chore: ignore codex local artifacts

.gitignore

@@ -51,11 +51,19 @@ node_modules
 outputs/
 /production/
 /smoke_test/
+/frontend/dist/
+/frontend/test-results/
 
 # Local tooling state
 .omc/
+/.codex/
+/.codex
 /.pydeps/
 /referance/
+/.pids/
+/.pytest_cache/
+/.ruff_cache/
+/evotraders.egg-info/
 
 # Run outputs
 /runs/

.pids/agent_service.pid

@@ -1 +0,0 @@
-73343

.pids/frontend.pid

@@ -1 +0,0 @@
-73348

.pids/gateway.pid

@@ -1 +0,0 @@
-66939

.pids/news_service.pid

@@ -1 +0,0 @@
-73345

.pids/openclaw_service.pid

@@ -1 +0,0 @@
-73347

.pids/runtime_service.pid

@@ -1 +0,0 @@
-73346

.pids/trading_service.pid

@@ -1 +0,0 @@
-73344

CLAUDE.md

@@ -115,7 +115,8 @@ npm run test         # Vitest 单元测试
 
 | 文件 | 职责 |
 |------|------|
-| `pipeline.py` | TradingPipeline - 核心编排器（分析→沟通→决策→执行→评估） |
+| `pipeline.py` | TradingPipeline - 核心编排器（分析→沟通→决策→执行→评估），支持断点 Checkpoint |
+| `apo.py` | PolicyOptimizer - (APO) 自动策略优化器，根据 P&L 自动修改 Agent POLICY.md |
 | `pipeline_runner.py` | REST API 触发的独立执行，5 阶段启动 |
 | `scheduler.py` | BacktestScheduler、Scheduler - 回测/实盘调度 |
 | `state_sync.py` | StateSync - 状态同步和广播 |
@@ -168,7 +169,8 @@ backend/
 │   └── models.py           # ProcessRun、ProcessRunState
 │
 ├── core/                  # Pipeline 执行
-│   ├── pipeline.py         # TradingPipeline（核心编排器）
+│   ├── pipeline.py         # TradingPipeline（核心编排器，支持恢复）
+│   ├── apo.py              # PolicyOptimizer（自动调优）
 │   ├── pipeline_runner.py  # 独立 Pipeline 执行
 │   ├── scheduler.py        # 调度器
 │   └── state_sync.py       # 状态同步

README.md

@@ -21,8 +21,11 @@ The repository name still uses `evotraders`, but the product-facing branding now
 **Multi-agent trading team**
 Six roles collaborate like a real desk: four specialist analysts (fundamentals, technical, sentiment, valuation), one portfolio manager, and one risk manager.
 
-**Continuous learning**
-Agents can persist long-term memory with ReMe, reflect after each cycle, and evolve their decision patterns over time.
+**Continuous learning & Evolution**
+Agents persist long-term memory with ReMe and reflect after each cycle. The **Autonomous Policy Optimizer (APO)** automatically tunes agent operational policies (`POLICY.md`) based on daily P&L feedback to correct recurring mistakes.
+
+**Robust execution with recovery**
+The trading pipeline supports **phase-based checkpointing**. If a run is interrupted, it can resume from the last successful phase (Analysis, Risk, Discussion, Decision, Execution, or Settlement), ensuring resilience in production.
 
 **Backtest and live modes**
 The same runtime model supports historical simulation and live execution with real-time market data.
@@ -68,6 +71,7 @@ Runtime state is stored in `runs/<run_id>/` — this is the **runtime source of
 
 ### Documentation
 
+- [docs/README.md](./docs/README.md) — documentation index and reading order
 - [docs/current-architecture.md](./docs/current-architecture.md) — canonical architecture facts
 - [services/README.md](./services/README.md) — service boundaries and migration details
 - [docs/current-architecture.excalidraw](./docs/current-architecture.excalidraw) — visual diagram
@@ -84,15 +88,11 @@ Runtime state is stored in `runs/<run_id>/` — this is the **runtime source of
 # clone this repository, then:
 cd evotraders
 
-# backend runtime dependencies
-uv pip install -r requirements.txt
-
 # install package entrypoint in editable mode
 uv pip install -e .
 
-# optional
+# optional dev dependencies
 # uv pip install -e ".[dev]"
-# pip install -e .
 ```
 
 Frontend dependencies:
@@ -103,7 +103,7 @@ npm ci
 cd ..
 ```
 
-Production deployment should prefer `requirements.txt` for backend and `npm ci` for frontend so the pulled environment matches the checked-in lockfiles and version pins.
+Production deployment should prefer the checked-in Python package metadata in `pyproject.toml` for backend installation and `npm ci` for frontend so the pulled environment matches the checked-in dependency declarations and lockfiles.
 
 ### 2. Configure environment
 
@@ -206,13 +206,13 @@ Recommended local development flow:
 ./start-dev.sh
 ```
 
-This starts:
+This starts directly from the script:
 
 - `agent_service` at `http://localhost:8000`
 - `trading_service` at `http://localhost:8001`
 - `news_service` at `http://localhost:8002`
 - `runtime_service` at `http://localhost:8003`
-- gateway WebSocket at `ws://localhost:8765`
+- gateway WebSocket at `ws://localhost:8765` via `runtime_service` managed startup
 
 Then start the frontend in another terminal:
 
@@ -229,31 +229,34 @@ python -m uvicorn backend.apps.agent_service:app --host 0.0.0.0 --port 8000 --re
 python -m uvicorn backend.apps.trading_service:app --host 0.0.0.0 --port 8001 --reload
 python -m uvicorn backend.apps.news_service:app --host 0.0.0.0 --port 8002 --reload
 python -m uvicorn backend.apps.runtime_service:app --host 0.0.0.0 --port 8003 --reload
-# compatibility gateway path, not the recommended primary dev entrypoint
-python -m backend.main --mode live --host 0.0.0.0 --port 8765
+
+# then create a runtime so runtime_service can spawn the Gateway subprocess
+curl -X POST http://localhost:8003/api/runtime/start \
+  -H "Content-Type: application/json" \
+  -d '{"launch_mode":"fresh","tickers":["AAPL","MSFT"],"mode":"live"}'
 ```
 ### 4. Run backtest or live mode
 
 Backtest:
 
 ```bash
-python backend/main.py --mode backtest --config-name smoke_fullstack --start-date 2025-11-01 --end-date 2025-12-01
-python backend/main.py --mode backtest --config-name smoke_fullstack --start-date 2025-11-01 --end-date 2025-12-01 --enable-memory
+curl -X POST http://localhost:8003/api/runtime/start \
+  -H "Content-Type: application/json" \
+  -d '{"launch_mode":"fresh","mode":"backtest","tickers":["AAPL","MSFT"],"start_date":"2025-11-01","end_date":"2025-12-01"}'
 ```
 
 Live:
 
 ```bash
-python backend/main.py --mode live --config-name live
-python backend/main.py --mode live --config-name live --enable-memory
-python backend/main.py --mode live --config-name live --interval-minutes 60
-python backend/main.py --mode live --config-name live --trigger-time 22:30
+curl -X POST http://localhost:8003/api/runtime/start \
+  -H "Content-Type: application/json" \
+  -d '{"launch_mode":"fresh","mode":"live","tickers":["AAPL","MSFT"]}'
 ```
 
 Help:
 
 ```bash
-python backend/main.py --help
+python backend/main.py --help  # compatibility standalone entrypoint only
 ```
 ### Offline backtest data
 
@@ -311,7 +314,7 @@ If these are not set, the frontend falls back to its local defaults and compatib
 
 ```text
 Market data -> independent analyst work -> team communication -> portfolio decision ->
-risk review -> execution/settlement -> reflection/memory update
+risk review -> execution/settlement -> reflection/memory update -> APO policy tuning
 ```
 
 The runtime manager also tracks:

README_zh.md

@@ -66,7 +66,9 @@ frontend -> runtime_service/control APIs -> gateway/runtime manager -> market se
 
 ### 文档
 
+- [docs/README.md](./docs/README.md) — 文档索引与阅读顺序
 - [docs/current-architecture.md](./docs/current-architecture.md) — 权威架构事实
+- [docs/project-layout.md](./docs/project-layout.md) — 当前目录结构与职责说明
 - [services/README.md](./services/README.md) — 服务边界和迁移详情
 - [docs/current-architecture.excalidraw](./docs/current-architecture.excalidraw) — 架构图
 - [docs/development-roadmap.md](./docs/development-roadmap.md) — 下一步执行计划
@@ -82,15 +84,11 @@ frontend -> runtime_service/control APIs -> gateway/runtime manager -> market se
 # 克隆仓库后进入项目目录
 cd evotraders
 
-# 安装后端运行时依赖
-uv pip install -r requirements.txt
-
 # 安装项目入口（可编辑模式）
 uv pip install -e .
 
-# 可选
+# 可选开发依赖
 # uv pip install -e ".[dev]"
-# pip install -e .
 ```
 
 前端依赖：
@@ -101,7 +99,7 @@ npm ci
 cd ..
 ```
 
-生产环境部署建议后端使用 `requirements.txt`，前端使用 `npm ci`，这样拉起的环境会严格跟随仓库中锁定的依赖版本。
+生产环境部署建议后端以 `pyproject.toml` 中声明的包元数据为准进行安装，前端使用 `npm ci`，这样拉起的环境会严格跟随仓库中声明的依赖和锁定版本。
 
 ### 2. 配置环境变量
 
@@ -178,7 +176,7 @@ python3 scripts/smoke_evo_runtime.py --agent-id fundamentals_analyst
 - `trading_service`：`http://localhost:8001`
 - `news_service`：`http://localhost:8002`
 - `runtime_service`：`http://localhost:8003`
-- gateway WebSocket：`ws://localhost:8765`
+- gateway WebSocket：`ws://localhost:8765`，由 `runtime_service` 托管拉起
 
 然后在另一个终端启动前端：
 
@@ -195,8 +193,11 @@ python -m uvicorn backend.apps.agent_service:app --host 0.0.0.0 --port 8000 --re
 python -m uvicorn backend.apps.trading_service:app --host 0.0.0.0 --port 8001 --reload
 python -m uvicorn backend.apps.news_service:app --host 0.0.0.0 --port 8002 --reload
 python -m uvicorn backend.apps.runtime_service:app --host 0.0.0.0 --port 8003 --reload
-# 兼容性 gateway 路径，不是推荐的主要开发入口
-python -m backend.main --mode live --host 0.0.0.0 --port 8765
+
+# 然后通过 runtime_service 创建运行时，由它拉起 Gateway 子进程
+curl -X POST http://localhost:8003/api/runtime/start \
+  -H "Content-Type: application/json" \
+  -d '{"launch_mode":"fresh","tickers":["AAPL","MSFT"],"mode":"live"}'
 ```
 
 仓库里部署脚本使用的 `production` 只是一个示例 run label，不应再把它理解成
@@ -207,24 +208,19 @@ python -m backend.main --mode live --host 0.0.0.0 --port 8765
 回测：
 
 ```bash
-python backend/main.py --mode backtest --config-name smoke_fullstack --start-date 2025-11-01 --end-date 2025-12-01
-python backend/main.py --mode backtest --config-name smoke_fullstack --start-date 2025-11-01 --end-date 2025-12-01 --enable-memory
+curl -X POST http://localhost:8003/api/runtime/start \
+  -H "Content-Type: application/json" \
+  -d '{"launch_mode":"fresh","mode":"backtest","tickers":["AAPL","MSFT"],"start_date":"2025-11-01","end_date":"2025-12-01"}'
 ```
 
 实盘：
 
