=provider.provider_owner,
+                model_settings=provider.model_settings,
             )
 
             logger.info(f"已添加适配器{provider_title}")
@@ -122,7 +125,7 @@ class ProviderManager:
     async def delete_provider(self, provider_title: str, postgres_database) -> None:
         """从内存注册表 + Postgres 中一并删除指定 Provider；不存在时静默返回。"""
         if provider_title in self.provider_register:
-            await postgres_database.delete_provider_by_title.remote(
+            await postgres_database.delete_provider_by_title(
                 provider_title=provider_title
             )
             del self.provider_register[provider_title]

kilostar/core/global_workflow_manager/global_workflow_manager.py

@@ -1,7 +1,5 @@
 import asyncio
 from typing import Dict
-from kilostar.utils.ray_compat import actor_class
-from kilostar.utils.ray_hook import ray_actor_hook
 from kilostar.utils.logger import get_logger
 
 
@@ -11,7 +9,6 @@ class TraceQueues:
         self.receive: asyncio.Queue[str] = asyncio.Queue()
 
 
-@actor_class
 class GlobalWorkflowManager:
     def __init__(self):
         self._traces: Dict[str, TraceQueues] = {}

kilostar/core/individual/consciousness_node/consciousness_node.py

@@ -14,7 +14,6 @@
 
 
 from typing import Union, overload
-from kilostar.utils.ray_compat import actor_class
 from kilostar.core.individual.consciousness_node.template import (
     ConsciousnessNodeDeps,
     ForregulatoryNode,
@@ -28,11 +27,9 @@ from pydantic_ai import Agent, RunContext
 from kilostar.core.global_state_machine.global_state_machine import GlobalStateMachine
 from kilostar.core.global_state_machine.model_provider.base_provider import Provider
 from kilostar.adapter.model_adapter.agent_factory import AgentFactory
-from kilostar.utils.ray_hook import ray_actor_hook
 from kilostar.utils.prompts import agent_prompt
 
 
-@actor_class
 class ConsciousnessNode:
     def __init__(self) -> None:
         from kilostar.utils.logger import get_logger
@@ -40,6 +37,7 @@ class ConsciousnessNode:
         self.logger = get_logger("consciousness_node")
         self.agent: None | Agent = None
         self.locale: str = "zh"
+        self._model_settings: dict = {}
 
     async def create_agent(
         self,
@@ -73,6 +71,7 @@ class ConsciousnessNode:
             tools=tools,
             toolsets=toolsets,
         )
+        self._model_settings = AgentFactory.resolve_model_settings(provider, model_id)
 
         @self.agent.system_prompt
         async def dynamic_prompt(ctx: RunContext[ConsciousnessNodeDeps]):
@@ -132,8 +131,10 @@ class ConsciousnessNode:
             f"ConsciousnessNode: 开始为 trace_id {trace_id} 设计工作流。原始命令：{command}"
         )
         # 获取可用技能 (示例)
-        postgres_database = ray_actor_hook("postgres_database").postgres_database
-        skills_entities = await postgres_database.get_all_worker_individual.remote()
+        from kilostar.utils.actor import get_postgres
+
+        postgres_database = get_postgres()
+        skills_entities = await postgres_database.get_all_worker_individual()
         available_skills = []
         if skills_entities:
             for skill in skills_entities:
@@ -149,10 +150,10 @@ class ConsciousnessNode:
             original_command=command, available_skills=available_skills
         )
 
-        global_workflow_manager = ray_actor_hook(
-            "global_workflow_manager"
-        ).global_workflow_manager
-        await global_workflow_manager.put_pending.remote(
+        from kilostar.utils.actor import get_gwm
+
+        global_workflow_manager = get_gwm()
+        await global_workflow_manager.put_pending(
             trace_id, "正在为您构建并规划工作流任务节点，请稍候..."
         )
 