 ```bash
-python backend/main.py --mode live --config-name live
-python backend/main.py --mode live --config-name live --enable-memory
-python backend/main.py --mode live --config-name live --interval-minutes 60
-python backend/main.py --mode live --config-name live --trigger-time 22:30
+curl -X POST http://localhost:8003/api/runtime/start \
+  -H "Content-Type: application/json" \
+  -d '{"launch_mode":"fresh","mode":"live","tickers":["AAPL","MSFT"]}'
 ```
 
-帮助：
-
-```bash
-python backend/main.py --help
-```
 ### 离线回测数据
 
 如果只是想快速体验回测，不依赖外部行情 API，可以下载离线数据包并解压到 `backend/data`：

backend/agents/dynamic_team_types.py

@@ -0,0 +1,372 @@
+# -*- coding: utf-8 -*-
+"""Dynamic Team Types - Core data types for PM-driven analyst team management.
+
+This module provides data structures for:
+- Analyst persona definitions (custom analyst types)
+- Analyst creation configuration (custom SOUL.md, AGENTS.md, etc.)
+- Dynamic team runtime state tracking
+
+These types enable the Portfolio Manager to dynamically create, clone, and manage
+analyst agents with custom configurations beyond the predefined 4 analyst types.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Optional, Dict, Any, List
+from datetime import datetime
+
+
+@dataclass
+class AnalystPersona:
+    """Analyst role definition - extends or replaces personas.yaml entries.
+
+    Defines the identity, focus areas, and characteristics of an analyst type.
+    Can be used to create entirely new analyst types at runtime.
+
+    Attributes:
+        name: Display name for the analyst (e.g., "期权策略分析师")
+        focus: List of focus areas (e.g., ["期权定价", "波动率交易"])
+        description: Detailed description of the analyst's role and expertise
+        preferred_tools: Optional list of preferred tool types or categories
+        icon: Optional icon identifier for frontend display
+    """
+    name: str
+    focus: List[str]
+    description: str
+    preferred_tools: Optional[List[str]] = None
+    icon: Optional[str] = None
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary for serialization."""
+        return {
+            "name": self.name,
+            "focus": self.focus,
+            "description": self.description,
+            "preferred_tools": self.preferred_tools,
+            "icon": self.icon,
+        }
+
+    @classmethod
+    def from_dict(cls, data: Dict[str, Any]) -> AnalystPersona:
+        """Create from dictionary."""
+        return cls(
+            name=data["name"],
+            focus=data.get("focus", []),
+            description=data.get("description", ""),
+            preferred_tools=data.get("preferred_tools"),
+            icon=data.get("icon"),
+        )
+
+
+@dataclass
+class AnalystConfig:
+    """Complete configuration for dynamically creating an analyst.
+
+    This dataclass allows the PM to specify all aspects of analyst creation,
+    including custom workspace files, model overrides, and skill selections.
+
+    Attributes:
+        persona: Complete persona definition (if creating custom type)
+        analyst_type: Reference to predefined type (e.g., "technical_analyst")
+        soul_md: Custom SOUL.md content (overrides default generation)
+        agents_md: Custom AGENTS.md content (overrides default generation)
+        profile_md: Custom PROFILE.md content (overrides default generation)
+        skills: List of skill IDs to enable for this analyst
+        model_name: Override default model for this analyst
+        memory_config: Custom memory system configuration
+        tags: Classification tags (e.g., ["options", "derivatives"])
+        parent_id: If cloned, the source analyst ID
+    """
+    # Identity configuration
+    persona: Optional[AnalystPersona] = None
+    analyst_type: Optional[str] = None  # Reference to predefined type
+
+    # Workspace file contents (override default generation)
+    soul_md: Optional[str] = None
+    agents_md: Optional[str] = None
+    profile_md: Optional[str] = None
+    bootstrap_md: Optional[str] = None
+
+    # Runtime configuration
+    skills: Optional[List[str]] = field(default_factory=list)
+    model_name: Optional[str] = None
+    memory_config: Optional[Dict[str, Any]] = field(default_factory=dict)
+
+    # Metadata
+    tags: Optional[List[str]] = field(default_factory=list)
+    parent_id: Optional[str] = None  # For clone tracking
+
+    def __post_init__(self):
+        """Initialize default collections."""
+        if self.skills is None:
+            self.skills = []
+        if self.memory_config is None:
+            self.memory_config = {}
+        if self.tags is None:
+            self.tags = []
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary for serialization."""
+        return {
+            "persona": self.persona.to_dict() if self.persona else None,
+            "analyst_type": self.analyst_type,
+            "soul_md": self.soul_md,
+            "agents_md": self.agents_md,
+            "profile_md": self.profile_md,
+            "bootstrap_md": self.bootstrap_md,
+            "skills": self.skills,
+            "model_name": self.model_name,
+            "memory_config": self.memory_config,
+            "tags": self.tags,
+            "parent_id": self.parent_id,
+        }
+
+    @classmethod
+    def from_dict(cls, data: Dict[str, Any]) -> AnalystConfig:
+        """Create from dictionary."""
+        persona_data = data.get("persona")
+        return cls(
+            persona=AnalystPersona.from_dict(persona_data) if persona_data else None,
+            analyst_type=data.get("analyst_type"),
+            soul_md=data.get("soul_md"),
+            agents_md=data.get("agents_md"),
+            profile_md=data.get("profile_md"),
+            bootstrap_md=data.get("bootstrap_md"),
+            skills=data.get("skills", []),
+            model_name=data.get("model_name"),
+            memory_config=data.get("memory_config", {}),
+            tags=data.get("tags", []),
+            parent_id=data.get("parent_id"),
+        )
+
+    def get_effective_analyst_type(self) -> Optional[str]:
+        """Get the effective analyst type for tool selection.
+
+        Returns analyst_type if set, otherwise derives from persona name.
+        """
+        if self.analyst_type:
+            return self.analyst_type
+        if self.persona:
+            # Derive type ID from persona name (e.g., "期权策略分析师" -> "options_strategist")
+            return self._derive_type_id(self.persona.name)
+        return None
+
+    @staticmethod
+    def _derive_type_id(name: str) -> str:
+        """Derive a type ID from a display name."""
+        import re
+        # Convert Chinese or mixed names to snake_case
+        # Remove special characters, keep alphanumeric and spaces
+        cleaned = re.sub(r'[^\w\s]', '', name)
+        # Convert to lowercase and replace spaces with underscores
+        return cleaned.lower().strip().replace(' ', '_')
+
+
+@dataclass
+class DynamicAnalystInstance:
+    """Runtime information about a dynamically created analyst.
+
+    Tracks the creation metadata and current state of a dynamic analyst.
+
+    Attributes:
+        agent_id: Unique identifier for this analyst instance
+        config: The configuration used to create this analyst
+        created_at: Timestamp when the analyst was created
+        created_by: Identifier of the agent that created this analyst (usually PM)
+        status: Current status (active, paused, removed)
+    """
+    agent_id: str
+    config: AnalystConfig
+    created_at: str = field(default_factory=lambda: datetime.now().isoformat())
+    created_by: str = "portfolio_manager"
+    status: str = "active"  # active, paused, removed
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary for serialization."""
+        return {
+            "agent_id": self.agent_id,
+            "config": self.config.to_dict(),
+            "created_at": self.created_at,
+            "created_by": self.created_by,
+            "status": self.status,
+        }
+
+    @classmethod
+    def from_dict(cls, data: Dict[str, Any]) -> DynamicAnalystInstance:
+        """Create from dictionary."""
+        return cls(
+            agent_id=data["agent_id"],
+            config=AnalystConfig.from_dict(data.get("config", {})),
+            created_at=data.get("created_at", datetime.now().isoformat()),
+            created_by=data.get("created_by", "portfolio_manager"),
+            status=data.get("status", "active"),
+        )
+
+
+@dataclass
+class DynamicTeamState:
+    """Complete runtime state for dynamic analyst team management.
+
+    This state is persisted alongside TEAM_PIPELINE.yaml and tracks:
+    - Custom analyst types registered at runtime
+    - All dynamically created analyst instances
+    - Configuration snapshots for cloning
+
+    Attributes:
+        run_id: The run configuration this state belongs to
+        registered_types: Runtime-registered analyst type definitions
+        instances: Dynamically created analyst instances
+        version: State format version for migration handling
+    """
+    run_id: str
+    registered_types: Dict[str, AnalystPersona] = field(default_factory=dict)
+    instances: Dict[str, DynamicAnalystInstance] = field(default_factory=dict)
+    version: int = 1
+
+    def register_type(self, type_id: str, persona: AnalystPersona) -> bool:
+        """Register a new analyst type.
+
+        Returns:
+            True if registered, False if type_id already exists
+        """
+        if type_id in self.registered_types:
+            return False
+        self.registered_types[type_id] = persona
+        return True
+
+    def add_instance(self, instance: DynamicAnalystInstance) -> None:
+        """Add a new analyst instance."""
+        self.instances[instance.agent_id] = instance
+
+    def remove_instance(self, agent_id: str) -> bool:
+        """Mark an instance as removed.
+
+        Returns:
+            True if instance was found and removed
+        """
+        if agent_id in self.instances:
+            self.instances[agent_id].status = "removed"
+            return True
+        return False
+
+    def get_active_instances(self) -> List[DynamicAnalystInstance]:
+        """Get all active (non-removed) analyst instances."""
+        return [
+            inst for inst in self.instances.values()
+            if inst.status == "active"
+        ]
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary for serialization."""
+        return {
+            "run_id": self.run_id,
+            "registered_types": {
+                k: v.to_dict() for k, v in self.registered_types.items()
+            },
+            "instances": {
+                k: v.to_dict() for k, v in self.instances.items()
+            },
+            "version": self.version,
+        }
+
+    @classmethod
+    def from_dict(cls, data: Dict[str, Any]) -> DynamicTeamState:
+        """Create from dictionary."""
+        registered_types = {
+            k: AnalystPersona.from_dict(v)
+            for k, v in data.get("registered_types", {}).items()
+        }
+        instances = {
+            k: DynamicAnalystInstance.from_dict(v)
+            for k, v in data.get("instances", {}).items()
+        }
+        return cls(
+            run_id=data.get("run_id", "unknown"),
+            registered_types=registered_types,
+            instances=instances,
+            version=data.get("version", 1),
+        )
+
+
+@dataclass
+class CreateAnalystResult:
+    """Result of creating a dynamic analyst.
+
+    Attributes:
+        success: Whether creation was successful
+        agent_id: The ID of the created analyst (if successful)
+        message: Human-readable result message
+        error: Error details (if failed)
+    """
+    success: bool
+    agent_id: Optional[str] = None
+    message: str = ""
+    error: Optional[str] = None
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary for API responses."""
+        return {
+            "success": self.success,
+            "agent_id": self.agent_id,
+            "message": self.message,
+            "error": self.error,
+        }
+
+
+@dataclass
+class CloneAnalystRequest:
+    """Request to clone an existing analyst.
+
+    Attributes:
+        source_id: ID of the analyst to clone
+        new_id: ID for the new analyst
+        config_overrides: Configuration fields to override
+    """
+    source_id: str
+    new_id: str
+    config_overrides: Optional[Dict[str, Any]] = field(default_factory=dict)
+
+    def __post_init__(self):
+        if self.config_overrides is None:
+            self.config_overrides = {}
+
+
+@dataclass
+class AnalystTypeInfo:
+    """Information about an available analyst type.
+
+    Used for listing all available types (predefined + runtime-registered).
+
+    Attributes:
+        type_id: Unique identifier for this type
+        name: Display name
+        description: Type description
+        is_builtin: Whether this is a built-in type or runtime-registered
+        source: Source of this type (e.g., "constants", "runtime", "config")
+    """
+    type_id: str
+    name: str
+    description: str
+    is_builtin: bool
+    source: str
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary for API responses."""
+        return {
+            "type_id": self.type_id,
+            "name": self.name,
+            "description": self.description,
+            "is_builtin": self.is_builtin,
+            "source": self.source,
+        }
+
+
+__all__ = [
+    "AnalystPersona",
+    "AnalystConfig",
+    "DynamicAnalystInstance",
+    "DynamicTeamState",
+    "CreateAnalystResult",
+    "CloneAnalystRequest",
+    "AnalystTypeInfo",
+]

backend/agents/toolkit_factory.py

@@ -14,6 +14,14 @@ from backend.agents.agent_workspace import load_agent_workspace_config
 from backend.agents.skills_manager import SkillsManager
 from backend.agents.skill_metadata import parse_skill_metadata
 from backend.config.bootstrap_config import get_bootstrap_config_for_run
+from backend.tools.dynamic_team_tools import (
+    create_analyst,
+    clone_analyst,
+    remove_analyst,
+    list_analyst_types,
+    get_analyst_info,
+    get_team_summary,
+)
 
 
 def load_agent_profiles() -> Dict[str, Dict[str, Any]]:
@@ -138,6 +146,23 @@ def _register_portfolio_tool_groups(toolkit: Any, pm_agent: Any) -> None:
             group_name="portfolio_ops",
         )
 
+    # Register dynamic team management tools
+    toolkit.create_tool_group(
+        group_name="dynamic_team",
+        description="Dynamic analyst team management tools.",
+        active=False,
+        notes=(
+            "Use these tools to create, clone, and manage analyst agents dynamically. "
+            "Only available when allow_dynamic_team_update is enabled."
+        ),
+    )
+    toolkit.register_tool_function(create_analyst, group_name="dynamic_team")
+    toolkit.register_tool_function(clone_analyst, group_name="dynamic_team")
+    toolkit.register_tool_function(remove_analyst, group_name="dynamic_team")
+    toolkit.register_tool_function(list_analyst_types, group_name="dynamic_team")
+    toolkit.register_tool_function(get_analyst_info, group_name="dynamic_team")
+    toolkit.register_tool_function(get_team_summary, group_name="dynamic_team")
+
 
 def _register_risk_tool_groups(toolkit: Any) -> None:
     """注册风险工具组"""

backend/api/__init__.py

@@ -13,6 +13,7 @@ from .workspaces import router as workspaces_router
 from .guard import router as guard_router
 from .runtime import router as runtime_router
 from .runs import router as runs_router
+from .dynamic_team import router as dynamic_team_router
 
 __all__ = [
     "agents_router",
@@ -20,4 +21,5 @@ __all__ = [
     "guard_router",
     "runtime_router",
     "runs_router",
+    "dynamic_team_router",
 ]

backend/api/agents.py

@@ -1,46 +1,18 @@
 # -*- coding: utf-8 -*-
-"""
-Agent API Routes
-
-Provides REST API endpoints for both:
-
-- design-time agent management under `workspaces/`
-- run-scoped agent asset access under `runs/<run_id>/`
-"""
+"""Agent API routes for design-time workspace registry CRUD only."""
 import logging
-import os
-import tempfile
 from pathlib import Path
 from typing import Any, Dict, List, Optional
 
-from fastapi import APIRouter, HTTPException, Depends, Body, UploadFile, File, Form
+from fastapi import APIRouter, HTTPException, Depends
 from pydantic import BaseModel, Field
 
 from backend.agents import AgentFactory, get_registry
-from backend.agents.workspace_manager import RunWorkspaceManager
-from backend.agents.agent_workspace import load_agent_workspace_config
-from backend.agents.skills_manager import SkillsManager
-from backend.agents.toolkit_factory import load_agent_profiles
-from backend.config.bootstrap_config import get_bootstrap_config_for_run
-from backend.llm.models import get_agent_model_info
 
 logger = logging.getLogger(__name__)
 
 router = APIRouter(prefix="/api/workspaces/{workspace_id}/agents", tags=["agents"])
 DESIGN_SCOPE = "design_workspace"
-RUNTIME_SCOPE = "runtime_run"
-RUNTIME_SCOPE_NOTE = (
-    "For profile, skills, and editable agent files, `workspace_id` is treated "
-    "as the active run id under `runs/<run_id>/`, not as the design-time "
-    "`workspaces/` registry."
-)
-
-
-def _runtime_scope_fields() -> dict[str, str]:
-    return {
-        "scope_type": RUNTIME_SCOPE,
-        "scope_note": RUNTIME_SCOPE_NOTE,
-    }
 
 
 def _design_scope_fields() -> dict[str, str]:
@@ -65,26 +37,9 @@ class CreateAgentRequest(BaseModel):
 
 
 class UpdateAgentRequest(BaseModel):
-    """Request to update an agent."""
+    """Request to update design-time agent metadata."""
     name: Optional[str] = None
     description: Optional[str] = None
-    enabled_skills: Optional[List[str]] = None
-    disabled_skills: Optional[List[str]] = None
-
-
-class InstallExternalSkillRequest(BaseModel):
-    """Request to install an external skill for one agent."""
-    source: str = Field(..., description="Directory path, zip path, or http(s) zip URL")
-    name: Optional[str] = Field(None, description="Optional override skill name")
-    activate: bool = Field(True, description="Whether to enable skill immediately")
-
-
-class LocalSkillRequest(BaseModel):
-    skill_name: str = Field(..., description="Local skill name")
-
-
-class LocalSkillContentRequest(BaseModel):
-    content: str = Field(..., description="Updated SKILL.md content")
 
 
 class AgentResponse(BaseModel):
@@ -99,54 +54,12 @@ class AgentResponse(BaseModel):
     scope_note: Optional[str] = None
 
 
-class AgentFileResponse(BaseModel):
-    """Agent file content response."""
-    filename: str
-    content: str
-    scope_type: str = RUNTIME_SCOPE
-    scope_note: Optional[str] = None
-
-
-class AgentProfileResponse(BaseModel):
-    agent_id: str
-    workspace_id: str
-    profile: Dict[str, Any]
-    scope_type: str = RUNTIME_SCOPE
-    scope_note: Optional[str] = None
-
-
-class AgentSkillsResponse(BaseModel):
-    agent_id: str
-    workspace_id: str
-    skills: List[Dict[str, Any]]
-    scope_type: str = RUNTIME_SCOPE
-    scope_note: Optional[str] = None
-
-
-class SkillDetailResponse(BaseModel):
-    agent_id: str
-    workspace_id: str
-    skill: Dict[str, Any]
-    scope_type: str = RUNTIME_SCOPE
-    scope_note: Optional[str] = None
-
-
 # Dependencies
 def get_agent_factory():
     """Get AgentFactory instance."""
     return AgentFactory()
 
 
-def get_workspace_manager():
-    """Get run-scoped asset manager for one runtime workspace/run id."""
-    return RunWorkspaceManager()
-
-
-def get_skills_manager():
-    """Get SkillsManager instance."""
-    return SkillsManager()
-
-
 # Routes
 @router.post("", response_model=AgentResponse)
 async def create_agent(
@@ -270,119 +183,6 @@ async def get_agent(
     )
 
 
-@router.get("/{agent_id}/profile", response_model=AgentProfileResponse)
-async def get_agent_profile(
-    workspace_id: str,
-    agent_id: str,
-    skills_manager: SkillsManager = Depends(get_skills_manager),
-):
-    asset_dir = skills_manager.get_agent_asset_dir(workspace_id, agent_id)
-    agent_config = load_agent_workspace_config(asset_dir / "agent.yaml")
-    profiles = load_agent_profiles()
-    profile = profiles.get(agent_id, {})
-    bootstrap = get_bootstrap_config_for_run(skills_manager.project_root, workspace_id)
-    override = bootstrap.agent_override(agent_id)
-    active_tool_groups = override.get("active_tool_groups", agent_config.active_tool_groups or profile.get("active_tool_groups", []))
-    if not isinstance(active_tool_groups, list):
-        active_tool_groups = []
-    disabled_tool_groups = agent_config.disabled_tool_groups
-    if disabled_tool_groups:
-        disabled_set = set(disabled_tool_groups)
-        active_tool_groups = [group_name for group_name in active_tool_groups if group_name not in disabled_set]
-
-    default_skills = profile.get("skills", [])
-    if not isinstance(default_skills, list):
-        default_skills = []
-    resolved_skills = skills_manager.resolve_agent_skill_names(
-        config_name=workspace_id,
-        agent_id=agent_id,
-        default_skills=default_skills,
-    )
-    prompt_files = agent_config.prompt_files or ["SOUL.md", "PROFILE.md", "AGENTS.md", "POLICY.md", "MEMORY.md"]
-    model_name, model_provider = get_agent_model_info(agent_id)
-
-    return AgentProfileResponse(
-        agent_id=agent_id,
-        workspace_id=workspace_id,
-        profile={
-            "model_name": model_name,
-            "model_provider": model_provider,
-            "prompt_files": prompt_files,
-            "default_skills": default_skills,
-            "resolved_skills": resolved_skills,
-            "active_tool_groups": active_tool_groups,
-            "disabled_tool_groups": disabled_tool_groups,
-            "enabled_skills": agent_config.enabled_skills,
-            "disabled_skills": agent_config.disabled_skills,
-        },
-        **_runtime_scope_fields(),
-    )
-
-
-@router.get("/{agent_id}/skills", response_model=AgentSkillsResponse)
-async def get_agent_skills(
-    workspace_id: str,
-    agent_id: str,
-    skills_manager: SkillsManager = Depends(get_skills_manager),
-):
-    agent_asset_dir = skills_manager.get_agent_asset_dir(workspace_id, agent_id)
-    agent_config = load_agent_workspace_config(agent_asset_dir / "agent.yaml")
-    resolved_skills = set(skills_manager.resolve_agent_skill_names(config_name=workspace_id, agent_id=agent_id, default_skills=[]))
-    enabled = set(agent_config.enabled_skills)
-    disabled = set(agent_config.disabled_skills)
-
-    payload = []
-    for item in skills_manager.list_agent_skill_catalog(workspace_id, agent_id):
-        if item.skill_name in disabled:
-            status = "disabled"
-        elif item.skill_name in enabled:
-            status = "enabled"
-        elif item.skill_name in resolved_skills:
-            status = "active"
-        else:
-            status = "available"
-        payload.append({
-            "skill_name": item.skill_name,
-            "name": item.name,
-            "description": item.description,
-            "version": item.version,
-            "source": item.source,
-            "tools": item.tools,
-            "status": status,
-        })
-
-    return AgentSkillsResponse(
-        agent_id=agent_id,
-        workspace_id=workspace_id,
-        skills=payload,
-        **_runtime_scope_fields(),
-    )
-
-
-@router.get("/{agent_id}/skills/{skill_name}", response_model=SkillDetailResponse)
-async def get_agent_skill_detail(
-    workspace_id: str,
-    agent_id: str,
-    skill_name: str,
-    skills_manager: SkillsManager = Depends(get_skills_manager),
-):
-    try:
-        detail = skills_manager.load_agent_skill_document(
-            config_name=workspace_id,
-            agent_id=agent_id,
-            skill_name=skill_name,
-        )
-    except FileNotFoundError:
-        raise HTTPException(status_code=404, detail=f"Unknown skill: {skill_name}")
-
-    return SkillDetailResponse(
-        agent_id=agent_id,
-        workspace_id=workspace_id,
-        skill=detail,
-        **_runtime_scope_fields(),
-    )
-
-
 @router.delete("/{agent_id}")
 async def delete_agent(
     workspace_id: str,
@@ -448,16 +248,6 @@ async def update_agent(
     if metadata_updates:
         registry.update_metadata(agent_id, metadata_updates)
 
-    # Update skills if provided
-    if request.enabled_skills or request.disabled_skills:
-        skills_manager = SkillsManager()
-        skills_manager.update_agent_skill_overrides(
-            config_name=workspace_id,
-            agent_id=agent_id,
-            enable=request.enabled_skills or [],
-            disable=request.disabled_skills or [],
-        )
-
     # Get updated info
     agent_info = registry.get(agent_id)
     return AgentResponse(
@@ -469,301 +259,3 @@ async def update_agent(
         status=agent_info.status,
         **_design_scope_fields(),
     )
-
-
-@router.post("/{agent_id}/skills/{skill_name}/enable")
-async def enable_skill(
-    workspace_id: str,
-    agent_id: str,
-    skill_name: str,
-    registry = Depends(get_registry),
-):
-    """
-    Enable a skill for an agent.
-
-    Args:
-        workspace_id: Workspace identifier
-        agent_id: Agent identifier
-        skill_name: Skill name to enable
-
-    Returns:
-        Success message
-    """
-    agent_info = registry.get(agent_id)
-    if not agent_info or agent_info.workspace_id != workspace_id:
-        raise HTTPException(status_code=404, detail=f"Agent '{agent_id}' not found")
-
-    skills_manager = SkillsManager()
-    result = skills_manager.update_agent_skill_overrides(
-        config_name=workspace_id,
-        agent_id=agent_id,
-        enable=[skill_name],
-    )
-
-    return {
-        "message": f"Skill '{skill_name}' enabled for agent '{agent_id}'",
-        "enabled_skills": result["enabled_skills"],
-    }
-
-
-@router.post("/{agent_id}/skills/{skill_name}/disable")
-async def disable_skill(
-    workspace_id: str,
-    agent_id: str,
-    skill_name: str,
-    registry = Depends(get_registry),
-):
-    """
-    Disable a skill for an agent.
-
-    Args:
-        workspace_id: Workspace identifier
-        agent_id: Agent identifier
-        skill_name: Skill name to disable
-
-    Returns:
-        Success message
-    """
-    agent_info = registry.get(agent_id)
-    if not agent_info or agent_info.workspace_id != workspace_id:
-        raise HTTPException(status_code=404, detail=f"Agent '{agent_id}' not found")
-
-    skills_manager = SkillsManager()
-    result = skills_manager.update_agent_skill_overrides(
-        config_name=workspace_id,
-        agent_id=agent_id,
-        disable=[skill_name],
-    )
-
-    return {
-        "message": f"Skill '{skill_name}' disabled for agent '{agent_id}'",
-        "disabled_skills": result["disabled_skills"],
-    }
-
-
-@router.post("/{agent_id}/skills/install")
-async def install_external_skill(
-    workspace_id: str,
-    agent_id: str,
-    request: InstallExternalSkillRequest,
-    registry=Depends(get_registry),
-):
-    """Install an external skill into one agent's local skills."""
-    agent_info = registry.get(agent_id)
-    if not agent_info or agent_info.workspace_id != workspace_id:
-        raise HTTPException(status_code=404, detail=f"Agent '{agent_id}' not found")
-
-    skills_manager = SkillsManager()
-    try:
-        result = skills_manager.install_external_skill_for_agent(
-            config_name=workspace_id,
-            agent_id=agent_id,
-            source=request.source,
-            skill_name=request.name,
-            activate=request.activate,
-        )
-    except (FileNotFoundError, ValueError) as exc:
-        raise HTTPException(status_code=400, detail=str(exc))
-
-    return {
-        "message": f"Installed external skill '{result['skill_name']}' for '{agent_id}'",
-        **result,
-    }
-
-
-@router.post("/{agent_id}/skills/local")
-async def create_local_skill(
-    workspace_id: str,
-    agent_id: str,
-    request: LocalSkillRequest,
-    registry=Depends(get_registry),
-):
-    agent_info = registry.get(agent_id)
-    if not agent_info or agent_info.workspace_id != workspace_id:
-        raise HTTPException(status_code=404, detail=f"Agent '{agent_id}' not found")
-
-    skills_manager = SkillsManager()
-    try:
-        skills_manager.create_agent_local_skill(
-            config_name=workspace_id,
-            agent_id=agent_id,
-            skill_name=request.skill_name,
-        )
-    except (ValueError, FileExistsError) as exc:
-        raise HTTPException(status_code=400, detail=str(exc))
-
-    return {"message": f"Created local skill '{request.skill_name}' for '{agent_id}'"}
-
-
-@router.put("/{agent_id}/skills/local/{skill_name}")
-async def update_local_skill(
-    workspace_id: str,
-    agent_id: str,
-    skill_name: str,
-    request: LocalSkillContentRequest,
-    registry=Depends(get_registry),
-):
-    agent_info = registry.get(agent_id)
-    if not agent_info or agent_info.workspace_id != workspace_id:
-        raise HTTPException(status_code=404, detail=f"Agent '{agent_id}' not found")
-
-    skills_manager = SkillsManager()
-    try:
-        skills_manager.update_agent_local_skill(
-            config_name=workspace_id,
-            agent_id=agent_id,
-            skill_name=skill_name,
-            content=request.content,
-        )
-    except (ValueError, FileNotFoundError) as exc:
-        raise HTTPException(status_code=400, detail=str(exc))
-
-    return {"message": f"Updated local skill '{skill_name}' for '{agent_id}'"}
-
-
-@router.delete("/{agent_id}/skills/local/{skill_name}")
-async def delete_local_skill(
-    workspace_id: str,
-    agent_id: str,
-    skill_name: str,
-    registry=Depends(get_registry),
-):
-    agent_info = registry.get(agent_id)
-    if not agent_info or agent_info.workspace_id != workspace_id:
-        raise HTTPException(status_code=404, detail=f"Agent '{agent_id}' not found")
-
-    skills_manager = SkillsManager()
-    try:
-        skills_manager.delete_agent_local_skill(
-            config_name=workspace_id,
-            agent_id=agent_id,
-            skill_name=skill_name,
-        )
-        skills_manager.forget_agent_skill_overrides(
-            config_name=workspace_id,
-            agent_id=agent_id,
-            skill_names=[skill_name],
-        )
-    except (ValueError, FileNotFoundError) as exc:
-        raise HTTPException(status_code=400, detail=str(exc))
-
-    return {"message": f"Deleted local skill '{skill_name}' for '{agent_id}'"}
-
-
-@router.post("/{agent_id}/skills/upload")
-async def upload_external_skill(
-    workspace_id: str,
-    agent_id: str,
-    file: UploadFile = File(...),
-    name: Optional[str] = Form(None),
-    activate: bool = Form(True),
-    registry=Depends(get_registry),
-):
-    """Upload a zip skill package from frontend and install for one agent."""
-    agent_info = registry.get(agent_id)
-    if not agent_info or agent_info.workspace_id != workspace_id:
-        raise HTTPException(status_code=404, detail=f"Agent '{agent_id}' not found")
-
-    original_name = (file.filename or "").strip()
-    if not original_name.lower().endswith(".zip"):
-        raise HTTPException(status_code=400, detail="Uploaded file must be a .zip archive")
-
-    suffix = Path(original_name).suffix or ".zip"
-    temp_path: Optional[str] = None
-    try:
-        with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
-            temp_path = tmp.name
-            content = await file.read()
-            tmp.write(content)
-
-        skills_manager = SkillsManager()
-        result = skills_manager.install_external_skill_for_agent(
-            config_name=workspace_id,
-            agent_id=agent_id,
-            source=temp_path,
-            skill_name=name,
-            activate=activate,
-        )
-    except (FileNotFoundError, ValueError) as exc:
-        raise HTTPException(status_code=400, detail=str(exc))
-    finally:
-        try:
-            await file.close()
-        except Exception as e:
-            logger.warning(f"Failed to close uploaded file: {e}")
-        if temp_path and os.path.exists(temp_path):
-            os.remove(temp_path)
-
-    return {
-        "message": f"Uploaded and installed external skill '{result['skill_name']}' for '{agent_id}'",
-        **result,
-    }
-
-
-@router.get("/{agent_id}/files/{filename}", response_model=AgentFileResponse)
-async def get_agent_file(
-    workspace_id: str,
-    agent_id: str,
-    filename: str,
-    workspace_manager: RunWorkspaceManager = Depends(get_workspace_manager),
-):
-    """
-    Read an agent file from the run-scoped asset tree under `runs/<run_id>/`.
-
-    Args:
-        workspace_id: Workspace identifier
-        agent_id: Agent identifier
-        filename: File to read (e.g., SOUL.md, PROFILE.md)
-
-    Returns:
-        File content
-    """
-    try:
-        content = workspace_manager.load_agent_file(
-            config_name=workspace_id,
-            agent_id=agent_id,
-            filename=filename,
-        )
-        return AgentFileResponse(
-            filename=filename,
-            content=content,
-            **_runtime_scope_fields(),
-        )
-    except FileNotFoundError:
-        raise HTTPException(status_code=404, detail=f"File '{filename}' not found")
-
-
-@router.put("/{agent_id}/files/{filename}", response_model=AgentFileResponse)
-async def update_agent_file(
-    workspace_id: str,
-    agent_id: str,
-    filename: str,
-    content: str = Body(..., media_type="text/plain"),
-    workspace_manager: RunWorkspaceManager = Depends(get_workspace_manager),
-):
-    """
-    Update an agent file in the run-scoped asset tree under `runs/<run_id>/`.
-
-    Args:
-        workspace_id: Workspace identifier
-        agent_id: Agent identifier
-        filename: File to update
-        content: New file content
-
-    Returns:
-        Updated file information
-    """
-    try:
-        workspace_manager.update_agent_file(
-            config_name=workspace_id,
-            agent_id=agent_id,
-            filename=filename,
-            content=content,
-        )
-        return AgentFileResponse(
-            filename=filename,
-            content=content,
-            **_runtime_scope_fields(),
-        )
-    except Exception as e:
-        raise HTTPException(status_code=500, detail=str(e))

backend/api/dynamic_team.py

@@ -0,0 +1,404 @@
+# -*- coding: utf-8 -*-
+"""Dynamic Team API - REST endpoints for managing analyst team dynamically.
+
+This module provides API endpoints for:
+- Creating new analysts with custom configuration
+- Cloning existing analysts
+- Removing analysts
+- Listing available analyst types
+- Getting analyst information
+- Managing team composition
+
+These endpoints allow both the PM agent (via tool calls) and frontend
+(via HTTP) to manage the analyst team dynamically.
+"""
+from __future__ import annotations
+
+import logging
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+from fastapi import APIRouter, HTTPException
+from pydantic import BaseModel, Field
+
+from backend.agents.dynamic_team_types import (
+    AnalystPersona,
+    AnalystConfig,
+    AnalystTypeInfo,
+)
+from backend.config.constants import ANALYST_TYPES
+from backend.agents.prompt_loader import get_prompt_loader
+
+logger = logging.getLogger(__name__)
+
+router = APIRouter(prefix="/api/dynamic-team", tags=["dynamic-team"])
+
+PROJECT_ROOT = Path(__file__).resolve().parents[2]
+
+
+# Pydantic models for API requests/responses
+
+class AnalystPersonaRequest(BaseModel):
+    """Request model for analyst persona definition."""
+    name: str = Field(..., description="Display name for the analyst")
+    focus: List[str] = Field(default_factory=list, description="List of focus areas")
+    description: str = Field(..., description="Detailed description")
+    preferred_tools: Optional[List[str]] = Field(None, description="Preferred tool categories")
+    icon: Optional[str] = Field(None, description="Icon identifier")
+
+
+class CreateAnalystRequest(BaseModel):
+    """Request model for creating a new analyst."""
+    agent_id: str = Field(..., description="Unique identifier for the new analyst")
+    analyst_type: str = Field(..., description="Base type or custom identifier")
+    persona: Optional[AnalystPersonaRequest] = Field(None, description="Custom persona definition")
+    soul_md: Optional[str] = Field(None, description="Custom SOUL.md content")
+    agents_md: Optional[str] = Field(None, description="Custom AGENTS.md content")
+    profile_md: Optional[str] = Field(None, description="Custom PROFILE.md content")
+    bootstrap_md: Optional[str] = Field(None, description="Custom BOOTSTRAP.md content")
+    model_name: Optional[str] = Field(None, description="Override default LLM model")
+    skills: Optional[List[str]] = Field(None, description="List of skill IDs to enable")
+    tags: Optional[List[str]] = Field(None, description="Classification tags")
+
+
+class CloneAnalystRequest(BaseModel):
+    """Request model for cloning an analyst."""
+    source_id: str = Field(..., description="ID of the analyst to clone")
+    new_id: str = Field(..., description="Unique identifier for the new analyst")
+    name: Optional[str] = Field(None, description="New display name")
+    focus_additions: Optional[List[str]] = Field(None, description="Additional focus areas")
+    description_override: Optional[str] = Field(None, description="New description")
+    model_name: Optional[str] = Field(None, description="Override model from source")
+
+
+class RegisterTypeRequest(BaseModel):
+    """Request model for registering a new analyst type."""
+    type_id: str = Field(..., description="Unique identifier for this type")
+    name: str = Field(..., description="Display name")
+    focus: List[str] = Field(..., description="List of focus areas")
+    description: str = Field(..., description="Detailed description")
+    preferred_tools: Optional[List[str]] = Field(None, description="Preferred tool categories")
+
+
+class AnalystResponse(BaseModel):
+    """Response model for analyst operations."""
+    success: bool
+    agent_id: Optional[str] = None
+    message: str
+    error: Optional[str] = None
+
+
+class AnalystTypeResponse(BaseModel):
+    """Response model for analyst type information."""
+    type_id: str
+    name: str
+    description: str
+    is_builtin: bool
+    source: str
+
+
+class AnalystInfoResponse(BaseModel):
+    """Response model for detailed analyst information."""
+    found: bool
+    agent_id: str
+    config: Optional[Dict[str, Any]] = None
+    is_custom: bool = False
+    is_clone: bool = False
+    parent_id: Optional[str] = None
+    message: Optional[str] = None
+
+
+class TeamSummaryResponse(BaseModel):
+    """Response model for team summary."""
+    total_analysts: int
+    custom_analysts: int
+    cloned_analysts: int
+    analysts: List[Dict[str, Any]]
+    registered_types: int
+
+
+# Helper function to get the current pipeline instance
+
+def _get_pipeline(run_id: str) -> Optional[Any]:
+    """Get the TradingPipeline instance for a run.
+
+    Args:
+        run_id: The run configuration ID
+
+    Returns:
+        TradingPipeline instance or None if not found
+    """
+    # Import here to avoid circular imports
+    try:
+        from backend.apps.runtime_service import get_runtime_state
+        runtime_state = get_runtime_state()
+        if runtime_state and hasattr(runtime_state, 'pipeline'):
+            return runtime_state.pipeline
+    except Exception as e:
+        logger.warning(f"Could not get pipeline for run {run_id}: {e}")
+    return None
+
+
+def _get_controller(run_id: str) -> Optional[Any]:
+    """Get the DynamicTeamController for a run.
+
+    Args:
+        run_id: The run configuration ID
+
+    Returns:
+        DynamicTeamController instance or None if not available
+    """
+    try:
+        from backend.tools.dynamic_team_tools import get_controller
+        return get_controller()
+    except Exception as e:
+        logger.warning(f"Could not get controller for run {run_id}: {e}")
+    return None
+
+
+# API Endpoints
+
+@router.get("/types", response_model=List[AnalystTypeResponse])
+async def list_analyst_types() -> List[AnalystTypeResponse]:
+    """List all available analyst types.
+
+    Returns both built-in types (from ANALYST_TYPES) and runtime-registered types.
+    """
+    result = []
+
+    # Add built-in types
+    for type_id, info in ANALYST_TYPES.items():
+        result.append(AnalystTypeResponse(
+            type_id=type_id,
+            name=info.get("display_name", type_id),
+            description=info.get("description", ""),
+            is_builtin=True,
+            source="constants",
+        ))
+
+    # Try to get runtime registered types
+    controller = _get_controller("default")
+    if controller:
+        for type_id, persona in controller._registered_types.items():
+            result.append(AnalystTypeResponse(
+                type_id=type_id,
+                name=persona.name,
+                description=persona.description,
+                is_builtin=False,
+                source="runtime",
+            ))
+
+    return result
+
+
+@router.get("/personas")
+async def get_personas() -> Dict[str, Any]:
+    """Get all analyst personas from personas.yaml.
+
+    Returns the persona definitions used for analyst initialization.
+    """
+    try:
+        personas = get_prompt_loader().load_yaml_config("analyst", "personas")
+        return {"success": True, "personas": personas}
+    except Exception as e:
+        logger.error(f"Failed to load personas: {e}")
+        raise HTTPException(status_code=500, detail=f"Failed to load personas: {e}")
+
+
+@router.post("/runs/{run_id}/analysts", response_model=AnalystResponse)
+async def create_analyst(
+    run_id: str,
+    request: CreateAnalystRequest,
+) -> AnalystResponse:
+    """Create a new analyst in the specified run.
+
+    Args:
+        run_id: The run configuration ID
+        request: Analyst creation configuration
+
+    Returns:
+        Result of the creation operation
+    """
+    controller = _get_controller(run_id)
+    if not controller:
+        raise HTTPException(
+            status_code=503,
+            detail="Dynamic team controller not available. Is the pipeline running?"
+        )
+
+    # Build persona if provided
+    persona = None
+    if request.persona:
+        persona = AnalystPersona(
+            name=request.persona.name,
+            focus=request.persona.focus,
+            description=request.persona.description,
+            preferred_tools=request.persona.preferred_tools,
+            icon=request.persona.icon,
+        )
+
+    # Build config
+    config = AnalystConfig(
+        persona=persona,
+        analyst_type=request.analyst_type if request.analyst_type in ANALYST_TYPES else None,
+        soul_md=request.soul_md,
+        agents_md=request.agents_md,
+        profile_md=request.profile_md,
+        bootstrap_md=request.bootstrap_md,
+        model_name=request.model_name,
+        skills=request.skills or [],
+        tags=request.tags or [],
+    )
+
+    # Create the analyst
+    result = controller.create_analyst(
+        agent_id=request.agent_id,
+        analyst_type=request.analyst_type,
+        name=persona.name if persona else None,
+        focus=persona.focus if persona else None,
+        description=persona.description if persona else None,
+        soul_md=config.soul_md,
+        agents_md=config.agents_md,
+        model_name=config.model_name,
+    )
+
+    return AnalystResponse(**result)
+
+
+@router.post("/runs/{run_id}/analysts/clone", response_model=AnalystResponse)
+async def clone_analyst(
+    run_id: str,
+    request: CloneAnalystRequest,
+) -> AnalystResponse:
+    """Clone an existing analyst.
+
+    Args:
+        run_id: The run configuration ID
+        request: Clone configuration
+
+    Returns:
+        Result of the clone operation
+    """
+    controller = _get_controller(run_id)
+    if not controller:
+        raise HTTPException(
+            status_code=503,
+            detail="Dynamic team controller not available. Is the pipeline running?"
+        )
+
+    result = controller.clone_analyst(
+        source_id=request.source_id,
+        new_id=request.new_id,
+        name=request.name,
+        focus_additions=request.focus_additions,
+        description_override=request.description_override,
+        model_name=request.model_name,
+    )
+
+    return AnalystResponse(**result)
+
+
+@router.delete("/runs/{run_id}/analysts/{agent_id}", response_model=AnalystResponse)
+async def remove_analyst(run_id: str, agent_id: str) -> AnalystResponse:
+    """Remove a dynamically created analyst.
+
+    Args:
+        run_id: The run configuration ID
+        agent_id: The analyst to remove
+
+    Returns:
+        Result of the removal operation
+    """
+    controller = _get_controller(run_id)
+    if not controller:
+        raise HTTPException(
+            status_code=503,
+            detail="Dynamic team controller not available. Is the pipeline running?"
+        )
+
+    result = controller.remove_analyst(agent_id)
+    return AnalystResponse(**result)
+
+
+@router.get("/runs/{run_id}/analysts/{agent_id}", response_model=AnalystInfoResponse)
+async def get_analyst_info(run_id: str, agent_id: str) -> AnalystInfoResponse:
+    """Get information about a specific analyst.
+
+    Args:
+        run_id: The run configuration ID
+        agent_id: The analyst ID
+
+    Returns:
+        Analyst configuration and status
+    """
+    controller = _get_controller(run_id)
+    if not controller:
+        raise HTTPException(
+            status_code=503,
+            detail="Dynamic team controller not available. Is the pipeline running?"
+        )
+
+    result = controller.get_analyst_info(agent_id)
+    return AnalystInfoResponse(**result)
+
+
+@router.get("/runs/{run_id}/summary", response_model=TeamSummaryResponse)
+async def get_team_summary(run_id: str) -> TeamSummaryResponse:
+    """Get a summary of the current analyst team.
+
+    Args:
+        run_id: The run configuration ID
+
+    Returns:
+        Team composition information
+    """
+    controller = _get_controller(run_id)
+    if not controller:
+        raise HTTPException(
+            status_code=503,
+            detail="Dynamic team controller not available. Is the pipeline running?"
+        )
+
+    result = controller.get_team_summary()
+    return TeamSummaryResponse(**result)
+
+
+@router.post("/runs/{run_id}/types", response_model=AnalystTypeResponse)
+async def register_analyst_type(
+    run_id: str,
+    request: RegisterTypeRequest,
+) -> AnalystTypeResponse:
+    """Register a new analyst type.
+
+    Args:
+        run_id: The run configuration ID
+        request: Type registration configuration
+
+    Returns:
+        Registered type information
+    """
+    controller = _get_controller(run_id)
+    if not controller:
+        raise HTTPException(
+            status_code=503,
+            detail="Dynamic team controller not available. Is the pipeline running?"
+        )
+
+    result = controller.register_analyst_type(
+        type_id=request.type_id,
+        name=request.name,
+        focus=request.focus,
+        description=request.description,
+        preferred_tools=request.preferred_tools,
+    )
+
+    if not result.get("success", False):
+        raise HTTPException(status_code=400, detail=result.get("message", "Registration failed"))
+
+    return AnalystTypeResponse(
+        type_id=request.type_id,
+        name=request.name,
+        description=request.description,
+        is_builtin=False,
+        source="runtime",
+    )

backend/api/runs.py

@@ -6,7 +6,7 @@ Provides REST API endpoints for runtime agent asset access under `runs/<run_id>/
 
 This module separates runtime concerns from design-time workspace management:
 - `/api/runs/{run_id}/agents/*` - Runtime agent assets and configuration
-- `/api/workspaces/{workspace_id}/agents/*` - Design-time workspace registry (deprecated)
+- design-time workspace registry CRUD lives under `/api/workspaces/{workspace_id}/...`
 """
 import logging
 import os

backend/api/runtime.py

@@ -29,6 +29,17 @@ router = APIRouter(prefix="/api/runtime", tags=["runtime"])
 PROJECT_ROOT = Path(__file__).resolve().parents[2]
 
 
+def _normalize_schedule_mode(value: Any) -> str:
+    """Normalize schedule mode to the current public vocabulary.
+
+    `intraday` is kept as a backward-compatible alias for `interval`.
+    """
+    mode = str(value or "daily").strip().lower()
+    if mode == "intraday":
+        return "interval"
+    return mode or "daily"
+
+
 class RuntimeState:
     """Thread-safe singleton for managing runtime state.
 
@@ -439,6 +450,11 @@ def _is_gateway_running() -> bool:
 
     Checks both the internally-managed gateway process and falls back to
     port availability (for externally-managed gateway processes).
+
+    The fallback matters because this codebase may still encounter two startup
+    shapes while historical artifacts remain in-tree:
+    1. runtime_service-managed Gateway subprocesses
+    2. externally started historical Gateway processes outside the supported dev flow
     """
     process = _runtime_state.gateway_process
     if process is not None and process.poll() is None:
@@ -481,7 +497,11 @@ def _start_gateway_process(
     bootstrap: Dict[str, Any],
     port: int
 ) -> subprocess.Popen:
-    """Start Gateway as a separate process."""
+    """Start Gateway as a runtime_service-managed subprocess.
+
+    This path is used when runtime lifecycle is driven through the runtime API.
+    It is not the only supported way a Gateway may exist in the current repo.
+    """
     # Validate configuration before starting
     validation_errors = _validate_gateway_config(bootstrap)
     if validation_errors:
@@ -592,9 +612,9 @@ def _validate_gateway_config(bootstrap: Dict[str, Any]) -> List[str]:
                 errors.append("Dates must be in YYYY-MM-DD format")
 
     # Validate schedule mode
-    schedule_mode = bootstrap.get("schedule_mode", "daily")
-    if schedule_mode not in ("daily", "intraday"):
-        errors.append(f"Invalid schedule_mode '{schedule_mode}': must be 'daily' or 'intraday'")
+    schedule_mode = _normalize_schedule_mode(bootstrap.get("schedule_mode", "daily"))
+    if schedule_mode not in ("daily", "interval"):
+        errors.append(f"Invalid schedule_mode '{schedule_mode}': must be 'daily' or 'interval'")
 
     return errors
 
@@ -778,7 +798,7 @@ async def get_runtime_mode() -> RuntimeModeResponse:
             mode=mode,
             is_backtest=mode == "backtest",
             run_id=context.get("config_name"),
-            schedule_mode=bootstrap.get("schedule_mode"),
+            schedule_mode=_normalize_schedule_mode(bootstrap.get("schedule_mode")),
             is_running=True,
         )
     except HTTPException:
@@ -909,7 +929,7 @@ def _resolve_runtime_response(run_id: str) -> RuntimeConfigResponse:
         project_root=PROJECT_ROOT,
         config_name=run_id,
         enable_memory=bool(bootstrap.get("enable_memory", False)),
-        schedule_mode=str(bootstrap.get("schedule_mode", "daily")),
+        schedule_mode=_normalize_schedule_mode(bootstrap.get("schedule_mode", "daily")),
         interval_minutes=int(bootstrap.get("interval_minutes", 60) or 60),
         trigger_time=str(bootstrap.get("trigger_time", "09:30") or "09:30"),
     )
@@ -929,11 +949,11 @@ def _normalize_runtime_config_updates(
     updates: Dict[str, Any] = {}
 
     if request.schedule_mode is not None:
-        schedule_mode = str(request.schedule_mode).strip().lower()
-        if schedule_mode not in {"daily", "intraday"}:
+        schedule_mode = _normalize_schedule_mode(request.schedule_mode)
+        if schedule_mode not in {"daily", "interval"}:
             raise HTTPException(
                 status_code=400,
-                detail="schedule_mode must be 'daily' or 'intraday'",
+                detail="schedule_mode must be 'daily' or 'interval'",
             )
         updates["schedule_mode"] = schedule_mode
 

backend/apps/agent_service.py

@@ -31,8 +31,7 @@ def _build_scope_payload(project_root: Path) -> dict[str, object]:
         },
         "agent_route_note": (
             "Runtime routes use `/api/runs/{run_id}/agents/...`. "
-            "Legacy `/api/workspaces/{workspace_id}/agents/...` routes are deprecated "
-            "but remain for backward compatibility."
+            "Design-time CRUD routes use `/api/workspaces/{workspace_id}/agents/...`."
         ),
     }
 

backend/apps/runtime_service.py

@@ -5,7 +5,7 @@ from __future__ import annotations
 
 from fastapi import FastAPI
 
-from backend.api import runtime_router
+from backend.api import runtime_router, dynamic_team_router
 from backend.api.runtime import get_runtime_state, _check_gateway_health, _get_gateway_process_details
 from backend.apps.cors import add_cors_middleware
 
@@ -78,6 +78,7 @@ def create_app() -> FastAPI:
         }
 
     app.include_router(runtime_router)
+    app.include_router(dynamic_team_router)
     return app
 
 

backend/config/bootstrap_config.py

@@ -131,6 +131,13 @@ def _coerce_bool(value: Any) -> bool:
     return bool(value)
 
 
+def _normalize_schedule_mode(value: Any) -> str:
+    mode = str(value or "daily").strip().lower()
+    if mode == "intraday":
+        return "interval"
+    return mode or "daily"
+
+
 def resolve_runtime_config(
     project_root: Path,
     config_name: str,
@@ -162,9 +169,9 @@ def resolve_runtime_config(
                 get_env_int("MAX_COMM_CYCLES", 2),
             ),
         ),
-        "schedule_mode": str(
+        "schedule_mode": _normalize_schedule_mode(
             bootstrap.get("schedule_mode", schedule_mode),
-        ).strip().lower() or schedule_mode,
+        ),
         "interval_minutes": int(
             bootstrap.get(
                 "interval_minutes",

backend/core/apo.py

@@ -0,0 +1,197 @@
+# -*- coding: utf-8 -*-
+"""
+Autonomous Policy Optimizer (APO)
+Automatically tunes agent policies based on performance feedback.
+"""
+
+import logging
+import json
+import os
+from datetime import datetime
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+from agentscope.message import Msg
+from backend.llm.models import get_agent_model, get_agent_formatter
+from backend.agents.workspace_manager import WorkspaceManager
+
+logger = logging.getLogger(__name__)
+
+class PolicyOptimizer:
+    """
+    PolicyOptimizer analyzes trading performance and automatically updates
+    agent workspace files (POLICY.md, AGENTS.md) to improve future results.
+    """
+
+    def __init__(self, config_name: str, project_root: Optional[Path] = None):
+        self.config_name = config_name
+        self.workspace_manager = WorkspaceManager(project_root=project_root)
+        # Use a high-capability model for the optimizer (meta-agent)
+        self.model = get_agent_model("portfolio_manager") 
+        self.formatter = get_agent_formatter("portfolio_manager")
+
+    async def run_optimization(
+        self,
+        date: str,
+        reflection_content: str,
+        settlement_result: Dict[str, Any],
+        analyst_results: List[Dict[str, Any]],
+        decisions: Dict[str, Dict],
+    ) -> Dict[str, Any]:
+        """
+        Run the optimization loop if performance indicates a need for change.
+        """
+        total_pnl = settlement_result.get("portfolio_value", 0) - 100000.0 # Assuming 100k initial
+        # You might want to use a more sophisticated trigger, like 3 consecutive losses
+        if total_pnl >= 0:
+            logger.info(f"APO: Positive P&L (${total_pnl:,.2f}) for {date}, skipping optimization.")
+            return {"status": "skipped", "reason": "positive_pnl"}
+
+        logger.info(f"APO: Negative P&L (${total_pnl:,.2f}) detected for {date}. Starting optimization...")
+
+        # 1. Identify underperforming agents or logic
+        # 2. Generate policy updates
+        # 3. Apply updates
+        
+        optimizations = []
+        
+        # Focus on agents that gave high confidence but wrong direction
+        underperformers = self._identify_underperformers(settlement_result, analyst_results)
+        
+        for agent_id in underperformers:
+            update = await self._generate_policy_update(
+                agent_id, 
+                date, 
+                reflection_content, 
+                settlement_result,
+                analyst_results,
+                decisions
+            )
+            if update:
+                self._apply_update(agent_id, update)
+                optimizations.append({
+                    "agent_id": agent_id,
+                    "file": update.get("file", "POLICY.md"),
+                    "change": update.get("change", "")
+                })
+
+        return {
+            "status": "completed",
+            "date": date,
+            "total_pnl": total_pnl,
+            "optimizations": optimizations
+        }
+
+    def _identify_underperformers(
+        self, 
+        settlement_result: Dict[str, Any], 
+        analyst_results: List[Dict[str, Any]]
+    ) -> List[str]:
+        """Identify which agents might need policy adjustments."""
+        underperformers = []
+        
+        # Simple logic: if the overall day was a loss, all active analysts might need a check,
+        # but specifically those whose predictions didn't match the market.
+        # For now, let's include all analysts involved in the day.
+        for result in analyst_results:
+            agent_id = result.get("agent")
+            if agent_id:
+                underperformers.append(agent_id)
+        
+        # Also include PM and Risk Manager as they are critical
+        underperformers.append("portfolio_manager")
+        underperformers.append("risk_manager")
+        
+        return list(set(underperformers))
+
+    async def _generate_policy_update(
+        self,
+        agent_id: str,
+        date: str,
+        reflection_content: str,
+        settlement_result: Dict[str, Any],
+        analyst_results: List[Dict[str, Any]],
+        decisions: Dict[str, Dict],
+    ) -> Optional[Dict[str, str]]:
+        """Use LLM to generate a specific policy update for an agent."""
+        
+        # Load current policy
+        try:
+            current_policy = self.workspace_manager.load_agent_file(
+                config_name=self.config_name,
+                agent_id=agent_id,
+                filename="POLICY.md"
+            )
+        except Exception:
+            current_policy = "No existing policy found."
+
+        prompt = f"""
+As an Expert Meta-Optimizer for a multi-agent trading system, your task is to update the operational POLICY for an agent named '{agent_id}' based on recent performance failures.
+
+[Current Context]
+Date: {date}
+Daily Reflection:
+{reflection_content}
+
+[Agent's Current POLICY.md]
+{current_policy}
+
+[Task]
+Analyze why the system failed (loss occurred). Identify what '{agent_id}' could have done differently or what new constraint/heuristic should be added to its policy to prevent similar mistakes in the future.
+
+Provide a specific, concise addition or modification to the POLICY.md file. 
+The output MUST be a JSON object with:
+1. "reasoning": Brief explanation of why this change is needed.
+2. "file": Always "POLICY.md".
+3. "change": The EXACT markdown text to APPEND or REPLACE in the file. Keep it in Chinese as the system uses Chinese prompts.
+
+Output ONLY the JSON object.
+"""
+        msg = Msg(name="system", content=prompt, role="user")
+        response = await self.model.reply(msg)
+        
+        content = response.content
+        if isinstance(content, list):
+            content = content[0].get("text", "")
+            
+        # Clean JSON if wrapped in markdown
+        if "```json" in content:
+            content = content.split("```json")[1].split("```")[0].strip()
+        
+        try:
+            return json.loads(content)
+        except Exception as e:
+            logger.error(f"APO: Failed to parse optimization response for {agent_id}: {e}")
+            return None
+
+    def _apply_update(self, agent_id: str, update: Dict[str, str]) -> None:
+        """Apply the suggested update to the agent's workspace."""
+        filename = update.get("file", "POLICY.md")
+        change = update.get("change", "")
+        
+        if not change:
+            return
+
+        try:
+            current_content = self.workspace_manager.load_agent_file(
+                config_name=self.config_name,
+                agent_id=agent_id,
+                filename=filename
+            )
+            
+            # Check if change is already there to avoid duplicates
+            if change.strip() in current_content:
+                logger.info(f"APO: Change already present in {agent_id}/{filename}")
+                return
+
+            new_content = current_content + "\n\n### APO Update (" + datetime.now().strftime("%Y-%m-%d") + ")\n" + change
+            
+            self.workspace_manager.update_agent_file(
+                config_name=self.config_name,
+                agent_id=agent_id,
+                filename=filename,
+                content=new_content
+            )
+            logger.info(f"APO: Updated {agent_id}/{filename} with new heuristics.")
+        except Exception as e:
+            logger.error(f"APO: Failed to apply update to {agent_id}/{filename}: {e}")

backend/core/pipeline.py

@@ -33,6 +33,8 @@ from backend.agents.workspace_manager import WorkspaceManager
 from backend.agents.prompt_loader import get_prompt_loader
 from backend.llm.models import get_agent_formatter, get_agent_model
 from backend.config.constants import ANALYST_TYPES
+from backend.agents.dynamic_team_types import AnalystConfig
+from backend.tools.dynamic_team_tools import DynamicTeamController, set_controller
 
 
 def _resolve_evo_agent_ids() -> set[str]:
@@ -84,6 +86,9 @@ def _log(msg: str) -> None:
     logger.info(msg)
 
 
+from backend.core.apo import PolicyOptimizer
+
+
 class TradingPipeline:
     """
     Trading Pipeline - Orchestrates the complete trading cycle
@@ -127,7 +132,21 @@ class TradingPipeline:
         self.runtime_manager = runtime_manager
         self._session_key: Optional[str] = None
         self._dynamic_analysts: Dict[str, Any] = {}
+        self._dynamic_analyst_configs: Dict[str, AnalystConfig] = {}
 
+        # Initialize APO (Autonomous Policy Optimizer)
+        config_name = workspace_id or (runtime_manager.config_name if runtime_manager else "default")
+        self.apo = PolicyOptimizer(config_name=config_name)
+
+        # Initialize dynamic team controller and inject into PM
+        self._team_controller = DynamicTeamController(
+            create_callback=self._create_runtime_analyst,
+            remove_callback=self._remove_runtime_analyst,
+            get_analysts_callback=self._all_analysts,
+        )
+        set_controller(self._team_controller)
+
+        # Backward compatibility: also set individual callbacks if PM expects them
         if hasattr(self.pm, "set_team_controller"):
             self.pm.set_team_controller(
                 create_agent_callback=self._create_runtime_analyst,
@@ -150,23 +169,7 @@ class TradingPipeline:
         execute_decisions: bool = True,
     ) -> Dict[str, Any]:
         """
-        Run one complete trading cycle
-
-        Args:
-            tickers: List of stock tickers
-            date: Trading date (YYYY-MM-DD)
-            prices: Open prices {ticker: price} (for backtest)
-            close_prices: Close prices for settlement (for backtest)
-            market_caps: Optional market caps for baseline calculation
-            get_open_prices_fn: Async callback to wait for open prices (live mode)
-            get_close_prices_fn: Async callback to wait for close prices (live mode)
-
-        For live mode:
-        - Analysis runs immediately
-        - Execution waits for market open via get_open_prices_fn
-        - Settlement waits for market close via get_close_prices_fn
-
-        Each agent's result is broadcast immediately via StateSync.
+        Run one complete trading cycle with checkpointing support.
         """
         _log(f"Starting cycle {date} - {len(tickers)} tickers")
         session_key = TradingSessionKey(date=date).key()
@@ -176,14 +179,45 @@ class TradingPipeline:
             agents=active_analysts + [self.risk_manager, self.pm],
             session_key=session_key,
         )
+
+        # Load checkpoint if exists
+        checkpoint = self._load_checkpoint(session_key)
+        checkpoint_data = checkpoint.get("data", {}) if checkpoint else {}
+        last_phase = checkpoint.get("phase") if checkpoint else None
+
+        if checkpoint:
+            _log(f"Resuming from checkpoint: {last_phase}")
+            # Restore state from checkpoint
+            analyst_results = checkpoint_data.get("analyst_results", [])
+            risk_assessment = checkpoint_data.get("risk_assessment", {})
+            self.conference_summary = checkpoint_data.get("conference_summary")
+            final_predictions = checkpoint_data.get("final_predictions", [])
+            pm_result = checkpoint_data.get("pm_result", {})
+            execution_result = checkpoint_data.get("execution_result", {})
+            settlement_result = checkpoint_data.get("settlement_result")
+            # Prefer passed prices if not hold in checkpoint
+            if not prices:
+                prices = checkpoint_data.get("prices")
+            if not close_prices:
+                close_prices = checkpoint_data.get("close_prices")
+        else:
+            analyst_results = []
+            risk_assessment = {}
+            self.conference_summary = None
+            final_predictions = []
+            pm_result = {}
+            execution_result = {}
+            settlement_result = None
+
         if self.runtime_manager:
             self.runtime_manager.set_session_key(session_key)
-            self._runtime_log_event("cycle:start", {"tickers": tickers, "date": date})
+            self._runtime_log_event("cycle:start", {"tickers": tickers, "date": date, "resumed": checkpoint is not None})
             self._runtime_batch_status(active_analysts, "analysis_in_progress")
 
-        # Phase 0: Clear short-term memory to avoid cross-day context pollution
-        _log("Phase 0: Clearing memory")
-        await self._clear_all_agent_memory()
+        # Phase 0: Clear memory (only if not resuming or if resuming from very start)
+        if not last_phase:
+            _log("Phase 0: Clearing memory")
+            await self._clear_all_agent_memory()
 
         participants = self._all_analysts() + [self.risk_manager, self.pm]
 
@@ -196,125 +230,219 @@ class TradingPipeline:
                 "system",
             ),
         ):
-            # Phase 1.1: Analysts (parallel execution with TeamCoordinator)
-            _log("Phase 1.1: Analyst analysis (parallel)")
-            analyst_results = await self._run_analysts_parallel(
-                tickers,
-                date,
-                active_analysts=active_analysts,
-            )
+            # Phase 1.1: Analysts
+            if not last_phase or last_phase == "cleared":
+                _log("Phase 1.1: Analyst analysis (parallel)")
+                analyst_results = await self._run_analysts_parallel(
+                    tickers,
+                    date,
+                    active_analysts=active_analysts,
+                )
+                self._save_checkpoint(session_key, "analysis", {
+                    "analyst_results": analyst_results,
+                    "prices": prices,
+                    "close_prices": close_prices
+                })
+                last_phase = "analysis"
 
             # Phase 1.2: Risk Manager
-            _log("Phase 1.2: Risk assessment")
-            self._runtime_update_status(self.risk_manager, "risk_assessment")
-            risk_assessment = await self._run_risk_manager_with_sync(
-                tickers,
-                date,
-                prices,
-            )
+            if last_phase == "analysis":
+                _log("Phase 1.2: Risk assessment")
+                self._runtime_update_status(self.risk_manager, "risk_assessment")
+                risk_assessment = await self._run_risk_manager_with_sync(
+                    tickers,
+                    date,
+                    prices,
+                )
+                self._save_checkpoint(session_key, "risk_assessment", {
+                    "analyst_results": analyst_results,
+                    "risk_assessment": risk_assessment,
+                    "prices": prices,
+                    "close_prices": close_prices
+                })
+                last_phase = "risk_assessment"
 
-            # Phase 2.1: Conference discussion (within same MsgHub)
-            _log("Phase 2.1: Conference discussion")
-            conference_summary = await self._run_conference_cycles(
-                tickers=tickers,
-                date=date,
-                prices=prices,
-                analyst_results=analyst_results,
-                risk_assessment=risk_assessment,
-            )
-            self.conference_summary = conference_summary
+            # Phase 2.1: Conference discussion
+            if last_phase == "risk_assessment":
+                _log("Phase 2.1: Conference discussion")
+                conference_summary = await self._run_conference_cycles(
+                    tickers=tickers,
+                    date=date,
+                    prices=prices,
+                    analyst_results=analyst_results,
+                    risk_assessment=risk_assessment,
+                )
+                self.conference_summary = conference_summary
+                self._save_checkpoint(session_key, "conference", {
+                    "analyst_results": analyst_results,
+                    "risk_assessment": risk_assessment,
+                    "conference_summary": conference_summary,
+                    "prices": prices,
+                    "close_prices": close_prices
+                })
+                last_phase = "conference"
 
             # Phase 2.2: Analysts generate final structured predictions
-            _log("Phase 2.2: Analysts generate final structured predictions")
-            final_predictions = await self._collect_final_predictions(
-                tickers,
-                date,
-                active_analysts=active_analysts,
-            )
+            if last_phase == "conference":
+                _log("Phase 2.2: Analysts generate final structured predictions")
+                final_predictions = await self._collect_final_predictions(
+                    tickers,
+                    date,
+                    active_analysts=active_analysts,
+                )
+                self._save_checkpoint(session_key, "predictions", {
+                    "analyst_results": analyst_results,
+                    "risk_assessment": risk_assessment,
+                    "conference_summary": conference_summary,
+                    "final_predictions": final_predictions,
+                    "prices": prices,
+                    "close_prices": close_prices
+                })
+                last_phase = "predictions"
 
-            # Record final predictions for leaderboard ranking
-            if self.settlement_coordinator:
+            # Record final predictions
+            if last_phase == "predictions" and self.settlement_coordinator:
                 self.settlement_coordinator.record_analyst_predictions(
                     final_predictions,
                 )
 
-            # Live mode: wait for market open before execution
-            if get_open_prices_fn:
+            # Live mode: wait for market open
+            if not prices and get_open_prices_fn:
                 _log("Waiting for market open...")
                 prices = await get_open_prices_fn()
                 _log(f"Got open prices: {prices}")
+                # Update prices in checkpoint if we just got them
+                self._save_checkpoint(session_key, "predictions", {
+                    "analyst_results": analyst_results,
+                    "risk_assessment": risk_assessment,
+                    "conference_summary": conference_summary,
+                    "final_predictions": final_predictions,
+                    "prices": prices,
+                    "close_prices": close_prices
+                })
 
             # Phase 3: PM makes decisions
-            _log("Phase 3.1: PM makes decisions")
-            self._runtime_update_status(self.pm, "decision_phase")
-            pm_result = await self._run_pm_with_sync(
-                tickers,
-                date,
-                prices,
-                analyst_results,
-                risk_assessment,
-            )
+            if last_phase == "predictions":
+                _log("Phase 3.1: PM makes decisions")
+                self._runtime_update_status(self.pm, "decision_phase")
+                pm_result = await self._run_pm_with_sync(
+                    tickers,
+                    date,
+                    prices,
+                    analyst_results,
+                    risk_assessment,
+                )
+                self._save_checkpoint(session_key, "decisions", {
+                    "analyst_results": analyst_results,
+                    "risk_assessment": risk_assessment,
+                    "conference_summary": conference_summary,
+                    "final_predictions": final_predictions,
+                    "pm_result": pm_result,
+                    "prices": prices,
+                    "close_prices": close_prices
+                })
+                last_phase = "decisions"
 
-        decisions = pm_result.get("decisions", {})
-        execution_result = {
-            "executed_trades": [],
-            "portfolio": self.pm.get_portfolio_state(),
-        }
-        if execute_decisions:
-            _log("Phase 4: Executing trades")
-            self._runtime_update_status(self.pm, "executing")
-            execution_result = self._execute_decisions(decisions, prices, date)
-        else:
-            _log("Phase 4: Skipping trade execution")
+        # Outside MsgHub for execution and settlement
+        decisions = pm_result.get("decisions", {}) if pm_result else {}
+        if not execution_result:
+            execution_result = {
+                "executed_trades": [],
+                "portfolio": self.pm.get_portfolio_state(),
+            }
 
-        # Live mode: wait for market close before settlement
-        if get_close_prices_fn:
+        if last_phase == "decisions":
+            if execute_decisions:
+                _log("Phase 4: Executing trades")
+                self._runtime_update_status(self.pm, "executing")
+                execution_result = self._execute_decisions(decisions, prices, date)
+            else:
+                _log("Phase 4: Skipping trade execution")
+            
+            self._save_checkpoint(session_key, "execution", {
+                "analyst_results": analyst_results,
+                "risk_assessment": risk_assessment,
+                "conference_summary": conference_summary,
+                "final_predictions": final_predictions,
+                "pm_result": pm_result,
+                "execution_result": execution_result,
+                "prices": prices,
+                "close_prices": close_prices
+            })
+            last_phase = "execution"
+
+        # Live mode: wait for market close
+        if not close_prices and get_close_prices_fn:
             _log("Waiting for market close")
             close_prices = await get_close_prices_fn()
             _log(f"Got close prices: {close_prices}")
+            # Update close_prices in checkpoint
+            self._save_checkpoint(session_key, "execution", {
+                "analyst_results": analyst_results,
+                "risk_assessment": risk_assessment,
+                "conference_summary": conference_summary,
+                "final_predictions": final_predictions,
+                "pm_result": pm_result,
+                "execution_result": execution_result,
+                "prices": prices,
+                "close_prices": close_prices
+            })
 
-        # Phase 5: Settlement - run after close prices available
-        settlement_result = None
-        if close_prices and self.settlement_coordinator:
-            _log("Phase 5: Daily review and generate memories")
-            self._runtime_batch_status(
-                [self.risk_manager] + self._all_analysts() + [self.pm],
-                "settlement",
-            )
+        # Phase 5: Settlement
+        if last_phase == "execution":
+            if close_prices and self.settlement_coordinator:
+                _log("Phase 5: Daily review and generate memories")
+                self._runtime_batch_status(
+                    [self.risk_manager] + self._all_analysts() + [self.pm],
+                    "settlement",
+                )
 
-            agent_trajectories = await self._capture_agent_trajectories()
+                agent_trajectories = await self._capture_agent_trajectories()
 
-            if market_caps is None:
-                market_caps = {ticker: 1e9 for ticker in tickers}
+                if market_caps is None:
+                    market_caps = {ticker: 1e9 for ticker in tickers}
 
-            settlement_result = (
-                self.settlement_coordinator.run_daily_settlement(
+                settlement_result = (
+                    self.settlement_coordinator.run_daily_settlement(
+                        date=date,
+                        tickers=tickers,
+                        open_prices=prices,
+                        close_prices=close_prices,
+                        market_caps=market_caps,
+                        agent_portfolio=execution_result.get("portfolio", {}),
+                        analyst_results=analyst_results,
+                        pm_decisions=decisions,
+                    )
+                )
+
+                await self._run_reflection(
                     date=date,
-                    tickers=tickers,
+                    agent_trajectories=agent_trajectories,
+                    analyst_results=analyst_results,
+                    decisions=decisions,
+                    executed_trades=execution_result.get("executed_trades", []),
                     open_prices=prices,
                     close_prices=close_prices,
-                    market_caps=market_caps,
-                    agent_portfolio=execution_result.get("portfolio", {}),
-                    analyst_results=analyst_results,
-                    pm_decisions=decisions,
+                    settlement_result=settlement_result,
+                    conference_summary=self.conference_summary,
                 )
-            )
-
-            await self._run_reflection(
-                date=date,
-                agent_trajectories=agent_trajectories,
-                analyst_results=analyst_results,
-                decisions=decisions,
-                executed_trades=execution_result.get("executed_trades", []),
-                open_prices=prices,
-                close_prices=close_prices,
-                settlement_result=settlement_result,
-                conference_summary=self.conference_summary,
-            )
-            self._runtime_batch_status(
-                [self.risk_manager] + self._all_analysts() + [self.pm],
-                "reflection",
-            )
+                self._runtime_batch_status(
+                    [self.risk_manager] + self._all_analysts() + [self.pm],
+                    "reflection",
+                )
+            
+            self._save_checkpoint(session_key, "settlement", {
+                "analyst_results": analyst_results,
+                "risk_assessment": risk_assessment,
+                "conference_summary": conference_summary,
+                "final_predictions": final_predictions,
+                "pm_result": pm_result,
+                "execution_result": execution_result,
+                "settlement_result": settlement_result,
+                "prices": prices,
+                "close_prices": close_prices
+            })
+            last_phase = "settlement"
 
         _log(f"Cycle complete: {date}")
         self._runtime_batch_status(
@@ -323,6 +451,11 @@ class TradingPipeline:
         )
         self._runtime_log_event("cycle:end", {"tickers": tickers, "date": date})
 
+        # Optional: Clean up checkpoint after successful completion
+        # path = self._get_checkpoint_path(session_key)
+        # if path and path.exists():
+        #     path.unlink()
+
         return {
             "analyst_results": analyst_results,
             "risk_assessment": risk_assessment,
@@ -385,6 +518,44 @@ class TradingPipeline:
         await self.risk_manager.memory.clear()
         await self.pm.memory.clear()
 
+    def _get_checkpoint_path(self, session_key: str) -> Optional[Path]:
+        """Get the path to the pipeline checkpoint file."""
+        if not self.runtime_manager or not self.runtime_manager.run_dir:
+            return None
+        checkpoint_dir = self.runtime_manager.run_dir / "state" / "checkpoints"
+        checkpoint_dir.mkdir(parents=True, exist_ok=True)
+        return checkpoint_dir / f"pipeline_{session_key}.json"
+
+    def _save_checkpoint(self, session_key: str, phase: str, data: Dict[str, Any]) -> None:
+        """Save the current pipeline state to a checkpoint file."""
+        path = self._get_checkpoint_path(session_key)
+        if not path:
+            return
+
+        checkpoint = {
+            "session_key": session_key,
+            "phase": phase,
+            "timestamp": datetime.now().isoformat(),
+            "data": data
+        }
+        try:
+            path.write_text(json.dumps(checkpoint, ensure_ascii=False, indent=2, default=str), encoding="utf-8")
+            _log(f"Checkpoint saved: {phase} for {session_key}")
+        except Exception as e:
+            logger.error(f"Failed to save checkpoint: {e}")
+
+    def _load_checkpoint(self, session_key: str) -> Optional[Dict[str, Any]]:
+        """Load the pipeline state from a checkpoint file."""
+        path = self._get_checkpoint_path(session_key)
+        if not path or not path.exists():
+            return None
+
+        try:
+            return json.loads(path.read_text(encoding="utf-8"))
+        except Exception as e:
+            logger.error(f"Failed to load checkpoint: {e}")
+            return None
+
     async def _sync_memory_if_retrieved(self, agent: Any) -> None:
         """
         Check agent's short-term memory for retrieved long-term memory and sync to frontend.
@@ -585,6 +756,25 @@ class TradingPipeline:
                 content=reflection_content,
             )
 
+        # Phase 6: APO (Autonomous Policy Optimization)
+        # If the day was a loss, let APO suggest and apply policy updates.
+        if hasattr(self, "apo") and self.apo:
+            _log(f"Phase 6: APO - Running autonomous policy optimization for {date}")
+            try:
+                apo_result = await self.apo.run_optimization(
+                    date=date,
+                    reflection_content=reflection_content,
+                    settlement_result=settlement_result or {"portfolio_value": 100000.0 + total_pnl},
+                    analyst_results=analyst_results,
+                    decisions=decisions
+                )
+                if apo_result.get("status") == "completed":
+                    _log(f"APO: Successfully applied {len(apo_result.get('optimizations', []))} policy updates.")
+                    # Reload assets for next cycle to ensure they are picked up
+                    self.reload_runtime_assets()
+            except Exception as e:
+                logger.error(f"APO: Optimization failed: {e}")
+
     def _build_reflection_content(
         self,
         date: str,
@@ -1562,28 +1752,74 @@ class TradingPipeline:
         """Return static analysts plus runtime-created analysts."""
         return list(self.analysts) + list(self._dynamic_analysts.values())
 
-    def _create_runtime_analyst(self, agent_id: str, analyst_type: str) -> str:
-        """Create one runtime analyst instance."""
-        if analyst_type not in ANALYST_TYPES:
+    def _create_runtime_analyst(
+        self,
+        agent_id: str,
+        analyst_type: str,
+        custom_config: Optional[AnalystConfig] = None,