@@ -162,20 +163,21 @@ class ConsciousnessNode:
             workflow = result.workflow
             workflow.trace_id = trace_id
 
-            await global_workflow_manager.put_pending.remote(
+            await global_workflow_manager.put_pending(
                 trace_id, "工作流构建完成，即将开始执行！"
             )
 
-            # 直接以 ray task 形式 fire workflow，不再经过 WorkflowRunningEngine 这层中转：
-            # workflow 是一次性、有头有尾的执行，task 语义比常驻 actor 更贴。
+            # 以后台任务形式 fire workflow：workflow 是一次性、有头有尾的
+            # 执行，跑完即结束；持久化交给 PostgresStatePersistence 保证可恢复。
+            from kilostar.utils.actor import spawn_background
             from kilostar.core.work.workflow.workflow_engine import run_workflow_task
 
-            run_workflow_task.remote(workflow.model_dump(), trace_id)
+            spawn_background(run_workflow_task(workflow.model_dump(), trace_id))
         else:
-            await global_workflow_manager.put_pending.remote(
+            await global_workflow_manager.put_pending(
                 trace_id, "很抱歉，工作流生成失败。"
             )
-            await postgres_database.update_workflow_status.remote(trace_id, "failed")
+            await postgres_database.update_workflow_status(trace_id, "failed")
 
     async def working(
         self,
@@ -221,7 +223,7 @@ class ConsciousnessNode:
                 )
                 self.logger.debug("ConsciousnessNode: 开始生成工作流 (原生重试开启)")
                 prompt = "根据original_command制定严密的可执行workflow"
-                result = await self.agent.run(prompt, deps=deps)
+                result = await self.agent.run(prompt, deps=deps, model_settings=self._model_settings or None)
                 return result.output
 
             elif isinstance(payload, ForWorkflowInput):
@@ -236,6 +238,7 @@ class ConsciousnessNode:
                 result = await self.agent.run(
                     f"处理此工作流步骤信息：\n{payload.workflow_step.model_dump_json()}",
                     deps=deps,
+                    model_settings=self._model_settings or None,
                 )
                 return result.output
 
@@ -251,6 +254,7 @@ class ConsciousnessNode:
                 result = await self.agent.run(
                     f"基于以下工作流的执行记录，生成技术报告：\n{payload.workflow.model_dump_json()}",
                     deps=deps,
+                    model_settings=self._model_settings or None,
                 )
                 return result.output
         except Exception as e:

kilostar/core/individual/consciousness_node/template.py

@@ -14,24 +14,23 @@
 
 
 from kilostar.core.work.workflow.workflow import KiloStarWorkflow, WorkflowStep
-from kilostar.utils.agent_model import ResponseModel, DepsModel, RequestModel
-from pydantic import Field
+from pydantic import BaseModel, Field
 from typing import Optional, List
 
 
 # 意识节点回复类
-class ConsciousnessNodeResponse(ResponseModel):
+class ConsciousnessNodeResponse(BaseModel):
     """Consciousness response model,是意识节点所有回复类型的父类"""
     pass
 
-class ConsciousnessNodeDeps(DepsModel):
+class ConsciousnessNodeDeps(BaseModel):
     """ConsciousnessNode 在 pydantic-ai Agent 中使用的依赖：原始指令、当前指令以及可用 Skill 列表。"""
     original_command: str
     command: str
     available_skills: Optional[List[dict]] = None
     locale: str = "zh"
 