+    ) -> str:
+        """Create one runtime analyst instance.
+
+        Args:
+            agent_id: Unique identifier for the new analyst
+            analyst_type: Type of analyst (e.g., "technical_analyst")
+            custom_config: Optional custom configuration for the analyst,
+                          including persona, soul_md, agents_md, etc.
+
+        Returns:
+            Success or error message
+        """
+        # Validate analyst_type or custom_config
+        if analyst_type not in ANALYST_TYPES and not custom_config:
             return (
                 f"Unknown analyst_type '{analyst_type}'. "
-                f"Available: {', '.join(ANALYST_TYPES.keys())}"
+                f"Available: {', '.join(ANALYST_TYPES.keys())}. "
+                f"Or provide custom_config to create a custom analyst."
             )
         if agent_id in {agent.name for agent in self._all_analysts()}:
             return f"Analyst '{agent_id}' already exists."
 
         config_name = getattr(self.pm, "config", {}).get("config_name", "default")
         project_root = Path(__file__).resolve().parents[2]
-        personas = get_prompt_loader().load_yaml_config("analyst", "personas")
-        persona = personas.get(analyst_type, {})
+
+        # Get persona: use custom_config if provided, else load from personas.yaml
+        if custom_config and custom_config.persona:
+            persona = {
+                "name": custom_config.persona.name,
+                "focus": custom_config.persona.focus,
+                "description": custom_config.persona.description,
+            }
+        else:
+            personas = get_prompt_loader().load_yaml_config("analyst", "personas")
+            persona = personas.get(analyst_type, {})
         workspace_manager = WorkspaceManager(project_root=project_root)
+
+        # Build file contents: use custom if provided, else generate from persona
+        file_contents = {}
+        if custom_config:
+            if custom_config.soul_md:
+                file_contents["SOUL.md"] = custom_config.soul_md
+            if custom_config.agents_md:
+                file_contents["AGENTS.md"] = custom_config.agents_md
+            if custom_config.profile_md:
+                file_contents["PROFILE.md"] = custom_config.profile_md
+            if custom_config.bootstrap_md:
+                file_contents["BOOTSTRAP.md"] = custom_config.bootstrap_md
+
+        # Fill in any missing files with defaults
+        if not file_contents or len(file_contents) < 4:
+            default_files = workspace_manager.build_default_agent_files(
+                agent_id=agent_id,
+                persona=persona,
+            )
+            for key, value in default_files.items():
+                if key not in file_contents:
+                    file_contents[key] = value
+
         workspace_manager.ensure_agent_assets(
             config_name=config_name,
             agent_id=agent_id,
-            file_contents=workspace_manager.build_default_agent_files(
-                agent_id=agent_id,
-                persona=persona,
-            ),
+            file_contents=file_contents,
         )
 
         # Create EvoAgent with workspace-driven configuration
@@ -1594,11 +1830,23 @@ class TradingPipeline:
             agent_id,
         )
         agent_config = load_agent_workspace_config(workspace_dir / "agent.yaml")
+        # Support model override from custom_config
+        if custom_config and custom_config.model_name:
+            # Import create_model for custom model creation
+            from backend.llm.models import create_model
+            # Use specified model name, default to openai provider
+            model = create_model(
+                model_name=custom_config.model_name,
+                model_provider=custom_config.memory_config.get("model_provider", "openai") if custom_config.memory_config else "openai"
+            )
+        else:
+            model = get_agent_model(analyst_type)
+
         agent = EvoAgent(
             agent_id=agent_id,
             config_name=config_name,
             workspace_dir=workspace_dir,
-            model=get_agent_model(analyst_type),
+            model=model,
             formatter=get_agent_formatter(analyst_type),
             prompt_files=agent_config.prompt_files,
         )
@@ -1611,6 +1859,11 @@ class TradingPipeline:
         # Keep workspace_id for backward compatibility
         setattr(agent, "workspace_id", config_name)
         self._dynamic_analysts[agent_id] = agent
+
+        # Store custom config for future reference (e.g., cloning)
+        if custom_config:
+            self._dynamic_analyst_configs[agent_id] = custom_config
+
         update_active_analysts(
             project_root=project_root,
             config_name=config_name,
@@ -1624,6 +1877,8 @@ class TradingPipeline:
         if agent_id not in self._dynamic_analysts:
             return f"Runtime analyst '{agent_id}' not found."
         self._dynamic_analysts.pop(agent_id, None)
+        # Also remove stored config if exists
+        self._dynamic_analyst_configs.pop(agent_id, None)
         config_name = getattr(self.pm, "config", {}).get("config_name", "default")
         project_root = Path(__file__).resolve().parents[2]
         update_active_analysts(

backend/core/pipeline_runner.py

@@ -4,6 +4,12 @@ Pipeline Runner - Independent trading pipeline execution
 
 This module provides functions to start/stop trading pipelines
 that can be called from the REST API.
+
+COMPATIBILITY_NOTE:
+This module still carries selected fallback creation paths used by managed
+runtime startup and compatibility flows. New runtime behavior should be judged
+against the run-scoped helpers and current pipeline selection rules rather than
+assuming every constructor here is the long-term default.
 """
 
 from __future__ import annotations
@@ -11,6 +17,7 @@ from __future__ import annotations
 import asyncio
 import os
 from contextlib import AsyncExitStack
+from dataclasses import dataclass
 from pathlib import Path
 from typing import Any, Dict, List, Optional, Callable
 
@@ -22,7 +29,7 @@ from backend.agents.prompt_loader import get_prompt_loader
 from backend.agents.workspace_manager import WorkspaceManager
 from backend.config.constants import ANALYST_TYPES
 from backend.core.pipeline import TradingPipeline
-from backend.core.scheduler import BacktestScheduler, Scheduler
+from backend.core.scheduler import BacktestScheduler, Scheduler, normalize_schedule_mode
 from backend.llm.models import get_agent_formatter, get_agent_model
 from backend.runtime.manager import (
     TradingRuntimeManager,
@@ -46,6 +53,21 @@ _gateway_instance: Optional[Gateway] = None
 _long_term_memories: List[Any] = []
 
 
+@dataclass
+class GatewayRuntimeBundle:
+    """Assembled runtime components for a Gateway-backed execution path."""
+
+    runtime_manager: TradingRuntimeManager
+    market_service: MarketService
+    storage_service: StorageService
+    pipeline: TradingPipeline
+    gateway: Gateway
+    scheduler: Optional[Scheduler]
+    scheduler_callback: Optional[Callable]
+    long_term_memories: List[Any]
+    trading_dates: List[str]
+
+
 def _set_gateway(gateway: Optional[Gateway]) -> None:
     """Set global gateway reference."""
     global _gateway_instance
@@ -443,6 +465,151 @@ def create_agents(
     return analysts, risk_manager, portfolio_manager, long_term_memories
 
 
+def build_gateway_runtime_bundle(
+    *,
+    run_id: str,
+    run_dir: Path,
+    bootstrap: Dict[str, Any],
+    poll_interval: int = 10,
+) -> GatewayRuntimeBundle:
+    """Build the full Gateway runtime component graph for one run."""
+    tickers = bootstrap.get("tickers", ["AAPL", "MSFT", "GOOGL", "AMZN", "NVDA", "META", "TSLA", "AMD", "NFLX", "AVGO", "PLTR", "COIN"])
+    initial_cash = float(bootstrap.get("initial_cash", 100000.0))
+    margin_requirement = float(bootstrap.get("margin_requirement", 0.0))
+    max_comm_cycles = int(bootstrap.get("max_comm_cycles", 2))
+    schedule_mode = normalize_schedule_mode(bootstrap.get("schedule_mode", "daily"))
+    trigger_time = bootstrap.get("trigger_time", "09:30")
+    interval_minutes = int(bootstrap.get("interval_minutes", 60))
+    heartbeat_interval = int(bootstrap.get("heartbeat_interval", 0))
+    mode = bootstrap.get("mode", "live")
+    start_date = bootstrap.get("start_date")
+    end_date = bootstrap.get("end_date")
+    enable_memory = bootstrap.get("enable_memory", False)
+
+    is_backtest = mode == "backtest"
+
+    runtime_manager = TradingRuntimeManager(
+        config_name=run_id,
+        run_dir=run_dir,
+        bootstrap=bootstrap,
+    )
+    runtime_manager.prepare_run()
+
+    market_service = MarketService(
+        tickers=tickers,
+        poll_interval=poll_interval,
+        backtest_mode=is_backtest,
+        api_key=os.getenv("FINNHUB_API_KEY") if not is_backtest else None,
+        backtest_start_date=start_date if is_backtest else None,
+        backtest_end_date=end_date if is_backtest else None,
+    )
+
+    storage_service = StorageService(
+        dashboard_dir=run_dir / "team_dashboard",
+        initial_cash=initial_cash,
+        config_name=run_id,
+    )
+    if not storage_service.files["summary"].exists():
+        storage_service.initialize_empty_dashboard()
+    else:
+        storage_service.update_leaderboard_model_info()
+
+    analysts, risk_manager, pm, long_term_memories = create_agents(
+        run_id=run_id,
+        run_dir=run_dir,
+        initial_cash=initial_cash,
+        margin_requirement=margin_requirement,
+        enable_long_term_memory=enable_memory,
+    )
+    for agent in analysts + [risk_manager, pm]:
+        agent_id = getattr(agent, "agent_id", None) or getattr(agent, "name", None)
+        if agent_id:
+            runtime_manager.register_agent(agent_id)
+
+    portfolio_state = storage_service.load_portfolio_state()
+    pm.load_portfolio_state(portfolio_state)
+
+    settlement_coordinator = SettlementCoordinator(
+        storage=storage_service,
+        initial_capital=initial_cash,
+    )
+    pipeline = TradingPipeline(
+        analysts=analysts,
+        risk_manager=risk_manager,
+        portfolio_manager=pm,
+        settlement_coordinator=settlement_coordinator,
+        max_comm_cycles=max_comm_cycles,
+        runtime_manager=runtime_manager,
+    )
+
+    scheduler_callback = None
+    live_scheduler = None
+    trading_dates: List[str] = []
+
+    if is_backtest:
+        backtest_scheduler = BacktestScheduler(
+            start_date=start_date,
+            end_date=end_date,
+            trading_calendar="NYSE",
+            delay_between_days=0.5,
+        )
+        trading_dates = backtest_scheduler.get_trading_dates()
+
+        async def scheduler_callback_fn(callback):
+            await backtest_scheduler.start(callback)
+
+        scheduler_callback = scheduler_callback_fn
+    else:
+        live_scheduler = Scheduler(
+            mode=schedule_mode,
+            trigger_time=trigger_time,
+            interval_minutes=interval_minutes,
+            heartbeat_interval=heartbeat_interval if heartbeat_interval > 0 else None,
+            config={"config_name": run_id},
+        )
+
+        async def scheduler_callback_fn(callback):
+            await live_scheduler.start(callback)
+
+        scheduler_callback = scheduler_callback_fn
+
+    gateway = Gateway(
+        market_service=market_service,
+        storage_service=storage_service,
+        pipeline=pipeline,
+        scheduler_callback=scheduler_callback,
+        config={
+            "mode": mode,
+            "backtest_mode": is_backtest,
+            "tickers": tickers,
+            "config_name": run_id,
+            "schedule_mode": schedule_mode,
+            "interval_minutes": interval_minutes,
+            "trigger_time": trigger_time,
+            "heartbeat_interval": heartbeat_interval,
+            "initial_cash": initial_cash,
+            "margin_requirement": margin_requirement,
+            "max_comm_cycles": max_comm_cycles,
+            "enable_memory": enable_memory,
+        },
+        scheduler=live_scheduler,
+    )
+    if is_backtest:
+        gateway.set_backtest_dates(trading_dates)
+
+    return GatewayRuntimeBundle(
+        runtime_manager=runtime_manager,
+        market_service=market_service,
+        storage_service=storage_service,
+        pipeline=pipeline,
+        gateway=gateway,
+        scheduler=live_scheduler,
+        scheduler_callback=scheduler_callback,
+        long_term_memories=long_term_memories,
+        trading_dates=trading_dates,
+    )
+
+
 async def run_pipeline(
     run_id: str,
     run_dir: Path,
@@ -483,7 +650,7 @@ async def run_pipeline(
         initial_cash = float(bootstrap.get("initial_cash", 100000.0))
         margin_requirement = float(bootstrap.get("margin_requirement", 0.0))
         max_comm_cycles = int(bootstrap.get("max_comm_cycles", 2))
-        schedule_mode = bootstrap.get("schedule_mode", "daily")
+        schedule_mode = normalize_schedule_mode(bootstrap.get("schedule_mode", "daily"))
         trigger_time = bootstrap.get("trigger_time", "09:30")
         interval_minutes = int(bootstrap.get("interval_minutes", 60))
         heartbeat_interval = int(bootstrap.get("heartbeat_interval", 0))

backend/core/scheduler.py

@@ -17,6 +17,14 @@ NYSE_TZ = ZoneInfo("America/New_York")
 NYSE_CALENDAR = mcal.get_calendar("NYSE")
 
 
+def normalize_schedule_mode(mode: str | None) -> str:
+    """Normalize schedule mode to the current public vocabulary."""
+    value = str(mode or "daily").strip().lower()
+    if value == "intraday":
+        return "interval"
+    return value or "daily"
+
+
 class Scheduler:
     """
     Market-aware scheduler for live trading.
@@ -31,7 +39,7 @@ class Scheduler:
         heartbeat_interval: Optional[int] = None,
         config: Optional[dict] = None,
     ):
-        self.mode = mode
+        self.mode = normalize_schedule_mode(mode)
         self.trigger_time = trigger_time or "09:30"  # NYSE timezone
         self.trigger_now = self.trigger_time == "now"
         self.interval_minutes = interval_minutes or 60
@@ -107,7 +115,7 @@ class Scheduler:
 
         if self.mode == "daily":
             self._task = asyncio.create_task(self._run_daily(self._callback))
-        elif self.mode == "intraday":
+        elif self.mode == "interval":
             self._task = asyncio.create_task(
                 self._run_intraday(self._callback),
             )
@@ -124,8 +132,13 @@ class Scheduler:
         """Update scheduler parameters in-place and restart its timing loop."""
         changed = False
 
-        if mode and mode != self.mode:
-            self.mode = mode
+        if mode:
+            normalized_mode = normalize_schedule_mode(mode)
+        else:
+            normalized_mode = None
+
+        if normalized_mode and normalized_mode != self.mode:
+            self.mode = normalized_mode
             changed = True
 
         if trigger_time and trigger_time != self.trigger_time:
@@ -233,13 +246,13 @@ class Scheduler:
             await callback(date=current_date)
 
     async def _run_intraday(self, callback: Callable):
-        """Run every N minutes (for future use)"""
+        """Run every N minutes in interval mode."""
         while self.running:
             now = self._now_nyse()
             current_date = now.strftime("%Y-%m-%d")
 
             if self._is_trading_day(now):
-                logger.info(f"Triggering intraday cycle for {current_date}")
+                logger.info(f"Triggering interval cycle for {current_date}")
                 await callback(date=current_date)
 
             await asyncio.sleep(self.interval_minutes * 60)

backend/gateway_server.py

@@ -1,15 +1,14 @@
 # -*- coding: utf-8 -*-
-"""Gateway Server - Entry point for Gateway subprocess.
+"""Gateway Server - Entry point for the managed Gateway subprocess.
 
-This module is launched as a subprocess by the Control Plane (FastAPI)
-to run the Data Plane (Gateway + Pipeline).
+This module is launched by `runtime_service` when the runtime API is used to
+spawn a run-scoped Gateway process.
 """
 
 import argparse
 import asyncio
 import json
 import logging
-import os
 import sys
 from contextlib import AsyncExitStack
 from pathlib import Path
@@ -19,22 +18,13 @@ from dotenv import load_dotenv
 # Load environment variables
 load_dotenv()
 
-from backend.agents.prompt_loader import get_prompt_loader
-from backend.core.pipeline import TradingPipeline
-from backend.core.pipeline_runner import create_agents
-from backend.core.scheduler import BacktestScheduler, Scheduler
+from backend.core.pipeline_runner import build_gateway_runtime_bundle
 from backend.runtime.manager import (
-    TradingRuntimeManager,
     set_global_runtime_manager,
     clear_global_runtime_manager,
 )
-from backend.services.gateway import Gateway
-from backend.services.market import MarketService
-from backend.services.storage import StorageService
-from backend.utils.settlement import SettlementCoordinator
 
 logger = logging.getLogger(__name__)
-_prompt_loader = get_prompt_loader()
 
 
 INFO_LOGGER_PREFIXES = (
@@ -110,153 +100,24 @@ async def run_gateway(
     port: int
 ):
     """Run Gateway with Pipeline."""
-
-    # Extract config
-    tickers = bootstrap.get("tickers", ["AAPL", "MSFT", "GOOGL", "AMZN", "NVDA", "META", "TSLA", "AMD", "NFLX", "AVGO", "PLTR", "COIN"])
-    initial_cash = float(bootstrap.get("initial_cash", 100000.0))
-    margin_requirement = float(bootstrap.get("margin_requirement", 0.0))
-    max_comm_cycles = int(bootstrap.get("max_comm_cycles", 2))
-    schedule_mode = bootstrap.get("schedule_mode", "daily")
-    trigger_time = bootstrap.get("trigger_time", "09:30")
-    interval_minutes = int(bootstrap.get("interval_minutes", 60))
-    heartbeat_interval = int(bootstrap.get("heartbeat_interval", 0))  # 0 = disabled
-    mode = bootstrap.get("mode", "live")
-    start_date = bootstrap.get("start_date")
-    end_date = bootstrap.get("end_date")
-    enable_memory = bootstrap.get("enable_memory", False)
     poll_interval = int(bootstrap.get("poll_interval", 10))
 
-    is_backtest = mode == "backtest"
-
     logger.info(f"[Gateway Server] Starting run {run_id} on port {port}")
 
-    # Create runtime manager
-    runtime_manager = TradingRuntimeManager(
-        config_name=run_id,
-        run_dir=run_dir,
-        bootstrap=bootstrap,
-    )
-    runtime_manager.prepare_run()
-    set_global_runtime_manager(runtime_manager)
-
     try:
+        bundle = build_gateway_runtime_bundle(
+            run_id=run_id,
+            run_dir=run_dir,
+            bootstrap=bootstrap,
+            poll_interval=poll_interval,
+        )
+        set_global_runtime_manager(bundle.runtime_manager)
+
         async with AsyncExitStack() as stack:
-            # Create services
-            market_service = MarketService(
-                tickers=tickers,
-                poll_interval=poll_interval,
-                backtest_mode=is_backtest,
-                api_key=os.getenv("FINNHUB_API_KEY") if not is_backtest else None,
-                backtest_start_date=start_date if is_backtest else None,
-                backtest_end_date=end_date if is_backtest else None,
-            )
-
-            storage_service = StorageService(
-                dashboard_dir=run_dir / "team_dashboard",
-                initial_cash=initial_cash,
-                config_name=run_id,
-            )
-
-            if not storage_service.files["summary"].exists():
-                storage_service.initialize_empty_dashboard()
-            else:
-                storage_service.update_leaderboard_model_info()
-
-            # Create agents
-            analysts, risk_manager, pm, long_term_memories = create_agents(
-                run_id=run_id,
-                run_dir=run_dir,
-                initial_cash=initial_cash,
-                margin_requirement=margin_requirement,
-                enable_long_term_memory=enable_memory,
-            )
-
-            # Register agents
-            for agent in analysts + [risk_manager, pm]:
-                agent_id = getattr(agent, "agent_id", None) or getattr(agent, "name", None)
-                if agent_id:
-                    runtime_manager.register_agent(agent_id)
-
-            # Load portfolio state
-            portfolio_state = storage_service.load_portfolio_state()
-            pm.load_portfolio_state(portfolio_state)
-
-            # Create settlement coordinator
-            settlement_coordinator = SettlementCoordinator(
-                storage=storage_service,
-                initial_capital=initial_cash,
-            )
-
-            # Create pipeline
-            pipeline = TradingPipeline(
-                analysts=analysts,
-                risk_manager=risk_manager,
-                portfolio_manager=pm,
-                settlement_coordinator=settlement_coordinator,
-                max_comm_cycles=max_comm_cycles,
-                runtime_manager=runtime_manager,
-            )
-
-            # Create scheduler
-            scheduler_callback = None
-            live_scheduler = None
-
-            if is_backtest:
-                backtest_scheduler = BacktestScheduler(
-                    start_date=start_date,
-                    end_date=end_date,
-                    trading_calendar="NYSE",
-                    delay_between_days=0.5,
-                )
-
-                async def scheduler_callback_fn(callback):
-                    await backtest_scheduler.start(callback)
-
-                scheduler_callback = scheduler_callback_fn
-            else:
-                live_scheduler = Scheduler(
-                    mode=schedule_mode,
-                    trigger_time=trigger_time,
-                    interval_minutes=interval_minutes,
-                    heartbeat_interval=heartbeat_interval if heartbeat_interval > 0 else None,
-                    config={"config_name": run_id},
-                )
-
-                async def scheduler_callback_fn(callback):
-                    await live_scheduler.start(callback)
-
-                scheduler_callback = scheduler_callback_fn
-
-            # Enter long-term memory contexts
-            for memory in long_term_memories:
+            for memory in bundle.long_term_memories:
                 await stack.enter_async_context(memory)
-
-            # Create Gateway
-            gateway = Gateway(
-                market_service=market_service,
-                storage_service=storage_service,
-                pipeline=pipeline,
-                scheduler_callback=scheduler_callback,
-                config={
-                    "mode": mode,
-                    "backtest_mode": is_backtest,
-                    "tickers": tickers,
-                    "config_name": run_id,
-                    "schedule_mode": schedule_mode,
-                    "interval_minutes": interval_minutes,
-                    "trigger_time": trigger_time,
-                    "heartbeat_interval": heartbeat_interval,
-                    "initial_cash": initial_cash,
-                    "margin_requirement": margin_requirement,
-                    "max_comm_cycles": max_comm_cycles,
-                    "enable_memory": enable_memory,
-                },
-                scheduler=live_scheduler,
-            )
-
-            # Start Gateway (blocks until shutdown)
             logger.info(f"[Gateway Server] Gateway starting on port {port}")
-            await gateway.start(host="0.0.0.0", port=port)
+            await bundle.gateway.start(host="0.0.0.0", port=port)
 
     except asyncio.CancelledError:
         logger.info("[Gateway Server] Cancelled")

backend/main.py

@@ -1,596 +0,0 @@
-# -*- coding: utf-8 -*-
-"""
-Main Entry Point
-Supports: backtest, live modes
-"""
-import argparse
-import asyncio
-import logging
-import os
-from contextlib import AsyncExitStack
-from pathlib import Path
-import loguru
-
-from dotenv import load_dotenv
-
-from backend.agents import EvoAgent
-from backend.agents.agent_workspace import load_agent_workspace_config
-from backend.agents.skills_manager import SkillsManager
-from backend.agents.toolkit_factory import create_agent_toolkit, load_agent_profiles
-from backend.agents.prompt_loader import get_prompt_loader
-# WorkspaceManager is RunWorkspaceManager - provides run-scoped asset management
-# All runtime state lives under runs/<run_id>/
-from backend.agents.workspace_manager import WorkspaceManager
-from backend.config.bootstrap_config import resolve_runtime_config
-from backend.config.constants import ANALYST_TYPES
-from backend.core.pipeline import TradingPipeline
-from backend.core.scheduler import BacktestScheduler, Scheduler
-from backend.llm.models import get_agent_formatter, get_agent_model
-from backend.api.runtime import unregister_runtime_manager
-from backend.runtime.manager import (
-    TradingRuntimeManager,
-    set_global_runtime_manager,
-    clear_global_runtime_manager,
-)
-from backend.gateway_server import configure_gateway_logging
-from backend.services.gateway import Gateway
-from backend.services.market import MarketService
-from backend.services.storage import StorageService
-from backend.utils.settlement import SettlementCoordinator
-
-load_dotenv()
-logger = logging.getLogger(__name__)
-loguru.logger.disable("flowllm")
-loguru.logger.disable("reme_ai")
-configure_gateway_logging(verbose=os.getenv("LOG_LEVEL", "").upper() == "DEBUG")
-_prompt_loader = get_prompt_loader()
-
-
-def _get_run_dir(config_name: str) -> Path:
-    """Return the canonical run-scoped directory for a config.
-
-    This is the authoritative path for runtime state under runs/<run_id>/.
-    All runtime assets, state, and exports are scoped to this directory.
-    """
-    project_root = Path(__file__).resolve().parents[1]
-    # Use RunWorkspaceManager for run-scoped path resolution
-    return WorkspaceManager(project_root=project_root).get_run_dir(config_name)
-
-
-def _resolve_runtime_config(args) -> dict:
-    """Merge env defaults with run-scoped bootstrap config."""
-    project_root = Path(__file__).resolve().parents[1]
-    return resolve_runtime_config(
-        project_root=project_root,
-        config_name=args.config_name,
-        enable_memory=args.enable_memory,
-        schedule_mode=args.schedule_mode,
-        interval_minutes=args.interval_minutes,
-        trigger_time=args.trigger_time,
-    )
-
-
-def create_long_term_memory(agent_name: str, config_name: str):
-    """
-    Create ReMeTaskLongTermMemory for an agent
-
-    Requires DASHSCOPE_API_KEY env var
-    """
-    from agentscope.memory import ReMeTaskLongTermMemory
-    from agentscope.model import DashScopeChatModel
-    from agentscope.embedding import DashScopeTextEmbedding
-
-    api_key = os.getenv("MEMORY_API_KEY")
-    if not api_key:
-        logger.warning("MEMORY_API_KEY not set, long-term memory disabled")
-        return None
-
-    memory_dir = str(_get_run_dir(config_name) / "memory")
-
-    return ReMeTaskLongTermMemory(
-        agent_name=agent_name,
-        user_name=agent_name,
-        model=DashScopeChatModel(
-            model_name=os.getenv("MEMORY_MODEL_NAME", "qwen3-max"),
-            api_key=api_key,
-            stream=False,
-        ),
-        embedding_model=DashScopeTextEmbedding(
-            model_name=os.getenv(
-                "MEMORY_EMBEDDING_MODEL",
-                "text-embedding-v4",
-            ),
-            api_key=api_key,
-            dimensions=1024,
-        ),
-        **{
-            "vector_store.default.backend": "local",
-            "vector_store.default.params.store_dir": memory_dir,
-        },
-    )
-
-
-def _resolve_evo_agent_ids() -> set[str]:
-    """Return agent ids selected to use EvoAgent.
-
-    By default, all supported roles use EvoAgent.
-    EVO_AGENT_IDS can be used to limit to specific roles (legacy behavior).
-    Set EVO_AGENT_LEGACY=1 to disable EvoAgent entirely.
-
-    Supported roles:
-    - analyst roles (fundamentals, technical, sentiment, valuation)
-    - risk_manager
-    - portfolio_manager
-
-    Example:
-        EVO_AGENT_IDS=fundamentals_analyst,risk_manager,portfolio_manager
-    """
-    from backend.config.constants import ANALYST_TYPES
-
-    all_supported = set(ANALYST_TYPES) | {"risk_manager", "portfolio_manager"}
-
-    raw = os.getenv("EVO_AGENT_IDS", "")
-    if not raw.strip():
-        # Default: all supported roles use EvoAgent
-        return all_supported
-
-    if raw.strip().lower() in ("legacy", "old", "none"):
-        return set()
-
-    requested = {
-        item.strip()
-        for item in raw.split(",")
-        if item.strip()
-    }
-    return {
-        agent_id
-        for agent_id in requested
-        if agent_id in ANALYST_TYPES or agent_id in {"risk_manager", "portfolio_manager"}
-    }
-
-
-def _create_analyst_agent(
-    *,
-    analyst_type: str,
-    config_name: str,
-    model,
-    formatter,
-    skills_manager: SkillsManager,
-    active_skill_map: dict[str, list[Path]],
-    long_term_memory=None,
-):
-    """Create one analyst agent, optionally using EvoAgent."""
-    active_skill_dirs = active_skill_map.get(analyst_type, [])
-    toolkit = create_agent_toolkit(
-        analyst_type,
-        config_name,
-        active_skill_dirs=active_skill_dirs,
-    )
-
-    workspace_dir = skills_manager.get_agent_asset_dir(config_name, analyst_type)
-    agent_config = load_agent_workspace_config(workspace_dir / "agent.yaml")
-    agent = EvoAgent(
-        agent_id=analyst_type,
-        config_name=config_name,
-        workspace_dir=workspace_dir,
-        model=model,
-        formatter=formatter,
-        skills_manager=skills_manager,
-        prompt_files=agent_config.prompt_files,
-        long_term_memory=long_term_memory,
-    )
-    agent.toolkit = toolkit
-    setattr(agent, "run_id", config_name)
-    # Keep workspace_id for backward compatibility
-    setattr(agent, "workspace_id", config_name)
-    return agent
-
-
-def _create_risk_manager_agent(
-    *,
-    config_name: str,
-    model,
-    formatter,
-    skills_manager: SkillsManager,
-    active_skill_map: dict[str, list[Path]],
-    long_term_memory=None,
-):
-    """Create the risk manager, optionally using EvoAgent."""
-    active_skill_dirs = active_skill_map.get("risk_manager", [])
-    toolkit = create_agent_toolkit(
-        "risk_manager",
-        config_name,
-        active_skill_dirs=active_skill_dirs,
-    )
-
-    use_evo_agent = "risk_manager" in _resolve_evo_agent_ids()
-
-    if use_evo_agent:
-        workspace_dir = skills_manager.get_agent_asset_dir(config_name, "risk_manager")
-        agent_config = load_agent_workspace_config(workspace_dir / "agent.yaml")
-        agent = EvoAgent(
-            agent_id="risk_manager",
-            config_name=config_name,
-            workspace_dir=workspace_dir,
-            model=model,
-            formatter=formatter,
-            skills_manager=skills_manager,
-            prompt_files=agent_config.prompt_files,
-            long_term_memory=long_term_memory,
-        )
-        agent.toolkit = toolkit
-        setattr(agent, "run_id", config_name)
-        # Keep workspace_id for backward compatibility
-        setattr(agent, "workspace_id", config_name)
-        return agent
-
-    return RiskAgent(
-        model=model,
-        formatter=formatter,
-        name="risk_manager",
-        config={"config_name": config_name},
-        long_term_memory=long_term_memory,
-        toolkit=toolkit,
-    )
-
-
-def _create_portfolio_manager_agent(
-    *,
-    config_name: str,
-    model,
-    formatter,
-    initial_cash: float,
-    margin_requirement: float,
-    skills_manager: SkillsManager,
-    active_skill_map: dict[str, list[Path]],
-    long_term_memory=None,
-):
-    """Create the portfolio manager, optionally using EvoAgent."""
-    active_skill_dirs = active_skill_map.get("portfolio_manager", [])
-    use_evo_agent = "portfolio_manager" in _resolve_evo_agent_ids()
-
-    if use_evo_agent:
-        workspace_dir = skills_manager.get_agent_asset_dir(
-            config_name,
-            "portfolio_manager",
-        )
-        agent_config = load_agent_workspace_config(workspace_dir / "agent.yaml")
-        agent = EvoAgent(
-            agent_id="portfolio_manager",
-            config_name=config_name,
-            workspace_dir=workspace_dir,
-            model=model,
-            formatter=formatter,
-            skills_manager=skills_manager,
-            prompt_files=agent_config.prompt_files,
-            initial_cash=initial_cash,
-            margin_requirement=margin_requirement,
-            long_term_memory=long_term_memory,
-        )
-        agent.toolkit = create_agent_toolkit(
-            "portfolio_manager",
-            config_name,
-            owner=agent,
-            active_skill_dirs=active_skill_dirs,
-        )
-        setattr(agent, "run_id", config_name)
-        # Keep workspace_id for backward compatibility
-        setattr(agent, "workspace_id", config_name)
-        return agent
-
-    return PMAgent(
-        name="portfolio_manager",
-        model=model,
-        formatter=formatter,
-        initial_cash=initial_cash,
-        margin_requirement=margin_requirement,
-        config={"config_name": config_name},
-        long_term_memory=long_term_memory,
-        toolkit_factory=create_agent_toolkit,
-        toolkit_factory_kwargs={
-            "active_skill_dirs": active_skill_dirs,
-        },
-    )
-
-
-def create_agents(
-    config_name: str,
-    initial_cash: float,
-    margin_requirement: float,
-    enable_long_term_memory: bool = False,
-):
-    """Create all agents for the system
-
-    Returns:
-        tuple: (analysts, risk_manager, portfolio_manager, long_term_memories)
-        long_term_memories is a list of memory
-    """
-    analysts = []
-    long_term_memories = []
-    workspace_manager = WorkspaceManager()
-    workspace_manager.initialize_default_assets(
-        config_name=config_name,
-        agent_ids=list(ANALYST_TYPES.keys())
-        + ["risk_manager", "portfolio_manager"],
-        analyst_personas=_prompt_loader.load_yaml_config("analyst", "personas"),
-    )
-    profiles = load_agent_profiles()
-    skills_manager = SkillsManager()
-    active_skill_map = skills_manager.prepare_active_skills(
-        config_name=config_name,
-        agent_defaults={
-            agent_id: profile.get("skills", [])
-            for agent_id, profile in profiles.items()
-        },
-    )
-
-    for analyst_type in ANALYST_TYPES:
-        model = get_agent_model(analyst_type)
-        formatter = get_agent_formatter(analyst_type)
-
-        long_term_memory = None
-        if enable_long_term_memory:
-            long_term_memory = create_long_term_memory(
-                analyst_type,
-                config_name,
-            )
-            if long_term_memory:
-                long_term_memories.append(long_term_memory)
-
-        analyst = _create_analyst_agent(
-            analyst_type=analyst_type,
-            config_name=config_name,
-            model=model,
-            formatter=formatter,
-            skills_manager=skills_manager,
-            active_skill_map=active_skill_map,
-            long_term_memory=long_term_memory,
-        )
-        analysts.append(analyst)
-
-    risk_long_term_memory = None
-    if enable_long_term_memory:
-        risk_long_term_memory = create_long_term_memory(
-            "risk_manager",
-            config_name,
-        )
-        if risk_long_term_memory:
-            long_term_memories.append(risk_long_term_memory)
-
-    risk_manager = _create_risk_manager_agent(
-        config_name=config_name,
-        model=get_agent_model("risk_manager"),
-        formatter=get_agent_formatter("risk_manager"),
-        skills_manager=skills_manager,
-        active_skill_map=active_skill_map,
-        long_term_memory=risk_long_term_memory,
-    )
-
-    pm_long_term_memory = None
-    if enable_long_term_memory:
-        pm_long_term_memory = create_long_term_memory(
-            "portfolio_manager",
-            config_name,
-        )
-        if pm_long_term_memory:
-            long_term_memories.append(pm_long_term_memory)
-
-    portfolio_manager = _create_portfolio_manager_agent(
-        config_name=config_name,
-        model=get_agent_model("portfolio_manager"),
-        formatter=get_agent_formatter("portfolio_manager"),
-        initial_cash=initial_cash,
-        margin_requirement=margin_requirement,
-        skills_manager=skills_manager,
-        active_skill_map=active_skill_map,
-        long_term_memory=pm_long_term_memory,
-    )
-
-    return analysts, risk_manager, portfolio_manager, long_term_memories
-async def run_with_gateway(args):
-    """Run with WebSocket gateway"""
-    is_backtest = args.mode == "backtest"
-    runtime_config = _resolve_runtime_config(args)
-
-    config_name = args.config_name
-    tickers = runtime_config["tickers"]
-    initial_cash = runtime_config["initial_cash"]
-    margin_requirement = runtime_config["margin_requirement"]
-
-    runtime_manager = TradingRuntimeManager(
-        config_name=config_name,
-        run_dir=_get_run_dir(config_name),
-        bootstrap=runtime_config,
-    )
-    runtime_manager.prepare_run()
-    set_global_runtime_manager(runtime_manager)
-
-    # Create market service
-    market_service = MarketService(
-        tickers=tickers,
-        poll_interval=args.poll_interval,
-        backtest_mode=is_backtest,
-        api_key=os.getenv("FINNHUB_API_KEY") if not is_backtest else None,
-        backtest_start_date=args.start_date if is_backtest else None,
-        backtest_end_date=args.end_date if is_backtest else None,
-    )
-
-    # Create storage service
-    storage_service = StorageService(
-        dashboard_dir=_get_run_dir(config_name) / "team_dashboard",
-        initial_cash=initial_cash,
-        config_name=config_name,
-    )
-
-    if not storage_service.files["summary"].exists():
-        storage_service.initialize_empty_dashboard()
-    else:
-        storage_service.update_leaderboard_model_info()
-
-    # Create agents and pipeline
-    analysts, risk_manager, pm, long_term_memories = create_agents(
-        config_name=config_name,
-        initial_cash=initial_cash,
-        margin_requirement=margin_requirement,
-        enable_long_term_memory=runtime_config["enable_memory"],
-    )
-    for agent in analysts + [risk_manager, pm]:
-        agent_id = getattr(agent, "agent_id", None) or getattr(agent, "name", None)
-        if agent_id:
-            runtime_manager.register_agent(agent_id)
-    portfolio_state = storage_service.load_portfolio_state()
-    pm.load_portfolio_state(portfolio_state)
-
-    settlement_coordinator = SettlementCoordinator(
-        storage=storage_service,
-        initial_capital=initial_cash,
-    )
-
-    pipeline = TradingPipeline(
-        analysts=analysts,
-        risk_manager=risk_manager,
-        portfolio_manager=pm,
-        settlement_coordinator=settlement_coordinator,
-        max_comm_cycles=runtime_config["max_comm_cycles"],
-        runtime_manager=runtime_manager,
-    )
-
-    # Create scheduler callback
-    scheduler_callback = None
-    trading_dates = []
-    live_scheduler = None
-
-    if is_backtest:
-        backtest_scheduler = BacktestScheduler(
-            start_date=args.start_date,
-            end_date=args.end_date,
-            trading_calendar="NYSE",
-            delay_between_days=0.5,
-        )
-        trading_dates = backtest_scheduler.get_trading_dates()
-
-        async def scheduler_callback_fn(callback):
-            await backtest_scheduler.start(callback)
-
-        scheduler_callback = scheduler_callback_fn
-    else:
-        # Live mode: use daily or intraday scheduler with NYSE timezone
-        live_scheduler = Scheduler(
-            mode=runtime_config["schedule_mode"],
-            trigger_time=runtime_config["trigger_time"],
-            interval_minutes=runtime_config["interval_minutes"],
-            config={"config_name": config_name},
-        )
-
-        async def scheduler_callback_fn(callback):
-            await live_scheduler.start(callback)
-
-        scheduler_callback = scheduler_callback_fn
-
-    # Create gateway
-    gateway = Gateway(
-        market_service=market_service,
-        storage_service=storage_service,
-        pipeline=pipeline,
-        scheduler_callback=scheduler_callback,
-        config={
-            "mode": args.mode,
-            "backtest_mode": is_backtest,
-            "tickers": tickers,
-            "config_name": config_name,
-            "schedule_mode": runtime_config["schedule_mode"],
-            "interval_minutes": runtime_config["interval_minutes"],
-            "trigger_time": runtime_config["trigger_time"],
-            "initial_cash": initial_cash,
-            "margin_requirement": margin_requirement,
-            "max_comm_cycles": runtime_config["max_comm_cycles"],
-            "enable_memory": runtime_config["enable_memory"],
-        },
-        scheduler=live_scheduler if not is_backtest else None,
-    )
-
-    if is_backtest:
-        gateway.set_backtest_dates(trading_dates)
-
-    # Start long-term memory contexts and run gateway
-    async with AsyncExitStack() as stack:
-        try:
-            for memory in long_term_memories:
-                await stack.enter_async_context(memory)
-            await gateway.start(host=args.host, port=args.port)
-        finally:
-            # Persist long-term memories before cleanup
-            for memory in long_term_memories:
-                try:
-                    if hasattr(memory, 'save') and callable(getattr(memory, 'save')):
-                        await memory.save()
-                except Exception as e:
-                    logger.warning(f"Failed to persist memory: {e}")
-            unregister_runtime_manager()
-            clear_global_runtime_manager()
-
-
-def build_arg_parser() -> argparse.ArgumentParser:
-    """Build the CLI parser for the gateway runtime entrypoint."""
-    parser = argparse.ArgumentParser(description="Trading System")
-    parser.add_argument("--mode", choices=["live", "backtest"], default="live")
-    parser.add_argument(
-        "--config-name",
-        default="default_run",
-        help=(
-            "Run label under runs/<config_name>; not a special root-level "
-            "live/backtest/production directory."
-        ),
-    )
-    parser.add_argument("--host", default="0.0.0.0")
-    parser.add_argument("--port", type=int, default=8765)
-    parser.add_argument(
-        "--schedule-mode",
-        choices=["daily", "intraday"],
-        default="daily",
-    )
-    parser.add_argument("--trigger-time", default="09:30")  # NYSE market open
-    parser.add_argument("--interval-minutes", type=int, default=60)
-    parser.add_argument("--poll-interval", type=int, default=10)
-    parser.add_argument("--start-date")
-    parser.add_argument("--end-date")
-    parser.add_argument(
-        "--enable-memory",
-        action="store_true",
-        help="Enable ReMeTaskLongTermMemory for agents",
-    )
-    return parser
-
-
-def main():
-    """Main entry point"""
-    parser = build_arg_parser()
-
-    args = parser.parse_args()
-
-    # Load config from env for logging
-    runtime_config = _resolve_runtime_config(args)
-    tickers = runtime_config["tickers"]
-    initial_cash = runtime_config["initial_cash"]
-
-    logger.info("=" * 60)
-    logger.info(f"Mode: {args.mode}, Config: {args.config_name}")
-    logger.info(f"Tickers: {tickers}")
-    logger.info(f"Initial Cash: ${initial_cash:,.2f}")
-    logger.info(
-        "Long-term Memory: %s",
-        "enabled" if runtime_config["enable_memory"] else "disabled",
-    )
-    if args.mode == "backtest":
-        if not args.start_date or not args.end_date:
-            parser.error(
-                "--start-date and --end-date required for backtest mode",
-            )
-        logger.info(f"Backtest: {args.start_date} to {args.end_date}")
-    logger.info("=" * 60)
-
-    asyncio.run(run_with_gateway(args))
-
-
-if __name__ == "__main__":
-    main()

backend/services/gateway_admin_handlers.py

@@ -25,6 +25,13 @@ from backend.config.bootstrap_config import (
 from backend.llm.models import get_agent_model_info
 
 
+def _normalize_schedule_mode(value: Any) -> str:
+    mode = str(value or "daily").strip().lower()
+    if mode == "intraday":
+        return "interval"
+    return mode or "daily"
+
+
 async def handle_reload_runtime_assets(gateway: Any) -> None:
     config_name = gateway.config.get("config_name", "default")
     runtime_config = resolve_runtime_config(
@@ -44,10 +51,10 @@ async def handle_reload_runtime_assets(gateway: Any) -> None:
 async def handle_update_runtime_config(gateway: Any, websocket: Any, data: dict[str, Any]) -> None:
     updates: dict[str, Any] = {}
 
-    schedule_mode = str(data.get("schedule_mode", "")).strip().lower()
+    schedule_mode = _normalize_schedule_mode(data.get("schedule_mode", ""))
     if schedule_mode:
-        if schedule_mode not in {"daily", "intraday"}:
-            await websocket.send(json.dumps({"type": "error", "message": "schedule_mode must be 'daily' or 'intraday'."}, ensure_ascii=False))
+        if schedule_mode not in {"daily", "interval"}:
+            await websocket.send(json.dumps({"type": "error", "message": "schedule_mode must be 'daily' or 'interval'."}, ensure_ascii=False))
             return
         updates["schedule_mode"] = schedule_mode
 

backend/services/gateway_cycle_support.py

@@ -208,7 +208,7 @@ async def run_live_cycle(gateway: Any, date: str, tickers: list[str]) -> None:
     market_status = gateway.market_service.get_market_status()
     current_prices = gateway.market_service.get_all_prices()
 
-    if schedule_mode == "intraday":
+    if schedule_mode in {"interval", "intraday"}:
         execute_decisions = market_status.get("status") == "open"
         if execute_decisions:
             await gateway.state_sync.on_system_message("定时任务触发：当前处于交易时段，本轮将执行交易决策")

backend/services/gateway_runtime_support.py

@@ -8,6 +8,13 @@ from typing import Any
 from backend.data.provider_utils import normalize_symbol
 
 
+def _normalize_schedule_mode(value: Any) -> str:
+    mode = str(value or "daily").strip().lower()
+    if mode == "intraday":
+        return "interval"
+    return mode or "daily"
+
+
 def normalize_watchlist(raw_tickers: Any) -> list[str]:
     """Parse watchlist payloads from websocket messages."""
     if raw_tickers is None:
@@ -51,9 +58,11 @@ def apply_runtime_config(gateway: Any, runtime_config: dict[str, Any]) -> dict[s
 
     gateway.pipeline.max_comm_cycles = int(runtime_config["max_comm_cycles"])
     gateway.config["max_comm_cycles"] = gateway.pipeline.max_comm_cycles
-    gateway.config["schedule_mode"] = runtime_config.get(
-        "schedule_mode",
-        gateway.config.get("schedule_mode", "daily"),
+    gateway.config["schedule_mode"] = _normalize_schedule_mode(
+        runtime_config.get(
+            "schedule_mode",
+            gateway.config.get("schedule_mode", "daily"),
+        ),
     )
     gateway.config["interval_minutes"] = int(
         runtime_config.get(

backend/skills/builtin/dynamic_team_management/SKILL.md

@@ -0,0 +1,189 @@
+---
+name: dynamic_team_management
+description: 动态管理团队中的分析师Agent，包括创建、克隆、移除分析师，以及查看可用分析师类型。
+version: 1.0.0
+tools:
+  - create_analyst
+  - clone_analyst
+  - remove_analyst
+  - list_analyst_types
+  - get_analyst_info
+  - get_team_summary
+---
+
+# 动态团队管理
+
+当你需要调整分析师团队组成时，使用这个技能。投资经理可以动态创建新的分析师、克隆现有分析师进行定制、或移除不再需要的分析师。
+
+## 1) When to use
+
+- 当前团队缺乏特定领域的分析能力（如期权、加密货币、ESG等）
+- 需要多个相同类型但不同配置的分析师（如"激进型技术分析师"和"保守型技术分析师"）
+- 需要临时增加分析力量应对特殊市场环境
+- 发现某个分析师配置不当，需要移除并重建
+- 在团队讨论中发现需要新的分析视角
+
+## 2) Required inputs
+
+### 创建分析师 (create_analyst)
+- **agent_id**: 唯一标识符（如 "options_specialist_01"）
+- **analyst_type**: 基础类型（如 "technical_analyst"）或自定义标识
+- **可选**: name, focus, description, soul_md, agents_md, model_name
+
+### 克隆分析师 (clone_analyst)
+- **source_id**: 源分析师ID（如 "technical_analyst"）
+- **new_id**: 新分析师ID（如 "crypto_technical_01"）
+- **可选**: name, focus_additions, description_override, model_name
+
+### 移除分析师 (remove_analyst)
+- **agent_id**: 要移除的分析师ID
+
+## 3) Decision procedure
+
+1. **评估当前团队能力缺口**
+   - 查看当前活跃的分析师列表
+   - 识别缺失的分析视角或专业领域
+
+2. **选择创建策略**
+   - 基于现有类型创建：指定analyst_type，提供自定义配置
+   - 完全自定义：提供完整的persona定义
+   - 克隆并修改：从现有分析师复制并应用覆盖
+
+3. **配置分析师**
+   - 设置唯一agent_id
+   - 定义显示名称和关注点
+   - 可选：提供自定义SOUL.md内容以精确定义行为
+
+4. **验证创建结果**
+   - 检查返回的success状态
+   - 确认新分析师已加入活跃列表
+
+## 4) Tool call policy
+
+- **create_analyst**: 用于创建全新的分析师实例
+  - 必须提供唯一的agent_id
+  - 基于预定义类型时，analyst_type必须在可用类型列表中，或提供完整自定义配置
+  - 工具调用失败时，检查agent_id是否已存在
+
+- **clone_analyst**: 用于基于现有分析师创建变体
+  - 适用于：创建专注于特定行业的分析师（如从technical_analyst克隆crypto_technical）
+  - 新实例继承源配置，应用指定的覆盖
+
+- **remove_analyst**: 用于移除动态创建的分析师
+  - 只能移除通过本技能创建的分析师
+  - 系统预定义分析师（fundamentals_analyst等）不可移除
+
+- **list_analyst_types**: 用于查看所有可用分析师类型
+  - 返回预定义类型 + 运行时注册类型
+
+- **get_analyst_info**: 用于查看特定分析师的详细配置
+
+- **get_team_summary**: 用于查看团队整体构成
+
+## 5) Output schema
+
+### create_analyst / clone_analyst 输出
+```json
+{
+  "success": true,
+  "agent_id": "options_specialist_01",
+  "message": "Created runtime analyst 'options_specialist_01' (technical_analyst).",
+  "analyst_type": "technical_analyst"
+}
+```
+
+### remove_analyst 输出
+```json
+{
+  "success": true,
+  "agent_id": "options_specialist_01",
+  "message": "Removed runtime analyst 'options_specialist_01'."
+}
+```
+
+### list_analyst_types 输出
+```json
+[
+  {
+    "type_id": "fundamentals_analyst",
+    "name": "Fundamentals Analyst",
+    "description": "...",
+    "is_builtin": true,
+    "source": "constants"
+  }
+]
+```
+
+## 6) Failure fallback
+
+- **agent_id已存在**: 返回错误，提示选择新的agent_id或使用clone_analyst基于现有创建变体
+- **analyst_type未知**: 提示使用list_analyst_types查看可用类型，或提供完整的自定义persona
+- **创建失败**: 检查系统日志，可能原因包括：模型配置错误、工作空间权限问题
+- **移除失败**: 确认分析师是通过动态创建（系统预定义分析师不可移除）
+
+## 重要约定
+
+### Agent ID 命名规则
+
+为了使新创建的分析师能够正常工作，**agent_id 必须以 `_analyst` 结尾**。这是系统识别分析师类型并分配相应工具的关键约定。
+
+- ✅ **正确**: `options_specialist_analyst`, `crypto_technical_analyst`
+- ❌ **错误**: `options_specialist`, `crypto_expert`
+
+如果不遵循此约定，分析师将无法获得分析工具组（基本面、技术、情绪、估值等工具）。
+
+### 全新自定义类型 vs 基于现有类型
+
+**基于现有类型**（推荐用于快速创建）：
+- 使用 `analyst_type: "technical_analyst"` 等预定义类型
+- 可以覆盖 persona、SOUL.md 等配置
+- 工具组根据 `analyst_type` 自动选择
+
+**全新自定义类型**（用于完全自定义）：
+- 设置 `analyst_type` 为自定义标识（如 `"custom"`）或任意字符串
+- 必须提供完整的 `persona` 定义
+- 建议提供 `soul_md` 精确定义行为
+- **agent_id 必须仍然以 `_analyst` 结尾**
+
+## 最佳实践
+
+1. **命名约定**: 使用描述性agent_id，如 `industry_tech_analyst` 而非 `analyst_01`，**必须以 `_analyst` 结尾**
+2. **版本控制**: 克隆分析师时，在new_id中包含版本信息，如 `technical_v2_crypto_analyst`
+3. **文档记录**: 创建自定义分析师时，提供详细的description，便于后续理解和维护
+4. **资源管理**: 定期使用get_team_summary检查团队规模，移除不再需要的分析师
+
+## 示例场景
+
+### 场景1: 添加加密货币分析师
+```
+创建一个新的分析师，专注于加密货币技术分析：
+- agent_id: "crypto_technical_01"
+- analyst_type: "technical_analyst"
+- name: "加密货币技术分析师"
+- focus: ["链上数据分析", "DeFi协议", "加密货币技术指标"]
+```
+
+### 场景2: 克隆并定制
+```
+基于technical_analyst创建一个更激进的版本：
+- source_id: "technical_analyst"
+- new_id: "technical_aggressive_01"
+- name: "激进技术分析师"
+- focus_additions: ["高波动交易", "突破策略"]
+- description_override: "专注于高风险高回报的技术策略..."
+```
+
+### 场景3: 创建全新自定义类型（期权专家）
+```
+创建一个完全自定义的期权分析师（注意agent_id以_analyst结尾）：
+- agent_id: "options_strategist_analyst"
+- analyst_type: "custom"  # 使用非预定义类型
+- name: "期权策略分析师"
+- focus: ["期权定价", "希腊字母", "波动率曲面"]
+- soul_md: "# 角色定义\n你是期权策略专家，专注于..."
+```
+
+**说明**: 
+- 即使 `analyst_type` 是 "custom"（不在预定义类型中），只要提供完整的 `persona` 和 `soul_md`，系统就能创建功能完整的分析师
+- `agent_id` 必须以 `_analyst` 结尾才能获得分析工具
+- 模型使用全局默认，或通过 `model_name` 参数指定

backend/tests/test_agent_service_app.py

@@ -5,7 +5,7 @@
 from fastapi.testclient import TestClient
 
 from backend.apps.agent_service import create_app
-from backend.api import agents as agents_module
+from backend.api import runs as runs_module
 
 
 def test_agent_service_routes_include_control_plane_endpoints(tmp_path):
@@ -73,10 +73,10 @@ def test_agent_service_read_routes(monkeypatch, tmp_path):
         def load_agent_file(self, config_name, agent_id, filename):
             return f"{config_name}:{agent_id}:{filename}"
 
-    monkeypatch.setattr(agents_module, "load_agent_profiles", lambda: {"portfolio_manager": {"skills": ["demo_skill"]}})
-    monkeypatch.setattr(agents_module, "get_agent_model_info", lambda agent_id: ("deepseek-v3.2", "DASHSCOPE"))
+    monkeypatch.setattr(runs_module, "load_agent_profiles", lambda: {"portfolio_manager": {"skills": ["demo_skill"]}})
+    monkeypatch.setattr(runs_module, "get_agent_model_info", lambda agent_id: ("deepseek-v3.2", "DASHSCOPE"))
     monkeypatch.setattr(
-        agents_module,
+        runs_module,
         "load_agent_workspace_config",
         lambda path: type(
             "Cfg",
@@ -91,20 +91,20 @@ def test_agent_service_read_routes(monkeypatch, tmp_path):
         )(),
     )
     monkeypatch.setattr(
-        agents_module,
+        runs_module,
         "get_bootstrap_config_for_run",
         lambda project_root, config_name: type("Bootstrap", (), {"agent_override": lambda self, agent_id: {}})(),
     )
 
     app = create_app(project_root=tmp_path)
-    app.dependency_overrides[agents_module.get_skills_manager] = lambda: _FakeSkillsManager()
-    app.dependency_overrides[agents_module.get_workspace_manager] = lambda: _FakeWorkspaceManager()
+    app.dependency_overrides[runs_module.get_skills_manager] = lambda: _FakeSkillsManager()
+    app.dependency_overrides[runs_module.get_workspace_manager] = lambda: _FakeWorkspaceManager()
 
     with TestClient(app) as client:
-        profile = client.get("/api/workspaces/demo/agents/portfolio_manager/profile")
-        skills = client.get("/api/workspaces/demo/agents/portfolio_manager/skills")
-        detail = client.get("/api/workspaces/demo/agents/portfolio_manager/skills/demo_skill")
-        workspace_file = client.get("/api/workspaces/demo/agents/portfolio_manager/files/MEMORY.md")
+        profile = client.get("/api/runs/demo/agents/portfolio_manager/profile")
+        skills = client.get("/api/runs/demo/agents/portfolio_manager/skills")
+        detail = client.get("/api/runs/demo/agents/portfolio_manager/skills/demo_skill")
+        workspace_file = client.get("/api/runs/demo/agents/portfolio_manager/files/MEMORY.md")
 
     assert profile.status_code == 200
     assert profile.json()["profile"]["model_name"] == "deepseek-v3.2"
@@ -118,4 +118,3 @@ def test_agent_service_read_routes(monkeypatch, tmp_path):
     assert workspace_file.status_code == 200
     assert workspace_file.json()["content"] == "demo:portfolio_manager:MEMORY.md"
     assert workspace_file.json()["scope_type"] == "runtime_run"
-    assert "runs/<run_id>" in workspace_file.json()["scope_note"]

backend/tests/test_evo_agent_selection.py

@@ -6,14 +6,14 @@ from pathlib import Path
 
 
 def test_main_resolve_evo_agent_ids_filters_unsupported_roles(monkeypatch):
-    from backend import main as main_module
+    from backend.core import pipeline_runner as runner_module
 
     monkeypatch.setenv(
         "EVO_AGENT_IDS",
         "fundamentals_analyst,portfolio_manager,unknown,technical_analyst",
     )
 
-    resolved = main_module._resolve_evo_agent_ids()
+    resolved = runner_module._resolve_evo_agent_ids()
 
     assert resolved == {"fundamentals_analyst", "portfolio_manager", "technical_analyst"}
 
@@ -29,7 +29,7 @@ def test_pipeline_runner_resolve_evo_agent_ids_keeps_supported_roles(monkeypatch
 
 
 def test_main_create_analyst_agent_can_build_evo_agent(monkeypatch, tmp_path):
-    from backend import main as main_module
+    from backend.core import pipeline_runner as runner_module
 
     created = {}
 
@@ -49,12 +49,12 @@ def test_main_create_analyst_agent_can_build_evo_agent(monkeypatch, tmp_path):
             self.toolkit = None
 
     monkeypatch.setenv("EVO_AGENT_IDS", "fundamentals_analyst")
-    monkeypatch.setattr(main_module, "EvoAgent", DummyEvoAgent)
-    monkeypatch.setattr(main_module, "create_agent_toolkit", lambda *args, **kwargs: "toolkit")
+    monkeypatch.setattr(runner_module, "EvoAgent", DummyEvoAgent)
+    monkeypatch.setattr(runner_module, "create_agent_toolkit", lambda *args, **kwargs: "toolkit")
 
-    agent = main_module._create_analyst_agent(
+    agent = runner_module._create_analyst_agent(
         analyst_type="fundamentals_analyst",
-        config_name="demo",
+        run_id="demo",
         model="model",
         formatter="formatter",
         skills_manager=DummySkillsManager(),
@@ -71,7 +71,7 @@ def test_main_create_analyst_agent_can_build_evo_agent(monkeypatch, tmp_path):
 
 
 def test_main_create_risk_manager_can_build_evo_agent(monkeypatch, tmp_path):
-    from backend import main as main_module
+    from backend.core import pipeline_runner as runner_module
 
     created = {}
 
@@ -91,11 +91,11 @@ def test_main_create_risk_manager_can_build_evo_agent(monkeypatch, tmp_path):
             self.toolkit = None
 
     monkeypatch.setenv("EVO_AGENT_IDS", "risk_manager")
-    monkeypatch.setattr(main_module, "EvoAgent", DummyEvoAgent)
-    monkeypatch.setattr(main_module, "create_agent_toolkit", lambda *args, **kwargs: "risk-toolkit")
+    monkeypatch.setattr(runner_module, "EvoAgent", DummyEvoAgent)
+    monkeypatch.setattr(runner_module, "create_agent_toolkit", lambda *args, **kwargs: "risk-toolkit")
 
-    agent = main_module._create_risk_manager_agent(
-        config_name="demo",
+    agent = runner_module._create_risk_manager_agent(
+        run_id="demo",
         model="model",
         formatter="formatter",
         skills_manager=DummySkillsManager(),
@@ -112,7 +112,7 @@ def test_main_create_risk_manager_can_build_evo_agent(monkeypatch, tmp_path):
 
 
 def test_main_create_portfolio_manager_can_build_evo_agent(monkeypatch, tmp_path):
-    from backend import main as main_module
+    from backend.core import pipeline_runner as runner_module
 
     created = {}
 
@@ -132,15 +132,15 @@ def test_main_create_portfolio_manager_can_build_evo_agent(monkeypatch, tmp_path
             self.toolkit = None
 
     monkeypatch.setenv("EVO_AGENT_IDS", "portfolio_manager")
-    monkeypatch.setattr(main_module, "EvoAgent", DummyEvoAgent)
+    monkeypatch.setattr(runner_module, "EvoAgent", DummyEvoAgent)
     monkeypatch.setattr(
-        main_module,
+        runner_module,
         "create_agent_toolkit",
         lambda *args, **kwargs: "pm-toolkit",
     )
 
-    agent = main_module._create_portfolio_manager_agent(
-        config_name="demo",
+    agent = runner_module._create_portfolio_manager_agent(
+        run_id="demo",
         model="model",
         formatter="formatter",
         initial_cash=12345.0,
@@ -372,13 +372,13 @@ def test_pipeline_create_runtime_analyst_uses_legacy_when_not_in_evo_ids(monkeyp
 
 def test_main_resolve_evo_agent_ids_returns_all_by_default(monkeypatch):
     """Test that _resolve_evo_agent_ids returns all supported roles by default."""
-    from backend import main as main_module
+    from backend.core import pipeline_runner as runner_module
     from backend.config.constants import ANALYST_TYPES
 
     # Unset EVO_AGENT_IDS to test default behavior
     monkeypatch.delenv("EVO_AGENT_IDS", raising=False)
 
-    resolved = main_module._resolve_evo_agent_ids()
+    resolved = runner_module._resolve_evo_agent_ids()
 
     expected = set(ANALYST_TYPES) | {"risk_manager", "portfolio_manager"}
     assert resolved == expected

backend/tests/test_gateway_support_modules.py

@@ -159,11 +159,11 @@ def test_apply_runtime_config_updates_gateway_state():
     )
 
     assert gateway.config["tickers"] == ["MSFT", "NVDA"]
-    assert gateway.config["schedule_mode"] == "intraday"
+    assert gateway.config["schedule_mode"] == "interval"
     assert gateway.storage.initial_cash == 150000.0
     assert result["runtime_config_applied"]["max_comm_cycles"] == 4
     assert gateway.scheduler.calls[-1] == {
-        "mode": "intraday",
+        "mode": "interval",
         "trigger_time": "10:30",
         "interval_minutes": 30,
     }

backend/tests/test_runtime_service_app.py

@@ -86,7 +86,7 @@ def test_runtime_service_get_runtime_config(monkeypatch, tmp_path):
         "---\n"
         "tickers:\n"
         "  - AAPL\n"
-        "schedule_mode: intraday\n"
+        "schedule_mode: interval\n"
         "interval_minutes: 30\n"
         "trigger_time: '10:00'\n"
         "max_comm_cycles: 3\n"
@@ -102,7 +102,7 @@ def test_runtime_service_get_runtime_config(monkeypatch, tmp_path):
                     "run_dir": str(run_dir),
                     "bootstrap_values": {
                         "tickers": ["AAPL"],
-                        "schedule_mode": "intraday",
+                        "schedule_mode": "interval",
                         "interval_minutes": 30,
                         "trigger_time": "10:00",
                         "max_comm_cycles": 3,
@@ -123,7 +123,7 @@ def test_runtime_service_get_runtime_config(monkeypatch, tmp_path):
     assert response.status_code == 200
     payload = response.json()
     assert payload["run_id"] == "demo"
-    assert payload["bootstrap"]["schedule_mode"] == "intraday"
+    assert payload["bootstrap"]["schedule_mode"] == "interval"
     assert payload["resolved"]["interval_minutes"] == 30
     assert payload["resolved"]["enable_memory"] is True
 
@@ -190,7 +190,7 @@ def test_runtime_service_update_runtime_config_persists_bootstrap(monkeypatch, t
         response = client.put(
             "/api/runtime/config",
             json={
-                "schedule_mode": "intraday",
+                "schedule_mode": "interval",
                 "interval_minutes": 15,
                 "trigger_time": "10:15",
                 "max_comm_cycles": 4,
@@ -199,7 +199,7 @@ def test_runtime_service_update_runtime_config_persists_bootstrap(monkeypatch, t
 
     assert response.status_code == 200
     payload = response.json()
-    assert payload["bootstrap"]["schedule_mode"] == "intraday"
+    assert payload["bootstrap"]["schedule_mode"] == "interval"
     assert payload["resolved"]["interval_minutes"] == 15
     assert "interval_minutes: 15" in (run_dir / "BOOTSTRAP.md").read_text(encoding="utf-8")
 
@@ -547,7 +547,7 @@ def test_start_runtime_restore_reuses_historical_run_id(monkeypatch, tmp_path):
                     "run_dir": str(run_dir),
                     "bootstrap_values": {
                         "tickers": ["AAPL"],
-                        "schedule_mode": "intraday",
+                        "schedule_mode": "interval",
                         "interval_minutes": 30,
                         "trigger_time": "now",
                         "max_comm_cycles": 2,

backend/tools/dynamic_team_tools.py

@@ -0,0 +1,518 @@
+# -*- coding: utf-8 -*-
+"""Dynamic Team Management Tools - Tools for PM to manage analyst team dynamically.
+
+This module provides tools for the Portfolio Manager to:
+- Create new analysts with custom configuration
+- Clone existing analysts with variations
+- Remove analysts from the team
+- List available analyst types
+- Get analyst information
+
+These tools are registered with the PM's toolkit and enable dynamic team management
+as described in the Dynamic Team Architecture.
+"""
+from __future__ import annotations
+
+from typing import Any, Dict, List, Optional, Callable
+from dataclasses import asdict
+
+from backend.agents.dynamic_team_types import (
+    AnalystPersona,
+    AnalystConfig,
+    CreateAnalystResult,
+    AnalystTypeInfo,
+)
+from backend.config.constants import ANALYST_TYPES
+
+
+# Type alias for callbacks set by pipeline
+CreateAnalystCallback = Callable[[str, str, Optional[AnalystConfig]], str]
+RemoveAnalystCallback = Callable[[str], str]
+
+
+class DynamicTeamController:
+    """Controller for dynamic analyst team management.
+
+    This class is instantiated by TradingPipeline and injected into the PM agent
+    via set_team_controller(). It provides methods that the PM can call through
+    tools to manage the analyst team dynamically.
+
+    Attributes:
+        create_callback: Callback to _create_runtime_analyst in pipeline
+        remove_callback: Callback to _remove_runtime_analyst in pipeline
+        get_analysts_callback: Callback to get current analysts list
+        registered_types: Runtime-registered custom analyst types
+    """
+
+    def __init__(
+        self,
+        create_callback: CreateAnalystCallback,
+        remove_callback: RemoveAnalystCallback,
+        get_analysts_callback: Optional[Callable[[], List[Any]]] = None,
+    ):
+        """Initialize the controller with callbacks from pipeline.
+
+        Args:
+            create_callback: Function to create a runtime analyst
+            remove_callback: Function to remove a runtime analyst
+            get_analysts_callback: Optional function to get current analysts
+        """
+        self._create_callback = create_callback
+        self._remove_callback = remove_callback
+        self._get_analysts_callback = get_analysts_callback
+        self._registered_types: Dict[str, AnalystPersona] = {}
+        self._instance_configs: Dict[str, AnalystConfig] = {}
+
+    def create_analyst(
+        self,
+        agent_id: str,
+        analyst_type: str,
+        name: Optional[str] = None,
+        focus: Optional[List[str]] = None,
+        description: Optional[str] = None,
+        soul_md: Optional[str] = None,
+        agents_md: Optional[str] = None,
+        model_name: Optional[str] = None,
+        preferred_tools: Optional[List[str]] = None,
+    ) -> Dict[str, Any]:
+        """Create a new analyst with optional custom configuration.
+
+        This tool allows the Portfolio Manager to dynamically create new analysts
+        during a trading session. The analyst can be based on a predefined type
+        or fully customized with a unique persona.
+
+        Args:
+            agent_id: Unique identifier for the new analyst (e.g., "crypto_specialist_01")
+            analyst_type: Base type (e.g., "technical_analyst") or custom identifier
+            name: Display name for the analyst (overrides default)
+            focus: List of focus areas (overrides default)
+            description: Detailed description (overrides default)
+            soul_md: Custom SOUL.md content for the analyst's workspace
+            agents_md: Custom AGENTS.md content
+            model_name: Override the default LLM model
+            preferred_tools: List of preferred tool categories
+
+        Returns:
+            Dict with success status, message, and analyst info
+
+        Example:
+            >>> result = create_analyst(
+            ...     agent_id="options_specialist",
+            ...     analyst_type="technical_analyst",
+            ...     name="期权策略分析师",
+            ...     focus=["期权定价", "波动率交易"],
+            ...     description="专注于期权市场分析和波动率交易策略...",
+            ... )
+        """
+        # Build custom config if any customization is provided
+        custom_config = None
+        if any([name, focus, description, soul_md, agents_md, model_name, preferred_tools]):
+            persona = None
+            if name or focus or description:
+                persona = AnalystPersona(
+                    name=name or f"Custom {analyst_type}",
+                    focus=focus or ["General Analysis"],
+                    description=description or f"Custom analyst based on {analyst_type}",
+                    preferred_tools=preferred_tools,
+                )
+
+            custom_config = AnalystConfig(
+                persona=persona,
+                analyst_type=analyst_type if analyst_type in ANALYST_TYPES else None,
+                soul_md=soul_md,
+                agents_md=agents_md,
+                model_name=model_name,
+            )
+
+        # Call the pipeline's create method
+        result_message = self._create_callback(agent_id, analyst_type, custom_config)
+
+        # Parse result
+        success = result_message.startswith("Created")
+        if success:
+            self._instance_configs[agent_id] = custom_config if custom_config else AnalystConfig(
+                analyst_type=analyst_type
+            )
+
+        return {
+            "success": success,
+            "agent_id": agent_id if success else None,
+            "message": result_message,
+            "analyst_type": analyst_type,
+        }
+
+    def clone_analyst(
+        self,
+        source_id: str,
+        new_id: str,
+        name: Optional[str] = None,
+        focus_additions: Optional[List[str]] = None,
+        description_override: Optional[str] = None,
+        model_name: Optional[str] = None,
+    ) -> Dict[str, Any]:
+        """Clone an existing analyst with optional modifications.
+
+        Creates a new analyst by copying the configuration of an existing one
+        and applying specified overrides. Useful for creating specialized
+        variants (e.g., "crypto_technical" from "technical_analyst").
+
+        Args:
+            source_id: ID of the analyst to clone
+            new_id: Unique identifier for the new analyst
+            name: New display name (if different from source)
+            focus_additions: Additional focus areas to add
+            description_override: Completely new description
+            model_name: Override the model from source
+
+        Returns:
+            Dict with success status, message, and new analyst info
+
+        Example:
+            >>> result = clone_analyst(
+            ...     source_id="technical_analyst",
+            ...     new_id="crypto_technical_01",
+            ...     name="加密货币技术分析师",
+            ...     focus_additions=["链上数据", "DeFi协议分析"],
+            ... )
+        """
+        # Get source config if available
+        source_config = self._instance_configs.get(source_id)
+
+        # Determine base type and config
+        if source_config:
+            base_type = source_config.analyst_type or source_id
+            base_persona = source_config.persona
+        else:
+            # Assume source_id is a known type
+            base_type = source_id
+            base_persona = None
+
+        # Build new persona
+        new_focus = list(base_persona.focus) if base_persona else []
+        if focus_additions:
+            new_focus.extend(focus_additions)
+
+        new_name = name or (base_persona.name if base_persona else new_id)
+        new_description = description_override or (base_persona.description if base_persona else "")
+
+        # Create new config with parent reference
+        new_config = AnalystConfig(
+            persona=AnalystPersona(
+                name=new_name,
+                focus=new_focus,
+                description=new_description,
+                preferred_tools=base_persona.preferred_tools if base_persona else None,
+            ),
+            analyst_type=base_type if base_type in ANALYST_TYPES else None,
+            soul_md=source_config.soul_md if source_config else None,
+            agents_md=source_config.agents_md if source_config else None,
+            model_name=model_name or (source_config.model_name if source_config else None),
+            parent_id=source_id,
+        )
+
+        # Create the new analyst
+        result_message = self._create_callback(new_id, base_type, new_config)
+
+        success = result_message.startswith("Created")
+        if success:
+            self._instance_configs[new_id] = new_config
+
+        return {
+            "success": success,
+            "agent_id": new_id if success else None,
+            "parent_id": source_id,
+            "message": result_message,
+        }
+
+    def remove_analyst(self, agent_id: str) -> Dict[str, Any]:
+        """Remove a dynamically created analyst from the team.
+
+        Args:
+            agent_id: ID of the analyst to remove
+
+        Returns:
+            Dict with success status and message
+
+        Example:
+            >>> result = remove_analyst("options_specialist")
+        """
+        result_message = self._remove_callback(agent_id)
+        success = result_message.startswith("Removed") or "not found" not in result_message.lower()
+
+        if success and agent_id in self._instance_configs:
+            del self._instance_configs[agent_id]
+
+        return {
+            "success": success,
+            "agent_id": agent_id,
+            "message": result_message,
+        }
+
+    def list_analyst_types(self) -> List[Dict[str, Any]]:
+        """List all available analyst types.
+
+        Returns a list of all available analyst types, including:
+        - Built-in types from ANALYST_TYPES
+        - Runtime registered custom types
+
+        Returns:
+            List of analyst type information dictionaries
+
+        Example:
+            >>> types = list_analyst_types()
+            >>> print(types[0]["type_id"])  # "fundamentals_analyst"
+        """
+        result = []
+
+        # Add built-in types
+        for type_id, info in ANALYST_TYPES.items():
+            result.append({
+                "type_id": type_id,
+                "name": info.get("display_name", type_id),
+                "description": info.get("description", ""),
+                "is_builtin": True,
+                "source": "constants",
+            })
+
+        # Add runtime registered types
+        for type_id, persona in self._registered_types.items():
+            result.append({
+                "type_id": type_id,
+                "name": persona.name,
+                "description": persona.description,
+                "is_builtin": False,
+                "source": "runtime",
+            })
+
+        return result
+
+    def get_analyst_info(self, agent_id: str) -> Dict[str, Any]:
+        """Get information about a specific analyst.
+
+        Args:
+            agent_id: ID of the analyst
+
+        Returns:
+            Dict with analyst configuration and status
+        """
+        config = self._instance_configs.get(agent_id)
+        if not config:
+            return {
+                "found": False,
+                "agent_id": agent_id,
+                "message": f"No configuration found for '{agent_id}'",
+            }
+
+        return {
+            "found": True,
+            "agent_id": agent_id,
+            "config": config.to_dict(),
+            "is_custom": config.persona is not None,
+            "is_clone": config.parent_id is not None,
+            "parent_id": config.parent_id,
+        }
+
+    def register_analyst_type(
+        self,
+        type_id: str,
+        name: str,
+        focus: List[str],
+        description: str,
+        preferred_tools: Optional[List[str]] = None,
+    ) -> Dict[str, Any]:
+        """Register a new analyst type for later creation.
+
+        This allows defining reusable analyst personas that can be instantiated
+        multiple times with different configurations.
+
+        Args:
+            type_id: Unique identifier for this type (e.g., "options_analyst")
+            name: Display name
+            focus: List of focus areas
+            description: Detailed description
+            preferred_tools: Optional list of preferred tool categories
+
+        Returns:
+            Dict with success status and type info
+
+        Example:
+            >>> result = register_analyst_type(
+            ...     type_id="options_analyst",
+            ...     name="期权分析师",
+            ...     focus=["期权定价", "希腊字母分析"],
+            ...     description="专注于期权策略和波动率分析",
+            ... )
+        """
+        if type_id in self._registered_types or type_id in ANALYST_TYPES:
+            return {
+                "success": False,
+                "type_id": type_id,
+                "message": f"Type '{type_id}' already exists",
+            }
+
+        persona = AnalystPersona(
+            name=name,
+            focus=focus,
+            description=description,
+            preferred_tools=preferred_tools,
+        )
+        self._registered_types[type_id] = persona
+
+        return {
+            "success": True,
+            "type_id": type_id,
+            "persona": persona.to_dict(),
+            "message": f"Registered new analyst type '{type_id}'",
+        }
+
+    def get_team_summary(self) -> Dict[str, Any]:
+        """Get a summary of the current analyst team.
+
+        Returns:
+            Dict with team composition information
+        """
+        analysts = []
+        for agent_id, config in self._instance_configs.items():
+            analysts.append({
+                "agent_id": agent_id,
+                "name": config.persona.name if config.persona else agent_id,
+                "type": config.analyst_type,
+                "is_custom": config.persona is not None,
+                "is_clone": config.parent_id is not None,
+            })
+
+        return {
+            "total_analysts": len(analysts),
+            "custom_analysts": len([a for a in analysts if a["is_custom"]]),
+            "cloned_analysts": len([a for a in analysts if a["is_clone"]]),
+            "analysts": analysts,
+            "registered_types": len(self._registered_types),
+        }
+
+
+# Global controller instance - set by pipeline
+_controller_instance: Optional[DynamicTeamController] = None
+
+
+def set_controller(controller: DynamicTeamController) -> None:
+    """Set the global controller instance.
+
+    Called by TradingPipeline when initializing the PM agent.
+    """
+    global _controller_instance
+    _controller_instance = controller
+
+
+def get_controller() -> Optional[DynamicTeamController]:
+    """Get the global controller instance.
+
+    Returns:
+        DynamicTeamController instance or None if not set
+    """
+    return _controller_instance
+
+
+# Tool functions that wrap the controller methods
+# These are registered with the PM's toolkit
+
+def create_analyst(
+    agent_id: str,
+    analyst_type: str,
+    name: Optional[str] = None,
+    focus: Optional[str] = None,
+    description: Optional[str] = None,
+    soul_md: Optional[str] = None,
+    agents_md: Optional[str] = None,
+    model_name: Optional[str] = None,
+) -> Dict[str, Any]:
+    """Tool wrapper for create_analyst.
+
+    Note: focus parameter accepts comma-separated string for tool compatibility.
+    """
+    controller = get_controller()
+    if not controller:
+        return {"success": False, "error": "Dynamic team controller not available"}
+
+    focus_list = [f.strip() for f in focus.split(",")] if focus else None
+    return controller.create_analyst(
+        agent_id=agent_id,
+        analyst_type=analyst_type,
+        name=name,
+        focus=focus_list,
+        description=description,
+        soul_md=soul_md,
+        agents_md=agents_md,
+        model_name=model_name,
+    )
+
+
+def clone_analyst(
+    source_id: str,
+    new_id: str,
+    name: Optional[str] = None,
+    focus_additions: Optional[str] = None,
+    description_override: Optional[str] = None,
+    model_name: Optional[str] = None,
+) -> Dict[str, Any]:
+    """Tool wrapper for clone_analyst.
+
+    Note: focus_additions accepts comma-separated string.
+    """
+    controller = get_controller()
+    if not controller:
+        return {"success": False, "error": "Dynamic team controller not available"}
+
+    additions_list = [f.strip() for f in focus_additions.split(",")] if focus_additions else None
+    return controller.clone_analyst(
+        source_id=source_id,
+        new_id=new_id,
+        name=name,
+        focus_additions=additions_list,
+        description_override=description_override,
+        model_name=model_name,
+    )
+
+
+def remove_analyst(agent_id: str) -> Dict[str, Any]:
+    """Tool wrapper for remove_analyst."""
+    controller = get_controller()
+    if not controller:
+        return {"success": False, "error": "Dynamic team controller not available"}
+    return controller.remove_analyst(agent_id)
+
+
+def list_analyst_types() -> List[Dict[str, Any]]:
+    """Tool wrapper for list_analyst_types."""
+    controller = get_controller()
+    if not controller:
+        return []
+    return controller.list_analyst_types()
+
+
+def get_analyst_info(agent_id: str) -> Dict[str, Any]:
+    """Tool wrapper for get_analyst_info."""
+    controller = get_controller()
+    if not controller:
+        return {"found": False, "error": "Controller not available"}
+    return controller.get_analyst_info(agent_id)
+
+
+def get_team_summary() -> Dict[str, Any]:
+    """Tool wrapper for get_team_summary."""
+    controller = get_controller()
+    if not controller:
+        return {"error": "Controller not available"}
+    return controller.get_team_summary()
+
+
+__all__ = [
+    "DynamicTeamController",
+    "set_controller",
+    "get_controller",
+    "create_analyst",
+    "clone_analyst",
+    "remove_analyst",
+    "list_analyst_types",
+    "get_analyst_info",
+    "get_team_summary",
+]

deploy/README.md

@@ -10,37 +10,12 @@ For the planned convergence work, see
 
 ## Contents
 
-- [deploy/systemd/evotraders.service](./systemd/evotraders.service)
-  - systemd unit for the long-running 大时代 gateway process
-- [scripts/run_prod.sh](../scripts/run_prod.sh)
-  - production launch script used by the systemd unit
 - [deploy/nginx/bigtime.cillinn.com.conf](./nginx/bigtime.cillinn.com.conf)
   - HTTPS nginx config with WebSocket proxying
 - [deploy/nginx/bigtime.cillinn.com.http.conf](./nginx/bigtime.cillinn.com.http.conf)
   - plain HTTP/static-site variant
 
-## Deployment Topology Options
-
-This directory documents two deployment topologies:
-
-### 1. Compatibility Topology (backend.main) - CURRENT PRODUCTION DEFAULT
-
-The checked-in production path uses the **compatibility gateway** (`backend.main`):
-
-- nginx serves the built frontend from `/var/www/bigtime/current`
-- public domain examples use `bigtime.cillinn.com`
-- nginx proxies `/ws` to `127.0.0.1:8765`
-- systemd runs `scripts/run_prod.sh`
-- `scripts/run_prod.sh` starts `python3 -m backend.main` in live mode on `127.0.0.1:8765`
-
-This is a **monolithic gateway** that embeds all services internally. It is the
-current production default for simplicity but does not expose the split FastAPI
-services directly.
-
-**When to use**: Single-server deployments, simpler operational requirements,
-backwards compatibility with existing monitoring.
-
-### 2. Preferred Topology (Split Services) - RECOMMENDED FOR NEW DEPLOYMENTS
+## Recommended Topology
 
 The modern architecture exposes individual FastAPI services:
 
@@ -76,27 +51,9 @@ To deploy in split-service mode, you would:
 
 ## systemd
 
-The current systemd unit:
-
-- uses `WorkingDirectory=/root/code/evotraders`
-- executes [scripts/run_prod.sh](../scripts/run_prod.sh)
-- restarts automatically on failure
-
-Enable and start:
-
-```bash
-sudo cp deploy/systemd/evotraders.service /etc/systemd/system/evotraders.service
-sudo systemctl daemon-reload
-sudo systemctl enable evotraders
-sudo systemctl start evotraders
-```
-
-Check status and logs:
-
-```bash
-sudo systemctl status evotraders
-journalctl -u evotraders -f
-```
+No maintained systemd unit is checked into the repository anymore. If deployment
+work resumes, add units that mirror the split-service topology used in local
+development.
 
 ## nginx
 
@@ -124,7 +81,7 @@ The checked-in TLS config expects Let's Encrypt assets at:
 Before using the production scripts, ensure the runtime environment has:
 
 - a usable Python environment
-- backend dependencies installed from `requirements.txt`
+- backend dependencies installed from the checked-in Python package metadata in `pyproject.toml`
 - the package installed with `pip install -e .` or `uv pip install -e .`
 - frontend dependencies installed with `npm ci`
 - repo dependencies installed
@@ -136,30 +93,10 @@ Recommended production install sequence:
 ```bash
 python3 -m venv .venv
 source .venv/bin/activate
-pip install -r requirements.txt
 pip install -e .
 cd frontend && npm ci && npm run build && cd ..
 ```
 
-The production script currently sets:
-
-```bash
-PYTHONPATH=/root/code/evotraders/.pydeps:.
-TICKERS=${TICKERS:-AAPL,MSFT,GOOGL,AMZN,NVDA,META,TSLA,AMD,NFLX,AVGO,PLTR,COIN}
-```
-
-It then launches the current compatibility gateway/runtime process:
-
-```bash
-python3 -m backend.main \
-  --mode live \
-  --config-name production \
-  --host 127.0.0.1 \
-  --port 8765 \
-  --trigger-time now \
-  --poll-interval 15
-```
-
 ## Skill Sandbox Configuration
 
 Production deployments should enable Docker-based skill sandbox for security isolation:
@@ -172,7 +109,7 @@ pip install -e ".[docker-sandbox]"
 docker info
 ```
 