-class ConsciousnessNodeInput(RequestModel):
+class ConsciousnessNodeInput(BaseModel):
     """ConsciousnessNode 各类入参的共同基类，仅用于打 schema 标签。"""
     pass
 

kilostar/core/individual/control_node/__init__.py

@@ -1,17 +0,0 @@
-#  Copyright 2026 zhaoxi826
-#
-#  Licensed under the Apache License, Version 2.0 (the "License");
-#  you may not use this file except in compliance with the License.
-#  You may obtain a copy of the License at
-#
-#       http://www.apache.org/licenses/LICENSE-2.0
-#
-#  Unless required by applicable law or agreed to in writing, software
-#  distributed under the License is distributed on an "AS IS" BASIS,
-#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-#  See the License for the specific language governing permissions and
-#  limitations under the License.
-
-from .control_node import ControlNode
-
-__all__ = ["ControlNode"]

kilostar/core/individual/control_node/control_node.py

@@ -1,130 +0,0 @@
-#  Copyright 2026 zhaoxi826
-#
-#  Licensed under the Apache License, Version 2.0 (the "License");
-#  you may not use this file except in compliance with the License.
-#  You may obtain a copy of the License at
-#
-#       http://www.apache.org/licenses/LICENSE-2.0
-#
-#  Unless required by applicable law or agreed to in writing, software
-#  distributed under the License is distributed on an "AS IS" BASIS,
-#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-#  See the License for the specific language governing permissions and
-#  limitations under the License.
-
-from pydantic_ai import Agent, RunContext
-from kilostar.utils.ray_compat import actor_class
-from kilostar.core.global_state_machine.global_state_machine import GlobalStateMachine
-from kilostar.core.global_state_machine.model_provider.base_provider import Provider
-from kilostar.adapter.model_adapter.agent_factory import AgentFactory
-from kilostar.core.individual.control_node.template import (
-    ForWorkflow,
-    ForWorkflowInput,
-    ControlNodeDeps,
-)
-from kilostar.utils.prompts import agent_prompt
-
-
-@actor_class
-class ControlNode:
-    """ControlNode（控制节点）：工作流中具体子任务的执行 Actor。
-
-    它把 ConsciousnessNode 编排出的 ``workflow_step`` 拿来当作输入，借助
-    pydantic-ai Agent + 已绑定的工具集合产出 ``ForWorkflow`` 结构化输出。
-    """
-
-    def __init__(self):
-        from kilostar.utils.logger import get_logger
-
-        self.logger = get_logger("control_node")
-        self.agent: Agent | None = None
-
-    async def create_agent(
-        self,
-        global_state_machine: GlobalStateMachine,
-        provider_title: str,
-        model_id: str,
-        tools_list: list[str] = None,
-        toolsets=None,
-        locale: str | None = None,
-        custom_system_prompt: str | None = None,
-    ) -> None:
-        """
-        create_agent方法，将agent对象装配到Control的属性内
-        该方法通过provider_title从global_state_machine中获取provider对象，然后从provider对象中取出供应商形象，装配为pydantic_ai的
-        Agent实例，
-        并挂载到self.agent属性
-        Args:
-            global_state_machine: 全局状态机
-            provider_title: 供应商名
-            model_id: 模型id
-            locale: 语言代码（zh/en），控制system prompt语言
-            custom_system_prompt: 管理员自定义追加提示词（可选）
-
-        Returns:
-            无返回
-        """
-        system_prompt: str = agent_prompt("control_node", locale=locale, custom_system_prompt=custom_system_prompt)
-        output_type = ForWorkflow
-        from kilostar.utils.get_tool import load_tools_from_list
-        from kilostar.core.global_state_machine.gsm_snapshot import fetch_snapshot
-
-        snapshot = await fetch_snapshot(gsm_actor=global_state_machine)
-        provider: Provider = snapshot.providers.get(provider_title)
-        if provider is None:
-            from kilostar.utils.i18n import t
-            raise ValueError(t("provider_not_registered", locale=locale, provider_title=provider_title))
-        agent_factory = AgentFactory()
-
-        callables = load_tools_from_list(tools_list)
-        self.agent = agent_factory.create_agent(
-            provider=provider,
-            model_id=model_id,
-            output_type=output_type,
-            system_prompt=system_prompt,
-            deps_type=ControlNodeDeps,
-            agent_name="control_node",
-            tools=callables,
-            toolsets=toolsets,
-        )
-
-        @self.agent.system_prompt
-        async def dynamic_prompt(ctx: RunContext[ControlNodeDeps]):
-            """运行期动态拼接 system prompt：把当前 workflow_step 的关键字段塞进去。"""
-            prompt = system_prompt + "\n\n"
-            prompt += (
-                f"=== 当前任务步骤上下文 ===\n"
-                f"- 步骤名称 (Name): {ctx.deps.workflow_step.name}\n"
-                f"- 步骤目标/描述 (Description): {ctx.deps.workflow_step.desc}\n"
-                f"- 前置输入（input）: {ctx.deps.workflow_step.inputs}\n"
-            )
-            return prompt
-
-    async def working(self, payload: ForWorkflowInput) -> str:
-        """对外入口：执行一次步骤，吞掉异常并返回 ``None`` 以避免拖垮上游 Workflow。"""
-        try:
-            result: ForWorkflow = await self._run(payload)
-            return result
-        except Exception:
-            self.logger.exception("ControlNode在执行working时发生严重错误")
-            return None
-
-    async def _run(self, payload: ForWorkflowInput) -> ForWorkflow:
-        """实际执行步骤：组装 ``ControlNodeDeps``、调用 Agent，最终把 ``ForWorkflow`` 输出取出。"""
-        try:
-            self.agent.retries = 3
-            deps = ControlNodeDeps(workflow_step=payload.workflow_step)
-            self.logger.debug(
-                f"ControlNode: 开始执行工作流节点 [{payload.workflow_step.name}] (原生重试开启)"
-            )
-
-            result = await self.agent.run(
-                f"请根据提供的 workflow_step 上下文，执行此步骤并输出结果。\n详细指令或附加数据：{payload.workflow_step.model_dump_json()}",
-                deps=deps,
-            )
-            return result.output
-        except Exception as e:
-            self.logger.exception(
-                f"ControlNode 在执行步骤 [{payload.workflow_step.name}] 时最终失败: {str(e)}"
-            )
-            raise RuntimeError(f"ControlNode 执行步骤失败: {str(e)}") from e