-Environment variables (set by `scripts/run_prod.sh` with defaults):
+Example environment variables for a future deployment:
 
 | Variable | Default | Description |
 |----------|---------|-------------|
@@ -196,6 +133,5 @@ The checked-in deployment artifacts do not currently document or automate:
 - frontend build/publish steps
 - secret management
 
-If you move production fully to split-service mode, update this directory so it
-documents the new service topology explicitly instead of relying on the gateway-
-only path.
+If deployment returns to active development, rewrite this directory around the
+same split-service topology used by `start-dev.sh`.

deploy/systemd/evotraders.service

@@ -1,22 +0,0 @@
-[Unit]
-Description=大时代 Production Service
-After=network.target
-# COMPATIBILITY_SURFACE: stable
-# OWNER: ops-team
-# SEE: docs/legacy-inventory.md#gateway-first-production-example
-#
-# This systemd unit runs the gateway-first production topology.
-# It executes scripts/run_prod.sh which launches backend.main as the
-# primary gateway/runtime process. For split-service deployment topology,
-# see docs/current-architecture.md and deploy/README.md
-
-[Service]
-Type=simple
-WorkingDirectory=/root/code/evotraders
-ExecStart=/root/code/evotraders/scripts/run_prod.sh
-Restart=always
-RestartSec=5
-Environment=PYTHONUNBUFFERED=1
-
-[Install]
-WantedBy=multi-user.target

docs/CRITICAL_FIXES.md

@@ -1,239 +0,0 @@
-# 关键代码修复方案
-
-## 1. EvoAgent 长期记忆支持 ✅
-
-**状态**: EvoAgent 已支持 `long_term_memory` 参数，但需要移除 Legacy 回退逻辑
-
-**需要修改的文件**:
-- `backend/main.py` 第 158-176 行 - 移除记忆启用时的 Legacy 回退
-- `backend/core/pipeline.py` - 同样更新
-- `backend/core/pipeline_runner.py` - 同样更新
-
-**修复代码** (main.py):
-```python
-def _create_analyst_agent(...):
-    # ... 工具包创建代码 ...
-    
-    use_evo_agent = analyst_type in _resolve_evo_agent_ids()
-    
-    if use_evo_agent:
-        workspace_dir = skills_manager.get_agent_asset_dir(config_name, analyst_type)
-        agent_config = load_agent_workspace_config(workspace_dir / "agent.yaml")
-        agent = EvoAgent(
-            agent_id=analyst_type,
-            config_name=config_name,
-            workspace_dir=workspace_dir,
-            model=model,
-            formatter=formatter,
-            skills_manager=skills_manager,
-            prompt_files=agent_config.prompt_files,
-            long_term_memory=long_term_memory,  # 已支持
-            long_term_memory_mode="static_control",
-        )
-        agent.toolkit = toolkit
-        setattr(agent, "workspace_id", config_name)
-        return agent
-    
-    # Legacy fallback (deprecated)
-    return AnalystAgent(...)
-```
-
-## 2. Workspace ID 语义清理
-
-**问题**: `workspace_id` 同时用于 design-time 和 runtime 两个不同概念
-
-**修复方案**:
-
-```python
-# backend/api/workspaces.py
-# 明确区分两种资源
-
-# Design-time workspaces (CRUD)
-@router.get("/design-workspaces/{workspace_id}/...")
-async def get_design_workspace(workspace_id: str): ...
-
-# Runtime runs (只读)
-@router.get("/runs/{run_id}/agents/{agent_id}/...")
-async def get_runtime_agent(run_id: str, agent_id: str): ...
-```
-
-## 3. ToolGuard 与 Gateway 审批同步 ✅ 已完成
-
-**状态**: 审批同步已完善，添加了批量审批支持
-
-**API 端点**:
-- `POST /api/guard/check` - 检查工具调用是否需要审批
-- `POST /api/guard/approve` - 批准单个工具调用
-- `POST /api/guard/approve/batch` - ✅ 批量批准多个工具调用（新增）
-- `POST /api/guard/deny` - 拒绝工具调用
-- `GET /api/guard/pending` - 获取待审批列表
-
-**批量审批示例**:
-```python
-# 批量批准
-await approve_tool_calls(
-    BatchApprovalRequest(
-        approval_ids=["approval_001", "approval_002", "approval_003"],
-        one_time=True,
-    )
-)
-```
-
-**超时处理**: 默认 300 秒超时，可在 `ToolGuardMixin._init_tool_guard()` 中配置
-
-## 4. Smoke Test 依赖修复
-
-**需要的依赖**:
-```bash
-pip install pandas numpy matplotlib seaborn
-pip install finnhub-python yfinance
-pip install loguru rich
-pip install websockets
-pip install httpx requests
-pip install PyYAML
-pip install pandas-market-calendars exchange-calendars
-```
-
-## 5. 统一 Agent 工厂 ✅ 已完成
-
-**文件** `backend/agents/unified_factory.py`:
-
-统一工厂已创建，支持：
-- 所有 6 种 Agent 角色的创建
-- 自动 EvoAgent vs Legacy Agent 选择
-- Workspace 驱动配置
-- 长期记忆支持
-
-```python
-from backend.agents.unified_factory import UnifiedAgentFactory, get_agent_factory
-
-# 使用示例
-factory = UnifiedAgentFactory(
-    config_name="smoke_fullstack",
-    skills_manager=skills_manager,
-)
-
-# 创建分析师
-analyst = factory.create_analyst(
-    analyst_type="fundamentals_analyst",
-    model=model,
-    formatter=formatter,
-    long_term_memory=memory,
-)
-```
-
-## 6. EvoAgent 默认启用
-
-**修改** `backend/config/constants.py`:
-
-```python
-# 默认所有角色使用 EvoAgent
-DEFAULT_EVO_AGENT_ROLES = {
-    "fundamentals_analyst",
-    "technical_analyst", 
-    "sentiment_analyst",
-    "valuation_analyst",
-    "risk_manager",
-    "portfolio_manager",
-}
-
-# EVO_AGENT_IDS 现在用于选择性地禁用 EvoAgent
-# 如果设置，只启用指定的角色
-# 如果未设置，启用所有角色
-```
-
-**修改** `backend/main.py`:
-```python
-def _resolve_evo_agent_ids() -> set[str]:
-    """Return agent ids selected to use EvoAgent.
-    
-    By default, all supported roles use EvoAgent.
-    EVO_AGENT_IDS can be used to limit to specific roles.
-    """
-    from backend.config.constants import DEFAULT_EVO_AGENT_ROLES
-    
-    raw = os.getenv("EVO_AGENT_IDS", "")
-    if raw.strip():
-        # Filter to only valid roles
-        requested = {x.strip() for x in raw.split(",") if x.strip()}
-        return requested & DEFAULT_EVO_AGENT_ROLES
-    
-    # Default: all roles use EvoAgent
-    return DEFAULT_EVO_AGENT_ROLES
-```
-
-## 7. 遗留代码清理
-
-**可以删除的文件**:
-- `backend/agents/compat.py` ✅ 已删除
-- `frontend/src/hooks/useWebsocketSessionSync.js` ✅ 已删除
-
-**标记为废弃的文件** ✅ 已完成:
-- `backend/agents/analyst.py` - 已添加 DeprecationWarning
-- `backend/agents/risk_manager.py` - 已添加 DeprecationWarning
-- `backend/agents/portfolio_manager.py` - 已添加 DeprecationWarning
-
-## 8. 测试修复
-
-**更新** `backend/tests/test_evo_agent_selection.py`:
-
-移除这些测试 ✅ 已完成:
-- `test_main_create_analyst_agent_falls_back_to_legacy_when_memory_enabled`
-- `test_main_create_risk_manager_falls_back_to_legacy_when_memory_enabled`
-- `test_main_create_portfolio_manager_falls_back_to_legacy_when_memory_enabled`
-
-添加新测试 ✅ 已完成:
-- `test_evo_agent_supports_long_term_memory`
-- `test_all_roles_use_evo_agent_by_default`
-
-新增集成测试文件 ✅ 已完成:
-- `backend/tests/test_evo_agent_integration.py` - 13 个集成测试覆盖 Factory、ToolGuard、Workspace 集成
-
-## 9. 快速修复清单
-
-运行以下命令应用关键修复:
-
-```bash
-# 1. 修复 EvoAgent 记忆支持 (修改 main.py, pipeline.py, pipeline_runner.py)
-# 移除 long_term_memory 检查导致的 Legacy 回退
-
-# 2. 修复默认 EvoAgent 启用
-sed -i 's/def _resolve_evo_agent_ids():/def _resolve_evo_agent_ids() -> set[str]:/' backend/main.py
-
-# 3. 确保所有测试通过
-pytest backend/tests/test_evo_agent_selection.py -v
-
-# 4. 运行 smoke test
-python3 scripts/smoke_evo_runtime.py --test-all-roles
-```
-
-## 10. 实施进度
-
-### ✅ 已完成
-
-| 任务 | 状态 | 文件 |
-|------|------|------|
-| EvoAgent 长期记忆支持 | ✅ 已完成 | `evo_agent.py`, `main.py` |
-| 默认启用所有角色 EvoAgent | ✅ 已完成 | `main.py`, `pipeline.py` |
-| 统一 Agent 工厂 | ✅ 已完成 | `unified_factory.py` |
-| ToolGuard Gateway 同步 | ✅ 已完成 | `tool_guard.py`, `guard.py` |
-| ToolGuard 批量审批 | ✅ 已完成 | `guard.py` |
-| 废弃标记 Legacy Agent | ✅ 已完成 | `analyst.py`, `risk_manager.py`, `portfolio_manager.py` |
-| 集成测试 | ✅ 已完成 | `test_evo_agent_integration.py` |
-| 类型注解 | ✅ 已完成 | `unified_factory.py` |
-| Team 基础设施 | ✅ 已完成 | `messenger.py`, `task_delegator.py` |
-| Skills 沙盒执行 | ✅ 已完成 | `sandboxed_executor.py` |
-
-### 🚧 待完成
-
-| 优先级 | 任务 | 说明 |
-|--------|------|------|
-| P0 | Smoke Test 依赖修复 | 需要安装 pandas, finnhub, pandas-market-calendars 等 |
-| P1 | Workspace ID 语义清理 | ✅ 已添加 `run_id`，保留 `workspace_id` 用于向后兼容 |
-| P2 | 文档完善 | ✅ 已完成 |
-
-*最后更新: 2026-04-02*
-
----
-
-*文档生成时间: 2026-04-01*

docs/OPTIMIZATION_PLAN.md

@@ -1,249 +0,0 @@
-# 大时代项目优化和功能补齐计划
-
-## 当前状态评估
-
-### 已完成的工作
-1. ✅ EvoAgent 核心实现 (`backend/agents/base/evo_agent.py`)
-2. ✅ ToolGuardMixin 工具守卫 (`backend/agents/base/tool_guard.py`)
-3. ✅ Hooks 系统 (`backend/agents/base/hooks.py`)
-4. ✅ Smoke test 脚本 (`scripts/smoke_evo_runtime.py`)
-5. ✅ 选择性 EvoAgent 测试 (`backend/tests/test_evo_agent_selection.py`)
-6. ✅ 删除 `backend/agents/compat.py` 兼容性层
-7. ✅ 删除 `useWebsocketSessionSync.js` 旧钩子
-
-### 遗留问题清单
-
-#### 🔴 P0: 阻塞 EvoAgent 全面推出
-
-| # | 问题 | 位置 | 影响 | 解决方案 |
-|---|------|------|------|----------|
-| P0-1 | EvoAgent 不支持长期记忆 | `evo_agent.py:165-166` | 启用 memory 时回退到 Legacy Agent | 集成 ReMe 记忆系统 |
-| P0-2 | Pipeline 运行时分析师创建路径不一致 | `pipeline.py` | 运行时动态创建可能跳过 EvoAgent 路径 | 统一 `_create_runtime_analyst` 逻辑 |
-| P0-3 | Workspace 加载路径混乱 | `workspace.py`, `workspace_manager.py` | `workspace_id` vs `run_id` 语义混合 | 明确区分 design-time 和 runtime 路径 |
-| P0-4 | Smoke test 失败排查 | `scripts/smoke_evo_runtime.py` | 无法验证 EvoAgent 是否正确启动 | 修复测试并确保通过 |
-
-#### 🟡 P1: 功能完善
-
-| # | 问题 | 位置 | 影响 | 解决方案 |
-|---|------|------|------|----------|
-| P1-1 | Team 基础设施未完成 | `evo_agent.py:41-48` | Agent 间通信和任务委托不可用 | 完成 messenger 和 task_delegator |
-| P1-2 | ToolGuard 与 Gateway 审批流程集成 | `tool_guard.py`, `api/guard.py` | 审批状态同步可能不一致 | 统一审批存储和事件通知 |
-| P1-3 | Skills 沙盒执行 | `tools/sandboxed_executor.py` | 生产环境需要 Docker 隔离 | 完善沙盒执行器 |
-| P1-4 | 错误处理和重试机制 | 多处 | 部分错误未正确处理 | 添加统一的错误处理 |
-
-#### 🟢 P2: 代码质量和可维护性
-
-| # | 问题 | 位置 | 影响 | 解决方案 |
-|---|------|------|------|----------|
-| P2-1 | 重复的 Agent 创建逻辑 | `main.py`, `pipeline.py`, `pipeline_runner.py` | 维护困难，容易遗漏 | 提取统一的 Agent 工厂 |
-| P2-2 | 类型注解不完整 | 多处 | IDE 提示不足 | 完善类型注解 |
-| P2-3 | 缺少 EvoAgent 集成测试 | `backend/tests/` | 无法确保功能完整 | 添加集成测试 |
-| P2-4 | 文档和注释 | 多处 | 新贡献者理解困难 | 完善文档 |
-
----
-
-## 详细实施方案
-
-### Phase 1: P0 阻塞问题修复
-
-#### P0-1: EvoAgent 长期记忆支持
-
-**问题描述**:
-```python
-# main.py 中当前逻辑
-if long_term_memory and agent_id not in EVO_AGENT_IDS:
-    # 使用 Legacy Agent
-else:
-    # 使用 EvoAgent
-```
-
-**目标**: EvoAgent 支持 ReMe 长期记忆系统
-
-**实施步骤**:
-1. 在 `EvoAgent.__init__` 中正确接收 `long_term_memory` 参数
-2. 集成 ReMe 记忆系统的读写
-3. 在 Hooks 中添加记忆相关的生命周期管理
-4. 修改 `main.py`, `pipeline.py` 中移除 EvoAgent 的记忆回退逻辑
-
-**文件修改**:
-- `backend/agents/base/evo_agent.py`
-- `backend/main.py`
-- `backend/core/pipeline.py`
-
-#### P0-2: Pipeline 运行时分析师创建统一
-
-**问题描述**:
-`TradingPipeline._create_runtime_analyst` 方法需要确保:
-1. 检查 `EVO_AGENT_IDS` 环境变量
-2. 正确传递所有必要参数给 EvoAgent
-3. 处理 workspace 资产准备
-
-**实施步骤**:
-1. 统一 `pipeline.py` 和 `main.py` 中的 Agent 创建逻辑
-2. 确保 EvoAgent 路径和 Legacy 路径参数一致
-3. 添加运行时动态 Agent 创建的测试
-
-**文件修改**:
-- `backend/core/pipeline.py`
-- `backend/main.py`
-
-#### P0-3: Workspace 路径清理
-
-**问题描述**:
-- `workspace_id` 有时指 `workspaces/` 目录下的设计时 workspace
-- 有时指 `runs/<run_id>/` 下的运行时 workspace
-
-**解决方案**:
-1. 明确命名：`design_workspace_id` vs `run_id`
-2. 在 API 路由中区分两种资源
-3. 内部统一使用 `run_id` 作为运行时标识
-
-**文件修改**:
-- `backend/api/workspaces.py`
-- `backend/api/agents.py`
-- `backend/agents/workspace_manager.py`
-
-#### P0-4: Smoke Test 修复
-
-**当前测试**:
-```bash
-python3 scripts/smoke_evo_runtime.py --agent-id fundamentals_analyst
-```
-
-**验证点**:
-1. Gateway 正常启动
-2. EvoAgent 日志出现
-3. `runtime_state.json` 正确写入
-4. 审批流程正常工作
-
-**实施步骤**:
-1. 运行测试并识别失败点
-2. 修复 EvoAgent 初始化问题
-3. 确保所有 6 个角色都能通过测试
-
----
-
-### Phase 2: P1 功能完善
-
-#### P1-1: Team 基础设施
-
-**当前状态**:
-```python
-try:
-    from backend.agents.team.messenger import AgentMessenger
-    from backend.agents.team.task_delegator import TaskDelegator
-    TEAM_INFRA_AVAILABLE = True
-except ImportError:
-    TEAM_INFRA_AVAILABLE = False
-```
-
-**目标**: 完成 Agent 间通信和任务委托
-
-**实施步骤**:
-1. 完成 `AgentMessenger` 实现
-2. 完成 `TaskDelegator` 实现
-3. 添加 Agent 团队协调的测试
-
-#### P1-2: ToolGuard 与 Gateway 集成
-
-**当前状态**:
-- `ToolGuardStore` 是内存存储
-- Gateway 通过 `get_global_runtime_manager()` 访问
-
-**改进**:
-1. 确保审批状态在 Gateway 和 Agent 间同步
-2. 添加审批超时处理
-3. 支持批量审批
-
-#### P1-3: Skills 沙盒执行
-
-**当前状态**:
-```python
-SKILL_SANDBOX_MODE=none  # 开发模式，直接执行
-```
-
-**目标**: 生产环境使用 Docker 隔离
-
-**实施步骤**:
-1. 完成 `DockerSandboxBackend`
-2. 添加资源限制（CPU、内存、网络）
-3. 添加执行超时控制
-
----
-
-### Phase 3: P2 代码质量
-
-#### P2-1: 统一 Agent 工厂
-
-**目标**: 提取 `AgentFactory` 统一处理所有 Agent 创建
-
-**设计**:
-```python
-class AgentFactory:
-    def create_analyst(self, analyst_type: str, **kwargs) -> BaseAgent
-    def create_risk_manager(self, **kwargs) -> BaseAgent
-    def create_portfolio_manager(self, **kwargs) -> BaseAgent
-```
-
-#### P2-2: 类型注解
-
-**目标**: 所有公共 API 完整的类型注解
-
-#### P2-3: 集成测试
-
-**目标**: EvoAgent 完整的端到端测试
-
----
-
-## 实施顺序
-
-### Week 1: P0 阻塞问题
-1. [ ] P0-4: 运行 Smoke Test，识别失败点
-2. [ ] P0-1: EvoAgent 长期记忆支持
-3. [ ] P0-2: Pipeline 运行时统一
-4. [ ] P0-3: Workspace 路径清理
-5. [ ] 验证所有 Smoke Test 通过
-
-### Week 2: P1 功能完善
-1. [ ] P1-1: Team 基础设施
-2. [ ] P1-2: ToolGuard 集成优化
-3. [ ] P1-3: Skills 沙盒执行
-
-### Week 3: P2 代码质量
-1. [ ] P2-1: 统一 Agent 工厂
-2. [ ] P2-2: 类型注解
-3. [ ] P2-3: 集成测试
-4. [ ] P2-4: 文档完善
-
----
-
-## 成功标准
-
-### EvoAgent 全面推出标准
-1. ✅ 所有 6 个角色通过 smoke test
-2. ✅ 长期记忆功能正常工作
-3. ✅ 无需 `EVO_AGENT_IDS` 环境变量即可使用 EvoAgent
-4. ✅ Legacy Agent 代码标记为 deprecated
-5. ✅ 集成测试覆盖主要使用场景
-
-### 架构清理标准
-1. ✅ `runs/<run_id>/` 是唯一的运行时数据来源
-2. ✅ `workspaces/` 仅用于设计时注册表
-3. ✅ 所有服务边界清晰，无循环依赖
-4. ✅ 文档和代码一致
-
----
-
-## 风险和对策
-
-| 风险 | 可能性 | 影响 | 对策 |
-|------|--------|------|------|
-| EvoAgent 与 Legacy 行为不一致 | 中 | 高 | 并行运行对比测试 |
-| 长期记忆集成复杂 | 中 | 中 | 分阶段实现，先支持基础功能 |
-| 性能下降 | 低 | 高 | 基准测试，性能剖析 |
-| 迁移期间系统不稳定 | 中 | 高 | 保持 Legacy 作为回退 |
-
----
-
-*计划创建日期: 2026-04-01*
-*负责: Claude Code*

docs/README.md

@@ -0,0 +1,29 @@
+# Docs Index
+
+这个目录只保留当前仍有维护价值的文档。
+阅读顺序建议从“权威事实”开始，再看“兼容清理”和“规划说明”。
+
+## 权威事实
+
+- [`current-architecture.md`](./current-architecture.md) — 当前项目架构、运行路径和真相源定义
+- [`project-layout.md`](./project-layout.md) — 当前仓库目录职责与阅读方式
+- [`terminology.md`](./terminology.md) — runtime、run、workspace 等核心术语
+- [`runtime-api-changes.md`](./runtime-api-changes.md) — 当前 `runtime_service` API 约定与行为补充
+- [`current-architecture.excalidraw`](./current-architecture.excalidraw) — 架构图源文件
+
+## 兼容与遗留清理
+
+- [`legacy-inventory.md`](./legacy-inventory.md) — 兼容层、遗留代码和历史边界清单
+- [`compat-removal-plan.md`](./compat-removal-plan.md) — 兼容移除计划与收敛方向
+
+## 规划说明
+
+- [`development-roadmap.md`](./development-roadmap.md) — 后续演进方向和阶段性路线
+
+## 使用原则
+
+- 讨论“现在系统怎么工作”，优先看 `current-architecture.md`
+- 讨论“顶层目录应该怎么理解”，优先看 `project-layout.md`
+- 讨论“某个旧入口为什么还在”，优先看 `legacy-inventory.md`
+- 讨论“准备删哪些兼容路径”，优先看 `compat-removal-plan.md`
+- 不要把历史计划或阶段性修复记录当成当前事实源

docs/compat-removal-plan.md

@@ -67,11 +67,11 @@ backward-compatible behavior while migration settles.
 
 ## Remaining Migration Risks
 
-### Split service deployment is not yet the checked-in production default
+### Checked-in deployment artifacts still lag the development topology
 
 - The repo documents split-service local development clearly.
-- The checked-in production example still centers on `backend.main` and nginx
-  WebSocket proxying.
+- The checked-in deployment docs still center on historical nginx
+  WebSocket proxying rather than the active dev topology.
 - This is a topology mismatch to keep in mind when changing deploy docs or prod
   automation.
 
@@ -93,13 +93,13 @@ backward-compatible behavior while migration settles.
 Migration can be considered effectively complete when all of the following are
 true:
 
-1. Production deployment docs and scripts explicitly run the same split-service
-   topology used in development, or intentionally document a different stable
-   production topology.
+1. Deployment docs and scripts explicitly run the same split-service
+   topology used in development, or are removed from the repo.
 2. Critical read paths no longer require ambiguous fallback behavior to local
    module implementations.
 3. OpenClaw integration is documented as a stable contract with clear guidance
-   on when to use the WebSocket gateway versus the REST surface.
+   on when to use the WebSocket gateway versus the small set of CLI-backed
+   gateway read helpers.
 4. The frontend-service routing model is stable enough that direct-service and
    gateway-mediated paths are deliberate design choices rather than migration
    leftovers.
@@ -137,9 +137,6 @@ Recommended next action:
 These still have an operational reason to exist and should be documented rather
 than treated as accidental leftovers.
 
-- `backend.main`
-  - compatibility gateway/runtime process
-  - still relevant for websocket transport and current deploy topology
 - `runs/<run_id>/team_dashboard/*.json`
   - export/consumer compatibility layer
 - gateway-mediated websocket/event flow
@@ -147,8 +144,8 @@ than treated as accidental leftovers.
 
 Recommended next action:
 
-- keep these, but document them as intentional compatibility surfaces with
-  explicit ownership.
+- keep only surfaces with an active operational consumer, and avoid routing new
+  development through them.
 
 ### 3. Defer Until Topology Decisions Are Final
 
@@ -157,8 +154,8 @@ churn without simplifying the current runtime.
 
 - `workspaces/` design-time registry versus `runs/<run_id>/` runtime state
 - env-dependent service fallback behavior
-- checked-in deployment docs centered on `backend.main`
-- dual OpenClaw shapes: gateway integration and REST facade
+- checked-in deployment docs that have not yet been rewritten around split services
+- dual OpenClaw access patterns: gateway integration and CLI-backed read helpers
 
 Recommended next action:
 

docs/current-architecture.excalidraw

@@ -386,13 +386,13 @@
       "updated": 1,
       "link": null,
       "locked": false,
-      "text": "Gateway :8765\\nbackend.main\\nWebSocket + runtime orchestration",
+      "text": "Gateway :8765\\nGateway process\\nWebSocket + runtime orchestration",
       "fontSize": 18,
       "fontFamily": 5,
       "textAlign": "center",
       "verticalAlign": "middle",
       "containerId": null,
-      "originalText": "Gateway :8765\nbackend.main\nWebSocket + runtime orchestration",
+      "originalText": "Gateway :8765\nGateway process\nWebSocket + runtime orchestration",
       "lineHeight": 1.2
     },
     {
@@ -696,13 +696,13 @@
       "updated": 1,
       "link": null,
       "locked": false,
-      "text": "OpenClaw WS :18789\\noptional REST :8004",
+      "text": "OpenClaw WS :18789\\nCLI-backed reads via gateway",
       "fontSize": 20,
       "fontFamily": 5,
       "textAlign": "center",
       "verticalAlign": "middle",
       "containerId": null,
-      "originalText": "OpenClaw WS :18789\noptional REST :8004",
+      "originalText": "OpenClaw WS :18789\nCLI-backed reads via gateway",
       "lineHeight": 1.2
     },
     {

docs/current-architecture.md

@@ -1,8 +1,8 @@
 # Current Architecture
 
 This file describes the current code-supported architecture only. Historical
-paths and partial migrations are intentionally excluded unless called out as
-legacy compatibility.
+paths and partial migrations are intentionally excluded unless brief historical
+context is needed to explain the current shape.
 
 Reference material:
 
@@ -11,40 +11,21 @@ Reference material:
 - legacy inventory: [legacy-inventory.md](./legacy-inventory.md)
 - terminology guide: [terminology.md](./terminology.md)
 
-## Runtime Modes
+## Runtime Mode
 
-The system supports two distinct runtime modes:
+The supported runtime model is the split-service development architecture.
 
-### Standalone Mode (Legacy Compatibility)
-
-Direct Gateway startup via `backend.main` as a monolithic entrypoint.
-
-```bash
-python -m backend.main --mode live --port 8765
-```
-
-**Characteristics:**
-- Single process runs Gateway, Pipeline, Market Service, and Scheduler
-- No service discovery or process management
-- Suitable for single-node deployments and quick testing
-- All components share the same memory space
-
-**Use cases:**
-- Quick local testing without service orchestration
-- Single-node production deployments
-- Backward compatibility with legacy startup scripts
-
-### Microservice Mode (Default for Development)
-
-Split-service architecture with dedicated runtime_service managing the Gateway lifecycle.
+Split-service architecture with a dedicated runtime API surface and a separate
+Gateway process.
 
 ```bash
 ./start-dev.sh  # Starts all services including runtime_service and Gateway
 ```
 
 **Characteristics:**
-- `runtime_service` (:8003) acts as Gateway Process Manager
-- Gateway runs as a subprocess managed by runtime_service
+- `runtime_service` (:8003) provides runtime lifecycle APIs
+- the checked-in `start-dev.sh` starts split services and lets `runtime_service` spawn Gateway
+- manual split-service flows can also let `runtime_service` spawn Gateway
 - Clear separation between Control Plane (runtime_service) and Data Plane (Gateway)
 - Service discovery via environment variables
 - Independent scaling and deployment of each service
@@ -54,20 +35,7 @@ Split-service architecture with dedicated runtime_service managing the Gateway l
 - Multi-node deployments
 - Production environments requiring service isolation
 
-## Mode Comparison
-
-| Aspect | Standalone Mode | Microservice Mode |
-|--------|-----------------|-------------------|
-| **Entry point** | `python -m backend.main` | `./start-dev.sh` or individual services |
-| **Process model** | Single monolithic process | Multiple specialized processes |
-| **Gateway management** | Self-contained | Managed by runtime_service |
-| **Service discovery** | None (in-process) | Environment variable based |
-| **Hot reload** | Full restart required | Per-service reload |
-| **Scaling** | Vertical only | Horizontal possible |
-| **Complexity** | Lower | Higher |
-| **Use case** | Testing, simple deployments | Development, production |
-
-## Default Runtime Shape (Microservice Mode)
+## Default Runtime Shape
 
 The active runtime path is:
 
@@ -83,7 +51,6 @@ Current service surfaces:
   - read-only explain/news APIs
 - `backend.apps.runtime_service` on `:8003`
   - runtime lifecycle and gateway process management
-  - optional OpenClaw REST facade
 - gateway WebSocket on `:8765`
   - live feed/event transport and pipeline coordination
 
@@ -163,11 +130,11 @@ use `docker` mode with appropriate resource limits.
 
 ### Current State
 
-The system is in a transitional state:
+The system is in an active development state:
 
 1. **Microservice infrastructure is operational** - runtime_service can start/stop Gateway as subprocess
 2. **Pipeline logic remains in Gateway** - full Pipeline execution still happens within Gateway process
-3. **Standalone mode is preserved** - direct `backend.main` startup for compatibility
+3. **Direct gateway startup has been removed** - the repository now exposes a single supported startup model
 
 ### Future Direction
 
@@ -184,9 +151,9 @@ Phase 3: Pipeline decomposition (planned)
 - Gateway becomes a thin event router
 - runtime_service evolves into full orchestrator
 
-Phase 4: Standalone mode deprecation (future)
-- Remove direct `backend.main` entry point
-- All deployments use microservice mode
+Phase 4: Deployment convergence (future)
+- Remove or rewrite historical deployment artifacts
+- Keep all documented startup paths aligned with `runtime_service`
 
 ## Legacy Compatibility
 
@@ -194,8 +161,8 @@ These items still exist, but they are not the recommended source of truth for
 new development:
 
 - root-level runtime data directories such as `live/`, `production/`, `backtest/`
-- direct `backend.main` startup as the primary development path
+- historical documentation gaps that have not yet been fully rewritten
 
-The current runtime still creates legacy `AnalystAgent` / `RiskAgent` /
-`PMAgent` instances directly. EvoAgent remains an in-progress migration target,
-not the default execution path.
+Legacy fallback agent paths still exist in compatibility-oriented creation
+flows, but the default `TradingPipeline` runtime now prefers `EvoAgent` for the
+supported roles unless rollout settings explicitly reduce that set.

docs/development-roadmap.md

@@ -63,14 +63,14 @@ Goal: preserve only intentional compatibility layers.
 
 Planned work:
 
-- identify startup scripts and deploy artifacts that still center on
-  `backend.main` as a monolithic entrypoint
+- identify any remaining deployment docs that still lag the split-service topology
 - classify compatibility surfaces into:
   - stable and intentional
   - temporary and shrinking
   - removable once replacements are fully active
 - reduce env-dependent fallback ambiguity for read-only service routing where practical
-- document the difference between OpenClaw WebSocket integration and the optional REST facade
+- document the difference between OpenClaw WebSocket integration and the small
+  set of CLI-backed gateway read helpers
 
 Definition of done:
 

docs/legacy-inventory.md

@@ -21,7 +21,7 @@ These are the current defaults to build against:
   - `backend.apps.trading_service` on `:8001`
   - `backend.apps.news_service` on `:8002`
 - gateway process
-  - `backend.main`
+  - `backend.gateway_server` in the default managed-runtime path
   - `backend.services.gateway` on `:8765`
 
 ## Compatibility Surface Classification
@@ -35,7 +35,6 @@ compatibility surfaces with explicit ownership.
 
 | Surface | Location | Owner | Reason |
 |---------|----------|-------|--------|
-| Gateway-first production | `scripts/run_prod.sh`, `deploy/systemd/`, `deploy/nginx/` | ops-team | Current production example runs gateway directly and proxies `/ws` |
 | Dashboard export layer | `runs/<run_id>/team_dashboard/*.json` | frontend-team | Downstream dashboard consumers read these exports |
 | Design-time workspace registry | `workspaces/`, `backend.api.workspaces` | control-plane-team | Control-plane editing and registry-style management |
 | Gateway WebSocket transport | `backend.services.gateway` on `:8765` | runtime-team | Live event streaming contract for frontend |
@@ -50,8 +49,8 @@ in use.
 
 | Surface | Location | Replacement | ETA |
 |---------|----------|-------------|-----|
-| Legacy analyst agents | `backend.agents.analyst.*` | `EvoAgent` | After EvoAgent smoke tests pass |
-| Mixed workspace_id semantics | `/api/workspaces/{id}/agents/...` | ✅ `/api/runs/{run_id}/agents/...` routes added | Completed |
+| Legacy agent fallback paths | compatibility constructors in `backend.core.pipeline_runner` | `EvoAgent` | After fallback-free runtime cutover |
+| Mixed workspace_id semantics | removed from runtime agent routes | ✅ `/api/runs/{run_id}/agents/...` is the only runtime agent route | Completed |
 | Root-level runtime directories | `live/`, `backtest/`, `production/` | `runs/<run_id>/` | ✅ Removed, backed up to runs/_legacy/ |
 
 **Status**: Do not add new code using these surfaces. Migrate existing usage
@@ -67,43 +66,13 @@ topology and service-routing policy are frozen.
 |---------|---------------|-----------------|
 | OpenClaw integration | Gateway WebSocket (port 18789) | Stable — REST facade removed |
 | Env-dependent service fallbacks | `TRADING_SERVICE_URL`, `NEWS_SERVICE_URL` fallbacks to local modules | Remove fallbacks and require explicit URLs? |
-| Split-service production deploy | Docs show gateway-first, dev uses split-service | Align production with dev topology? |
+| Split-service deployment docs | Deployment docs are still partial compared with the dev topology | Rewrite deploy docs around split services? |
 
 **Status**: Document current behavior. Do not actively remove until topology
 decisions are finalized.
 
 ## Detailed Surface Documentation
 
-### Gateway-First Production Example
-
-**Files**:
-- `scripts/run_prod.sh` - Production launch script
-- `deploy/systemd/evotraders.service` - systemd unit
-- `deploy/nginx/bigtime.cillinn.com.conf` - HTTPS + WebSocket proxy
-- `deploy/nginx/bigtime.cillinn.com.http.conf` - HTTP variant
-
-**Behavior**:
-```bash
-# scripts/run_prod.sh launches:
-python3 -m backend.main \
-  --mode live \
-  --config-name production \
-  --host 127.0.0.1 \
-  --port 8765
-```
-
-**nginx proxies**:
-- `/ws` -> `127.0.0.1:8765` (WebSocket upgrade)
-- `/` -> static files in `/var/www/bigtime/current`
-
-**Why this exists**:
-- Simpler production deployment (single process + nginx)
-- WebSocket is the practical live event contract for frontend
-- Split-service topology adds operational complexity not needed for all deployments
-
-**Ownership**: ops-team
-**Status**: Stable and intentional
-
 ### Dashboard Export Layer
 
 **Files**: `runs/<run_id>/team_dashboard/*.json`
@@ -154,19 +123,14 @@ These remain in the tree, but they should not define the architecture for new wo
 - treat these as historical or compatibility-oriented data/layout artifacts
 - do not use them as the default runtime contract for new features
 
-### Mixed `workspace_id` semantics on agent routes
+### Historical mixed `workspace_id` semantics on agent routes
 
-- `/api/workspaces/{workspace_id}/agents/...`
+This compatibility shape has been removed from runtime agent routes.
 
-**Read**:
+**Current rule**:
 
-- design-time CRUD routes use `workspace_id` as a registry workspace id
-- profile, skills, and editable file routes use `workspace_id` as a run id
-
-**Mitigation already in repo**:
-
-- `agent_service /api/status` exposes scope metadata
-- runtime-read responses expose `scope_type` and `scope_note`
+- design-time CRUD routes use `/api/workspaces/{workspace_id}/agents/...`
+- runtime agent assets use `/api/runs/{run_id}/agents/...`
 
 ### Partial EvoAgent rollout
 
@@ -175,8 +139,10 @@ These remain in the tree, but they should not define the architecture for new wo
 
 **Read**:
 
-- EvoAgent is still a controlled rollout path
-- legacy analyst/risk/PM implementations remain the default runtime path for now
+- EvoAgent is the default selection path for supported roles in the current
+  pipeline
+- legacy implementations remain as compatibility fallbacks in selected startup
+  and runner paths
 
 ## Recommended Usage
 
@@ -185,13 +151,14 @@ When in doubt:
 1. trust `docs/current-architecture.md`
 2. trust `runs/<run_id>/` over root-level runtime directories
 3. treat `workspaces/` as control-plane registry, not runtime truth
-4. treat deploy artifacts as the current checked-in example, not the full system contract
+4. treat deploy artifacts as partial references, not the full system contract
 5. check this file's **Compatibility Surface Classification** before assuming something is legacy
+6. prefer `runtime_service`-managed startup for all new work
 
 ## Change Log
 
 | Date | Change |
 |------|--------|
 | 2026-03-31 | Added Compatibility Surface Classification (3 buckets) |
-| 2026-03-31 | Documented OpenClaw dual integration (REST vs WebSocket) |
+| 2026-03-31 | Clarified OpenClaw integration documentation |
 | 2026-03-31 | Added ownership and status to all surfaces |

docs/project-layout.md

@@ -0,0 +1,80 @@
+# Project Layout
+
+这个文件只描述当前仓库目录的职责划分，不记录历史迁移过程。
+
+如果你想知道“系统现在怎么工作”，优先看 `current-architecture.md`。
+如果你想知道“这个目录现在应该怎么理解”，看本文。
+
+## 顶层目录
+
+### 主线代码
+
+- `backend/`
+  - 后端运行时、服务、API、Pipeline、Agent、工具与测试
+- `frontend/`
+  - React/Vite 前端
+- `shared/`
+  - 前后端共享 schema 与 client 封装
+
+### 当前有效的数据与状态目录
+
+- `runs/`
+  - 运行态真相源
+  - 每个 `run_id` 下保存 BOOTSTRAP、agent 资产、state、logs、dashboard 导出
+- `workspaces/`
+  - 设计时注册表
+  - 用于 control-plane CRUD，不是默认运行时执行目录
+- `data/`
+  - 项目使用的数据资产与本地数据文件
+
+### 文档与部署
+
+- `docs/`
+  - 当前仍维护的架构、兼容边界、路线图、目录说明
+- `services/`
+  - 服务边界说明
+- `deploy/`
+  - 部署示例、systemd、nginx 配置
+- `scripts/`
+  - 启动、检查、烟测与辅助脚本
+
+### 项目入口与元数据
+
+- `README.md`
+  - 英文主说明
+- `README_zh.md`
+  - 中文主说明
+- `pyproject.toml`
+  - Python 项目元数据与依赖
+- `start-dev.sh`
+  - 本地默认开发入口
+- `start.sh`
+  - 偏生产风格的本地启动入口
+
+## 本地环境目录
+
+这些目录通常只对当前机器有效，不应作为架构事实源：
+
+- `.venv/`
+  - 本地 Python 虚拟环境
+- `.pydeps/`
+  - 本地依赖落地目录
+- `.omc/`
+  - 本地工具状态
+- `.codex/`
+  - 本地代理/工具状态
+
+## 阅读原则
+
+- 看运行时行为，优先从 `backend/`、`frontend/`、`runs/` 开始
+- 看控制面编辑与设计态资产，优先看 `workspaces/`
+- 看服务边界，优先看 `services/README.md`
+- 看目录时，不要把本地环境目录当成项目结构的一部分
+- 新增目录前，先判断它属于“主线代码 / 运行态数据 / 文档部署 / 本地环境”中的哪一类
+
+## 当前约束
+
+- 不再新增参考仓、副本仓、样例快照目录到顶层
+- 不再把测试输出、PID、构建产物、缓存目录提交到仓库
+- 运行态相关的新文件优先放到 `runs/<run_id>/`
+- 设计态注册相关的新文件优先放到 `workspaces/`

docs/runtime-api-changes.md

@@ -1,8 +1,26 @@
-# Runtime Service API 变更文档
+# Runtime Service API 说明
 
 ## 概述
 
-本文档描述了 `runtime_service` API 的改进，包括新增端点、增强的响应字段和改进的错误处理。
+本文档描述当前 `runtime_service` API 的关键端点、增强响应字段和启动/健康检查相关行为。
+它应被视为 runtime API 的补充说明，而不是迁移过程中的一次性变更记录。
+
+## 适用范围
+
+- 运行时控制面：`backend.apps.runtime_service`
+- 运行时 API 前缀：`/api/runtime/*`
+- 当前推荐配套阅读：
+  - `README.md`
+  - `README_zh.md`
+  - `docs/current-architecture.md`
+
+## 关于 `schedule_mode` 的当前约定
+
+当前对外约定统一使用 `daily` / `interval`。
+
+- `interval` 是当前公开的盘中轮询名称
+- `intraday` 仅作为向后兼容输入别名保留
+- runtime API 的响应和写回配置应优先输出 `interval`
 
 ## 新增端点
 
@@ -26,7 +44,8 @@
 - `mode`: 运行模式，`"live"`（实盘）或 `"backtest"`（回测），运行时停止时为 `"stopped"`
 - `is_backtest`: 是否为回测模式
 - `run_id`: 当前运行的任务 ID
-- `schedule_mode`: 调度模式，`"daily"` 或 `"intraday"`
+- `schedule_mode`: 调度模式字段。当前公开值为 `daily` 或 `interval`；
+  历史输入别名 `intraday` 会在服务端归一化为 `interval`。
 - `is_running`: Gateway 是否正在运行
 
 ---
@@ -113,7 +132,7 @@
 
 ---
 
-## 改进的端点
+## 关键端点说明
 
 ### GET /api/runtime/gateway/status
 
@@ -256,7 +275,8 @@ Gateway 启动前会自动验证以下配置：
    - `start_date` 必须早于 `end_date`
 
 6. **调度模式**
-   - `schedule_mode` 必须是 `"daily"` 或 `"intraday"`
+   - 当前公开校验值为 `"daily"` / `"interval"`
+   - 历史输入 `"intraday"` 会被兼容性归一化为 `"interval"`
 
 **验证失败响应**:
 

docs/terminology.md

@@ -51,19 +51,18 @@ Examples:
 
 ## Compatibility Rule
 
-Some API paths and fields still use legacy names:
+Some fields still use legacy names:
 
-- `/api/workspaces/{workspace_id}/agents/...`
 - `workspace_id` on approval records
 
-When reading those surfaces:
+Current API path rule:
 
-- design-time CRUD routes use `workspace_id` literally
-- runtime-read routes may use the same slot for `run_id`
+- design-time CRUD uses `/api/workspaces/{workspace_id}/...`
+- runtime agent assets use `/api/runs/{run_id}/agents/...`
 
 For new code:
 
-- prefer `runId` for runtime variables
+- prefer `runId` for runtime variables and runtime API paths
 - prefer `workspaceId` only for design-time registry flows
 
 ## UI Wording

frontend/README.md

@@ -24,7 +24,7 @@ That gives you:
 - trading service at `http://localhost:8001`
 - news service at `http://localhost:8002`
 - runtime service at `http://localhost:8003/api/runtime`
-- gateway WebSocket at `ws://localhost:8765`
+- gateway WebSocket at `ws://localhost:8765` started directly by `start-dev.sh`
 
 ## Frontend Environment Variables
 

frontend/src/components/RuntimeSettingsPanel.jsx

@@ -382,7 +382,7 @@ export default function RuntimeSettingsPanel({
                     }}
                   >
                     <option value="daily">每日定时</option>
-                    <option value="intraday">盘中轮询</option>
+                    <option value="interval">间隔轮询</option>
                   </select>
                 </label>
 

frontend/src/config/constants.js

@@ -154,6 +154,32 @@ export const WS_URL =
     ? `${FALLBACK_WS_PROTOCOL}//${FALLBACK_WS_HOST}:8765`
     : `${FALLBACK_WS_PROTOCOL}//${FALLBACK_WS_HOST}${FALLBACK_WS_PORT}/ws`);
 
+// Dynamic Team Management API
+const DEFAULT_DYNAMIC_TEAM_API_BASE = isLocalDevHost()
+  ? "http://localhost:8003/api/dynamic-team"
+  : `${DEFAULT_CONTROL_API_BASE}/dynamic-team`;
+export const DYNAMIC_TEAM_API_BASE =
+  trimTrailingSlash(import.meta.env.VITE_DYNAMIC_TEAM_API_BASE_URL || "") ||
+  DEFAULT_DYNAMIC_TEAM_API_BASE;
+
+// Dynamic Team API Endpoints
+export const DYNAMIC_TEAM_ENDPOINTS = {
+  // Get all available analyst types (builtin + runtime registered)
+  listTypes: () => `${DYNAMIC_TEAM_API_BASE}/types`,
+  // Get personas from personas.yaml
+  getPersonas: () => `${DYNAMIC_TEAM_API_BASE}/personas`,
+  // Create a new analyst
+  createAnalyst: (runId) => `${DYNAMIC_TEAM_API_BASE}/runs/${runId}/analysts`,
+  // Clone an existing analyst
+  cloneAnalyst: (runId) => `${DYNAMIC_TEAM_API_BASE}/runs/${runId}/analysts/clone`,
+  // Remove an analyst
+  removeAnalyst: (runId, agentId) => `${DYNAMIC_TEAM_API_BASE}/runs/${runId}/analysts/${agentId}`,
+  // Get analyst info
+  getAnalystInfo: (runId, agentId) => `${DYNAMIC_TEAM_API_BASE}/runs/${runId}/analysts/${agentId}`,
+  // Get team summary
+  getTeamSummary: (runId) => `${DYNAMIC_TEAM_API_BASE}/runs/${runId}/summary`,
+};
+
 // Initial ticker symbols for the production watchlist
 export const INITIAL_TICKERS = [
   { symbol: "AAPL", price: null, change: null },
@@ -170,3 +196,191 @@ export const INITIAL_TICKERS = [
   { symbol: "COIN", price: null, change: null }
 ];
 