kilostar/core/individual/control_node/template.py

@@ -1,51 +0,0 @@
-#  Copyright 2026 zhaoxi826
-#
-#  Licensed under the Apache License, Version 2.0 (the "License");
-#  you may not use this file except in compliance with the License.
-#  You may obtain a copy of the License at
-#
-#       http://www.apache.org/licenses/LICENSE-2.0
-#
-#  Unless required by applicable law or agreed to in writing, software
-#  distributed under the License is distributed on an "AS IS" BASIS,
-#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-#  See the License for the specific language governing permissions and
-#  limitations under the License.
-
-
-from pydantic import Field
-from kilostar.core.work.workflow.workflow import WorkflowStep
-from kilostar.utils.agent_model import ResponseModel, RequestModel, DepsModel
-
-
-class ControlNodeResponse(ResponseModel):
-    """控制节点回复的基类。"""
-
-    pass
-
-
-class ControlNodeInput(RequestModel):
-    """控制节点输入的基类，承载一次调度所需的入参。"""
-
-    pass
-
-
-class ControlNodeDeps(DepsModel):
-    """控制节点运行期依赖，注入到 pydantic-ai Agent 的 RunContext。"""
-
-    workflow_step: WorkflowStep
-    # In the future, this can be dynamically populated with tools specific to the current task execution
-
-
-class ForWorkflow(ControlNodeResponse):
-    """控制节点执行单个工作流步骤的输出模型。"""
-
-    output: str = Field(
-        ..., description="控制节点执行特定工作流步骤的结果。包含执行细节和输出数据。"
-    )
-
-
-class ForWorkflowInput(ControlNodeInput):
-    """控制节点针对工作流步骤的输入模型。"""
-
-    workflow_step: WorkflowStep