+// ============================================
+// Dynamic Analyst Team Management
+// ============================================
+
+/**
+ * Built-in analyst types that can be used as base for dynamic analysts
+ *
+ * IMPORTANT: When creating dynamic analysts, the agent_id MUST end with '_analyst'
+ * to receive analysis tool groups (fundamentals, technical, sentiment, valuation tools).
+ * Example: 'crypto_specialist_analyst' (correct) vs 'crypto_specialist' (incorrect)
+ */
+export const BUILTIN_ANALYST_TYPES = [
+  {
+    typeId: "fundamentals_analyst",
+    name: "基本面分析师",
+    description: "Uses LLM to intelligently select analysis tools, focuses on financial data and company fundamental analysis",
+    icon: "fundamentals",
+  },
+  {
+    typeId: "technical_analyst",
+    name: "技术分析师",
+    description: "Uses LLM to intelligently select analysis tools, focuses on technical indicators and chart analysis",
+    icon: "technical",
+  },
+  {
+    typeId: "sentiment_analyst",
+    name: "情绪分析师",
+    description: "Uses LLM to intelligently select analysis tools, analyzes market sentiment and news sentiment",
+    icon: "sentiment",
+  },
+  {
+    typeId: "valuation_analyst",
+    name: "估值分析师",
+    description: "Uses LLM to intelligently select analysis tools, focuses on company valuation and value assessment",
+    icon: "valuation",
+  },
+];
+
+/**
+ * Default colors for dynamically created analysts
+ * Cycles through these colors for new analysts
+ */
+export const DYNAMIC_ANALYST_COLORS = [
+  { bg: "#F9FDFF", text: "#1565C0", accent: "#1565C0" },  // Blue
+  { bg: "#FFF8F8", text: "#C62828", accent: "#C62828" },  // Red
+  { bg: "#FAFFFA", text: "#2E7D32", accent: "#2E7D32" },  // Green
+  { bg: "#FCFAFF", text: "#6A1B9A", accent: "#6A1B9A" },  // Purple
+  { bg: "#FFFCF7", text: "#E65100", accent: "#E65100" },  // Orange
+  { bg: "#F9FEFF", text: "#00838F", accent: "#00838F" },  // Cyan
+  { bg: "#FFF9F5", text: "#D84315", accent: "#D84315" },  // Deep Orange
+  { bg: "#F5F5FF", text: "#4527A0", accent: "#4527A0" },  // Deep Purple
+];
+
+/**
+ * Generate a color scheme for a dynamic analyst based on index
+ * @param {number} index - The index of the analyst
+ * @returns {Object} Color scheme object
+ */
+export const getDynamicAnalystColors = (index) => {
+  return DYNAMIC_ANALYST_COLORS[index % DYNAMIC_ANALYST_COLORS.length];
+};
+
+/**
+ * Generate a default avatar URL for dynamic analysts
+ * Uses a hash of the agentId to select from available avatars
+ * @param {string} agentId - The agent ID
+ * @returns {string} Avatar URL
+ */
+export const getDynamicAnalystAvatar = (agentId) => {
+  const avatars = [
+    CDN_ASSETS.companyRoom.agent_1,
+    CDN_ASSETS.companyRoom.agent_2,
+    CDN_ASSETS.companyRoom.agent_3,
+    CDN_ASSETS.companyRoom.agent_4,
+    CDN_ASSETS.companyRoom.agent_5,
+    CDN_ASSETS.companyRoom.agent_6,
+  ];
+  // Simple hash function to consistently map agentId to an avatar
+  const hash = agentId.split("").reduce((acc, char) => {
+    return acc + char.charCodeAt(0);
+  }, 0);
+  return avatars[hash % avatars.length];
+};
+
+/**
+ * Create a dynamic analyst configuration object
+ * @param {Object} config - Configuration object
+ * @param {string} config.agentId - Unique identifier
+ * @param {string} config.baseType - Base analyst type (e.g., "technical_analyst")
+ * @param {string} config.name - Display name
+ * @param {string[]} config.focus - Focus areas
+ * @param {string} config.description - Description
+ * @param {number} index - Index for color assignment
+ * @returns {Object} Complete agent configuration
+ */
+export const createDynamicAnalystConfig = ({
+  agentId,
+  baseType,
+  name,
+  focus = [],
+  description = "",
+  index = 0,
+}) => {
+  return {
+    id: agentId,
+    name: name || agentId,
+    role: name || agentId,
+    baseType,
+    focus,
+    description,
+    avatar: getDynamicAnalystAvatar(agentId),
+    colors: getDynamicAnalystColors(index),
+    isDynamic: true,
+    isCustom: true,
+  };
+};
+
+/**
+ * Check if an agent is a dynamic analyst
+ * @param {Object} agent - Agent object
+ * @returns {boolean}
+ */
+export const isDynamicAnalyst = (agent) => {
+  return agent?.isDynamic === true || agent?.id?.includes("_");
+};
+
+/**
+ * Validate agent ID format for dynamic analysts
+ * @param {string} agentId - Agent ID to validate
+ * @returns {Object} Validation result
+ */
+export const validateAgentId = (agentId) => {
+  const errors = [];
+  const warnings = [];
+
+  if (!agentId) {
+    errors.push("Agent ID is required");
+  } else if (typeof agentId !== "string") {
+    errors.push("Agent ID must be a string");
+  } else {
+    if (agentId.length < 3) {
+      errors.push("Agent ID must be at least 3 characters");
+    }
+    if (agentId.length > 50) {
+      errors.push("Agent ID must be at most 50 characters");
+    }
+    if (!/^[a-zA-Z0-9_]+$/.test(agentId)) {
+      errors.push("Agent ID can only contain letters, numbers, and underscores");
+    }
+    // Reserved IDs that cannot be used
+    const reservedIds = ["portfolio_manager", "risk_manager"];
+    if (reservedIds.includes(agentId)) {
+      errors.push(`"${agentId}" is a reserved ID and cannot be used`);
+    }
+    // Warning: agent_id should end with '_analyst' to get analysis tools
+    if (!agentId.endsWith("_analyst")) {
+      warnings.push(
+        "Agent ID should end with '_analyst' to receive analysis tool groups"
+      );
+    }
+  }
+
+  return {
+    valid: errors.length === 0,
+    errors,
+    warnings,
+  };
+};
+
+/**
+ * Generate a suggested agent ID from a name
+ * IMPORTANT: Agent ID must end with '_analyst' to receive analysis tools
+ * @param {string} name - Display name
+ * @param {string} baseType - Base analyst type
+ * @returns {string} Suggested agent ID (guaranteed to end with '_analyst')
+ */
+export const suggestAgentId = (name, baseType) => {
+  const timestamp = Date.now().toString(36).slice(-4);
+  const normalized = name
+    .toLowerCase()
+    .replace(/[^a-z0-9\s]/g, "")
+    .replace(/\s+/g, "_")
+    .replace(/_analyst$/, "") // Remove '_analyst' suffix if present to avoid duplication
+    .slice(0, 20);
+  // Must end with '_analyst' to get analysis tools registered
+  return `${normalized || baseType}_${timestamp}_analyst`;
+};
+

frontend/src/hooks/useAgentDataRequests.js

@@ -49,11 +49,24 @@ export function useAgentDataRequests(clientRef) {
     return runId;
   }, []);
 
+  const sendWs = useCallback((payload) => {
+    const client = clientRef.current;
+    if (!client) {
+      return false;
+    }
+    return client.send(payload);
+  }, [clientRef]);
+
   const requestAgentSkills = useCallback((agentId) => {
     const normalized = typeof agentId === 'string' ? agentId.trim() : '';
     if (!normalized) return false;
     setIsAgentSkillsLoading(true);
     setAgentSkillsFeedback(null);
+    const sent = sendWs({ type: 'get_agent_skills', agent_id: normalized });
+    if (sent) {
+      return true;
+    }
+
     void resolveRunId()
       .then((runId) => fetchAgentSkills(runId, normalized))
       .then((payload) => {
@@ -61,22 +74,19 @@ export function useAgentDataRequests(clientRef) {
         setIsAgentSkillsLoading(false);
       })
       .catch(() => {
-        if (!clientRef.current) {
-          setIsAgentSkillsLoading(false);
-          return;
-        }
-        console.debug('REST agent skills request failed, falling back to websocket compatibility path');
-        const success = clientRef.current.send({ type: 'get_agent_skills', agent_id: normalized });
-        if (!success) {
-          setIsAgentSkillsLoading(false);
-        }
+        setIsAgentSkillsLoading(false);
       });
     return true;
-  }, [clientRef, resolveRunId, setAgentSkillsByAgent, setIsAgentSkillsLoading, setAgentSkillsFeedback]);
+  }, [resolveRunId, sendWs, setAgentSkillsByAgent, setIsAgentSkillsLoading, setAgentSkillsFeedback]);
 
   const requestAgentProfile = useCallback((agentId) => {
     const normalized = typeof agentId === 'string' ? agentId.trim() : '';
     if (!normalized) return false;
+    const sent = sendWs({ type: 'get_agent_profile', agent_id: normalized });
+    if (sent) {
+      return true;
+    }
+
     void resolveRunId()
       .then((runId) => fetchAgentProfile(runId, normalized))
       .then((payload) => {
@@ -85,20 +95,20 @@ export function useAgentDataRequests(clientRef) {
           [normalized]: payload?.profile && typeof payload.profile === 'object' ? payload.profile : {}
         }));
       })
-      .catch(() => {
-        if (clientRef.current) {
-          console.debug('REST agent profile request failed, falling back to websocket compatibility path');
-          clientRef.current.send({ type: 'get_agent_profile', agent_id: normalized });
-        }
-      });
+      .catch(() => {});
     return true;
-  }, [clientRef, resolveRunId, setAgentProfilesByAgent]);
+  }, [resolveRunId, sendWs, setAgentProfilesByAgent]);
 
   const requestSkillDetail = useCallback((skillName) => {
     const normalized = typeof skillName === 'string' ? skillName.trim() : '';
     if (!normalized) return false;
     const detailKey = `${selectedSkillAgentId}:${normalized}`;
     setSkillDetailLoadingKey(detailKey);
+    const sent = sendWs({ type: 'get_skill_detail', agent_id: selectedSkillAgentId, skill_name: normalized });
+    if (sent) {
+      return true;
+    }
+
     void resolveRunId()
       .then((runId) => fetchAgentSkillDetail(runId, selectedSkillAgentId, normalized))
       .then((payload) => {
@@ -110,18 +120,10 @@ export function useAgentDataRequests(clientRef) {
         setSkillDetailLoadingKey(null);
       })
       .catch(() => {
-        if (!clientRef.current) {
-          setSkillDetailLoadingKey(null);
-          return;
-        }
-        console.debug('REST skill detail request failed, falling back to websocket compatibility path');
-        const success = clientRef.current.send({ type: 'get_skill_detail', agent_id: selectedSkillAgentId, skill_name: normalized });
-        if (!success) {
-          setSkillDetailLoadingKey(null);
-        }
+        setSkillDetailLoadingKey(null);
       });
     return true;
-  }, [clientRef, resolveRunId, selectedSkillAgentId, setSkillDetailLoadingKey, setSkillDetailsByName]);
+  }, [resolveRunId, selectedSkillAgentId, sendWs, setSkillDetailLoadingKey, setSkillDetailsByName]);
 
   const handleCreateLocalSkill = useCallback((skillName) => {
     const normalized = typeof skillName === 'string' ? skillName.trim() : '';
@@ -131,6 +133,11 @@ export function useAgentDataRequests(clientRef) {
     }
     setAgentSkillsSavingKey(`${selectedSkillAgentId}:${normalized}:create`);
     setAgentSkillsFeedback(null);
+    const sent = sendWs({ type: 'create_agent_local_skill', agent_id: selectedSkillAgentId, skill_name: normalized });
+    if (sent) {
+      return;
+    }
+
     void resolveRunId()
       .then((runId) => createAgentLocalSkill(runId, selectedSkillAgentId, normalized))
       .then(() => {
@@ -140,19 +147,10 @@ export function useAgentDataRequests(clientRef) {
         requestSkillDetail(normalized);
       })
       .catch(() => {
-        if (!clientRef.current) {
-          setAgentSkillsSavingKey(null);
-          setAgentSkillsFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
-          return;
-        }
-        console.debug('REST local skill create failed, falling back to websocket compatibility path');
-        const success = clientRef.current.send({ type: 'create_agent_local_skill', agent_id: selectedSkillAgentId, skill_name: normalized });
-        if (!success) {
-          setAgentSkillsSavingKey(null);
-          setAgentSkillsFeedback({ type: 'error', text: '发送失败，请检查连接状态' });
-        }
+        setAgentSkillsSavingKey(null);
+        setAgentSkillsFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
       });
-  }, [clientRef, requestAgentSkills, requestSkillDetail, resolveRunId, selectedSkillAgentId, setAgentSkillsFeedback, setAgentSkillsSavingKey]);
+  }, [requestAgentSkills, requestSkillDetail, resolveRunId, selectedSkillAgentId, sendWs, setAgentSkillsFeedback, setAgentSkillsSavingKey]);
 
   const handleLocalSkillDraftChange = useCallback((skillName, content) => {
     const detailKey = `${selectedSkillAgentId}:${skillName}`;
@@ -165,6 +163,11 @@ export function useAgentDataRequests(clientRef) {
     if (typeof content !== 'string') return;
     setAgentSkillsSavingKey(`${selectedSkillAgentId}:${skillName}:content`);
     setAgentSkillsFeedback(null);
+    const sent = sendWs({ type: 'update_agent_local_skill', agent_id: selectedSkillAgentId, skill_name: skillName, content });
+    if (sent) {
+      return;
+    }
+
     void resolveRunId()
       .then((runId) => updateAgentLocalSkill(runId, selectedSkillAgentId, skillName, content))
       .then(() => {
@@ -173,23 +176,19 @@ export function useAgentDataRequests(clientRef) {
         requestSkillDetail(skillName);
       })
       .catch(() => {
-        if (!clientRef.current) {
-          setAgentSkillsSavingKey(null);
-          setAgentSkillsFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
-          return;
-        }
-        console.debug('REST local skill save failed, falling back to websocket compatibility path');
-        const success = clientRef.current.send({ type: 'update_agent_local_skill', agent_id: selectedSkillAgentId, skill_name: skillName, content });
-        if (!success) {
-          setAgentSkillsSavingKey(null);
-          setAgentSkillsFeedback({ type: 'error', text: '发送失败，请检查连接状态' });
-        }
+        setAgentSkillsSavingKey(null);
+        setAgentSkillsFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
       });
-  }, [clientRef, localSkillDraftsByKey, requestSkillDetail, resolveRunId, selectedSkillAgentId, setAgentSkillsFeedback, setAgentSkillsSavingKey]);
+  }, [localSkillDraftsByKey, requestSkillDetail, resolveRunId, selectedSkillAgentId, sendWs, setAgentSkillsFeedback, setAgentSkillsSavingKey]);
 
   const handleLocalSkillDelete = useCallback((skillName) => {
     setAgentSkillsSavingKey(`${selectedSkillAgentId}:${skillName}:delete`);
     setAgentSkillsFeedback(null);
+    const sent = sendWs({ type: 'delete_agent_local_skill', agent_id: selectedSkillAgentId, skill_name: skillName });
+    if (sent) {
+      return;
+    }
+
     void resolveRunId()
       .then((runId) => deleteAgentLocalSkill(runId, selectedSkillAgentId, skillName))
       .then(() => {
@@ -198,23 +197,19 @@ export function useAgentDataRequests(clientRef) {
         requestAgentSkills(selectedSkillAgentId);
       })
       .catch(() => {
-        if (!clientRef.current) {
-          setAgentSkillsSavingKey(null);
-          setAgentSkillsFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
-          return;
-        }
-        console.debug('REST local skill delete failed, falling back to websocket compatibility path');
-        const success = clientRef.current.send({ type: 'delete_agent_local_skill', agent_id: selectedSkillAgentId, skill_name: skillName });
-        if (!success) {
-          setAgentSkillsSavingKey(null);
-          setAgentSkillsFeedback({ type: 'error', text: '发送失败，请检查连接状态' });
-        }
+        setAgentSkillsSavingKey(null);
+        setAgentSkillsFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
       });
-  }, [clientRef, requestAgentSkills, resolveRunId, selectedSkillAgentId, setAgentSkillsFeedback, setAgentSkillsSavingKey]);
+  }, [requestAgentSkills, resolveRunId, selectedSkillAgentId, sendWs, setAgentSkillsFeedback, setAgentSkillsSavingKey]);
 
   const handleRemoveSharedSkill = useCallback((skillName) => {
     setAgentSkillsSavingKey(`${selectedSkillAgentId}:${skillName}:remove`);
     setAgentSkillsFeedback(null);
+    const sent = sendWs({ type: 'remove_agent_skill', agent_id: selectedSkillAgentId, skill_name: skillName });
+    if (sent) {
+      return;
+    }
+
     void resolveRunId()
       .then((runId) => disableAgentSkill(runId, selectedSkillAgentId, skillName))
       .then(() => {
@@ -223,24 +218,20 @@ export function useAgentDataRequests(clientRef) {
         requestAgentSkills(selectedSkillAgentId);
       })
       .catch(() => {
-        if (!clientRef.current) {
-          setAgentSkillsSavingKey(null);
-          setAgentSkillsFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
-          return;
-        }
-        console.debug('REST shared skill remove failed, falling back to websocket compatibility path');
-        const success = clientRef.current.send({ type: 'remove_agent_skill', agent_id: selectedSkillAgentId, skill_name: skillName });
-        if (!success) {
-          setAgentSkillsSavingKey(null);
-          setAgentSkillsFeedback({ type: 'error', text: '发送失败，请检查连接状态' });
-        }
+        setAgentSkillsSavingKey(null);
+        setAgentSkillsFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
       });
-  }, [clientRef, requestAgentSkills, resolveRunId, selectedSkillAgentId, setAgentSkillsFeedback, setAgentSkillsSavingKey]);
+  }, [requestAgentSkills, resolveRunId, selectedSkillAgentId, sendWs, setAgentSkillsFeedback, setAgentSkillsSavingKey]);
 
   const handleAgentSkillToggle = useCallback((skillName, enabled) => {
     const agentId = selectedSkillAgentId;
     setAgentSkillsSavingKey(`${agentId}:${skillName}`);
     setAgentSkillsFeedback(null);
+    const sent = sendWs({ type: 'update_agent_skill', agent_id: agentId, skill_name: skillName, enabled });
+    if (sent) {
+      return;
+    }
+
     void resolveRunId()
       .then((runId) => enabled
         ? enableAgentSkill(runId, agentId, skillName)
@@ -251,19 +242,10 @@ export function useAgentDataRequests(clientRef) {
         requestAgentSkills(agentId);
       })
       .catch(() => {
-        if (!clientRef.current) {
-          setAgentSkillsSavingKey(null);
-          setAgentSkillsFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
-          return;
-        }
-        console.debug('REST skill toggle failed, falling back to websocket compatibility path');
-        const success = clientRef.current.send({ type: 'update_agent_skill', agent_id: agentId, skill_name: skillName, enabled });
-        if (!success) {
-          setAgentSkillsSavingKey(null);
-          setAgentSkillsFeedback({ type: 'error', text: '发送失败，请检查连接状态' });
-        }
+        setAgentSkillsSavingKey(null);
+        setAgentSkillsFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
       });
-  }, [clientRef, requestAgentSkills, resolveRunId, selectedSkillAgentId, setAgentSkillsFeedback, setAgentSkillsSavingKey]);
+  }, [requestAgentSkills, resolveRunId, selectedSkillAgentId, sendWs, setAgentSkillsFeedback, setAgentSkillsSavingKey]);
 
   const handleSkillAgentChange = useCallback((agentId) => {
     setSelectedSkillAgentId(agentId);
@@ -278,6 +260,11 @@ export function useAgentDataRequests(clientRef) {
     if (!normalizedAgentId || !normalizedFilename) return false;
     setIsWorkspaceFileLoading(true);
     setWorkspaceFileFeedback(null);
+    const sent = sendWs({ type: 'get_agent_workspace_file', agent_id: normalizedAgentId, filename: normalizedFilename });
+    if (sent) {
+      return true;
+    }
+
     void resolveRunId()
       .then((runId) => fetchAgentWorkspaceFile(runId, normalizedAgentId, normalizedFilename))
       .then((payload) => {
@@ -292,18 +279,10 @@ export function useAgentDataRequests(clientRef) {
         setIsWorkspaceFileLoading(false);
       })
       .catch(() => {
-        if (!clientRef.current) {
-          setIsWorkspaceFileLoading(false);
-          return;
-        }
-        console.debug('REST workspace file read failed, falling back to websocket compatibility path');
-        const success = clientRef.current.send({ type: 'get_agent_workspace_file', agent_id: normalizedAgentId, filename: normalizedFilename });
-        if (!success) {
-          setIsWorkspaceFileLoading(false);
-        }
+        setIsWorkspaceFileLoading(false);
       });
     return true;
-  }, [clientRef, resolveRunId, setIsWorkspaceFileLoading, setWorkspaceDraftContent, setWorkspaceFileFeedback, setWorkspaceFilesByAgent]);
+  }, [resolveRunId, sendWs, setIsWorkspaceFileLoading, setWorkspaceDraftContent, setWorkspaceFileFeedback, setWorkspaceFilesByAgent]);
 
   const handleWorkspaceFileChange = useCallback((filename) => {
     useAgentStore.getState().setSelectedWorkspaceFile(filename);
@@ -314,6 +293,16 @@ export function useAgentDataRequests(clientRef) {
     const key = `${selectedSkillAgentId}:${selectedWorkspaceFile}`;
     setWorkspaceFileSavingKey(key);
     setWorkspaceFileFeedback(null);
+    const sent = sendWs({
+      type: 'update_agent_workspace_file',
+      agent_id: selectedSkillAgentId,
+      filename: selectedWorkspaceFile,
+      content: workspaceDraftContent
+    });
+    if (sent) {
+      return;
+    }
+
     void resolveRunId()
       .then((runId) => updateAgentWorkspaceFile(runId, selectedSkillAgentId, selectedWorkspaceFile, workspaceDraftContent))
       .then((payload) => {
@@ -328,24 +317,10 @@ export function useAgentDataRequests(clientRef) {
         }));
       })
       .catch(() => {
-        if (!clientRef.current) {
-          setWorkspaceFileSavingKey(null);
-          setWorkspaceFileFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
-          return;
-        }
-        console.debug('REST workspace file save failed, falling back to websocket compatibility path');
-        const success = clientRef.current.send({
-          type: 'update_agent_workspace_file',
-          agent_id: selectedSkillAgentId,
-          filename: selectedWorkspaceFile,
-          content: workspaceDraftContent
-        });
-        if (!success) {
-          setWorkspaceFileSavingKey(null);
-          setWorkspaceFileFeedback({ type: 'error', text: '发送失败，请检查连接状态' });
-        }
+        setWorkspaceFileSavingKey(null);
+        setWorkspaceFileFeedback({ type: 'error', text: '连接未就绪，稍后重试' });
       });
-  }, [clientRef, resolveRunId, selectedSkillAgentId, selectedWorkspaceFile, setWorkspaceFileFeedback, setWorkspaceFileSavingKey, setWorkspaceFilesByAgent, workspaceDraftContent]);
+  }, [resolveRunId, selectedSkillAgentId, selectedWorkspaceFile, sendWs, setWorkspaceFileFeedback, setWorkspaceFileSavingKey, setWorkspaceFilesByAgent, workspaceDraftContent]);
 
   const handleUploadExternalSkill = useCallback(async (file) => {
     if (!(file instanceof File)) {

frontend/src/hooks/useStockDataRequests.js

@@ -26,6 +26,14 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
   const { setOhlcHistoryByTicker, setPriceHistoryByTicker, setHistorySourceByTicker,
     setNewsByTicker, setInsiderTradesByTicker } = useMarketStore();
 
+  const sendWs = useCallback((payload) => {
+    const client = clientRef.current;
+    if (!client) {
+      return false;
+    }
+    return client.send(payload);
+  }, [clientRef]);
+
   const requestStockHistory = useCallback((symbol, { force = false } = {}) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
     if (!normalized) return false;
@@ -40,6 +48,13 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
     start.setDate(start.getDate() - 120);
     const startDate = start.toISOString().slice(0, 10);
 
+    const wsPayload = { type: 'get_stock_history', ticker: normalized, lookback_days: 120 };
+    const wsSent = sendWs(wsPayload);
+    if (wsSent) {
+      requestedStockHistoryRef.current.add(normalized);
+      return true;
+    }
+
     if (hasDirectTradingService()) {
       void fetchStockHistoryDirect(normalized, startDate, endDate)
         .then((payload) => {
@@ -59,42 +74,36 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
           setHistorySourceByTicker((prev) => ({ ...prev, [normalized]: 'trading_service' }));
         })
         .catch((error) => {
-          console.error('Direct stock-history fetch failed, falling back to websocket:', error);
-          if (clientRef.current) {
-            const success = clientRef.current.send({
-              type: 'get_stock_history',
-              ticker: normalized,
-              lookback_days: 120
-            });
-            if (success) requestedStockHistoryRef.current.add(normalized);
-          }
+          console.error('Direct stock-history fetch failed:', error);
         });
       requestedStockHistoryRef.current.add(normalized);
       return true;
     }
 
-    if (!clientRef.current) return false;
-    const success = clientRef.current.send({ type: 'get_stock_history', ticker: normalized, lookback_days: 120 });
-    if (success) requestedStockHistoryRef.current.add(normalized);
-    return success;
-  }, [clientRef, currentDate, setOhlcHistoryByTicker, setPriceHistoryByTicker, setHistorySourceByTicker]);
+    return false;
+  }, [currentDate, hasDirectTradingService, sendWs, setOhlcHistoryByTicker, setPriceHistoryByTicker, setHistorySourceByTicker]);
 
   const requestStockExplainEvents = useCallback((symbol) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
-    if (!normalized || !clientRef.current) return false;
-    return clientRef.current.send({ type: 'get_stock_explain_events', ticker: normalized });
-  }, [clientRef]);
+    if (!normalized) return false;
+    return sendWs({ type: 'get_stock_explain_events', ticker: normalized });
+  }, [sendWs]);
 
   const requestStockNews = useCallback((symbol) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
-    if (!normalized || !clientRef.current) return false;
-    return clientRef.current.send({ type: 'get_stock_news', ticker: normalized, lookback_days: 45, limit: 12 });
-  }, [clientRef]);
+    if (!normalized) return false;
+    return sendWs({ type: 'get_stock_news', ticker: normalized, lookback_days: 45, limit: 12 });
+  }, [sendWs]);
 
   const requestStockNewsForDate = useCallback((symbol, date) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
     if (!normalized || !date) return false;
 
+    const wsSent = sendWs({ type: 'get_stock_news_for_date', ticker: normalized, date, limit: 20 });
+    if (wsSent) {
+      return true;
+    }
+
     if (hasDirectNewsService()) {
       void fetchNewsForDateDirect(normalized, date, 20)
         .then((payload) => {
@@ -111,23 +120,19 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
           }));
         })
         .catch((error) => {
-          console.error('Direct news-for-date fetch failed, falling back to websocket:', error);
-          if (clientRef.current) {
-            clientRef.current.send({ type: 'get_stock_news_for_date', ticker: normalized, date, limit: 20 });
-          }
+          console.error('Direct news-for-date fetch failed:', error);
         });
       return true;
     }
 
-    if (!clientRef.current) return false;
-    return clientRef.current.send({ type: 'get_stock_news_for_date', ticker: normalized, date, limit: 20 });
-  }, [clientRef, setNewsByTicker]);
+    return false;
+  }, [hasDirectNewsService, sendWs, setNewsByTicker]);
 
   const requestStockNewsTimeline = useCallback((symbol) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
-    if (!normalized || !clientRef.current) return false;
-    return clientRef.current.send({ type: 'get_stock_news_timeline', ticker: normalized, lookback_days: 90 });
-  }, [clientRef]);
+    if (!normalized) return false;
+    return sendWs({ type: 'get_stock_news_timeline', ticker: normalized, lookback_days: 90 });
+  }, [sendWs]);
 
   const requestStockNewsCategories = useCallback((symbol) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
@@ -141,6 +146,11 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
     start.setDate(start.getDate() - 90);
     const startDate = start.toISOString().slice(0, 10);
 
+    const wsSent = sendWs({ type: 'get_stock_news_categories', ticker: normalized, lookback_days: 90 });
+    if (wsSent) {
+      return true;
+    }
+
     if (hasDirectNewsService()) {
       void fetchNewsCategoriesDirect(normalized, startDate, endDate, 200)
         .then((payload) => {
@@ -157,22 +167,23 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
           }));
         })
         .catch((error) => {
-          console.error('Direct news-categories fetch failed, falling back to websocket:', error);
-          if (clientRef.current) {
-            clientRef.current.send({ type: 'get_stock_news_categories', ticker: normalized, lookback_days: 90 });
-          }
+          console.error('Direct news-categories fetch failed:', error);
         });
       return true;
     }
 
-    if (!clientRef.current) return false;
-    return clientRef.current.send({ type: 'get_stock_news_categories', ticker: normalized, lookback_days: 90 });
-  }, [clientRef, currentDate, setNewsByTicker]);
+    return false;
+  }, [currentDate, hasDirectNewsService, sendWs, setNewsByTicker]);
 
   const requestStockInsiderTrades = useCallback((symbol, startDate = null, endDate = null) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
     if (!normalized) return false;
 
+    const wsSent = sendWs({ type: 'get_stock_insider_trades', ticker: normalized, start_date: startDate, end_date: endDate, limit: 50 });
+    if (wsSent) {
+      return true;
+    }
+
     if (hasDirectTradingService()) {
       void fetchInsiderTradesDirect(normalized, startDate, endDate, 50)
         .then((payload) => {
@@ -183,28 +194,29 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
           }));
         })
         .catch((error) => {
-          console.error('Direct insider-trades fetch failed, falling back to websocket:', error);
-          if (clientRef.current) {
-            clientRef.current.send({ type: 'get_stock_insider_trades', ticker: normalized, start_date: startDate, end_date: endDate, limit: 50 });
-          }
+          console.error('Direct insider-trades fetch failed:', error);
         });
       return true;
     }
 
-    if (!clientRef.current) return false;
-    return clientRef.current.send({ type: 'get_stock_insider_trades', ticker: normalized, start_date: startDate, end_date: endDate, limit: 50 });
-  }, [clientRef, setInsiderTradesByTicker]);
+    return false;
+  }, [hasDirectTradingService, sendWs, setInsiderTradesByTicker]);
 
   const requestStockTechnicalIndicators = useCallback((symbol) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
-    if (!normalized || !clientRef.current) return false;
-    return clientRef.current.send({ type: 'get_stock_technical_indicators', ticker: normalized });
-  }, [clientRef]);
+    if (!normalized) return false;
+    return sendWs({ type: 'get_stock_technical_indicators', ticker: normalized });
+  }, [sendWs]);
 
   const requestStockRangeExplain = useCallback((symbol, startDate, endDate, articleIds = []) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
     if (!normalized || !startDate || !endDate) return false;
 
+    const wsSent = sendWs({ type: 'get_stock_range_explain', ticker: normalized, start_date: startDate, end_date: endDate, article_ids: Array.isArray(articleIds) ? articleIds : [] });
+    if (wsSent) {
+      return true;
+    }
+
     if (hasDirectNewsService()) {
       void fetchRangeExplainDirect(normalized, startDate, endDate, articleIds)
         .then((payload) => {
@@ -224,22 +236,23 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
           }));
         })
         .catch((error) => {
-          console.error('Direct range explain fetch failed, falling back to websocket:', error);
-          if (clientRef.current) {
-            clientRef.current.send({ type: 'get_stock_range_explain', ticker: normalized, start_date: startDate, end_date: endDate, article_ids: Array.isArray(articleIds) ? articleIds : [] });
-          }
+          console.error('Direct range explain fetch failed:', error);
         });
       return true;
     }
 
-    if (!clientRef.current) return false;
-    return clientRef.current.send({ type: 'get_stock_range_explain', ticker: normalized, start_date: startDate, end_date: endDate, article_ids: Array.isArray(articleIds) ? articleIds : [] });
-  }, [clientRef, setNewsByTicker]);
+    return false;
+  }, [hasDirectNewsService, sendWs, setNewsByTicker]);
 
   const requestStockStory = useCallback((symbol, asOfDate = null) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
     if (!normalized) return false;
 
+    const wsSent = sendWs({ type: 'get_stock_story', ticker: normalized, as_of_date: asOfDate });
+    if (wsSent) {
+      return true;
+    }
+
     if (hasDirectNewsService()) {
       void fetchStockStoryDirect(normalized, asOfDate)
         .then((payload) => {
@@ -258,22 +271,23 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
           }));
         })
         .catch((error) => {
-          console.error('Direct story fetch failed, falling back to websocket:', error);
-          if (clientRef.current) {
-            clientRef.current.send({ type: 'get_stock_story', ticker: normalized, as_of_date: asOfDate });
-          }
+          console.error('Direct story fetch failed:', error);
         });
       return true;
     }
 
-    if (!clientRef.current) return false;
-    return clientRef.current.send({ type: 'get_stock_story', ticker: normalized, as_of_date: asOfDate });
-  }, [clientRef, setNewsByTicker]);
+    return false;
+  }, [hasDirectNewsService, sendWs, setNewsByTicker]);
 
   const requestStockSimilarDays = useCallback((symbol, date, topK = 8) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
     if (!normalized || !date) return false;
 
+    const wsSent = sendWs({ type: 'get_stock_similar_days', ticker: normalized, date, top_k: topK });
+    if (wsSent) {
+      return true;
+    }
+
     if (hasDirectNewsService()) {
       void fetchSimilarDaysDirect(normalized, date, topK)
         .then((payload) => {
@@ -291,21 +305,17 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
           }));
         })
         .catch((error) => {
-          console.error('Direct similar-days fetch failed, falling back to websocket:', error);
-          if (clientRef.current) {
-            clientRef.current.send({ type: 'get_stock_similar_days', ticker: normalized, date, top_k: topK });
-          }
+          console.error('Direct similar-days fetch failed:', error);
         });
       return true;
     }
 
-    if (!clientRef.current) return false;
-    return clientRef.current.send({ type: 'get_stock_similar_days', ticker: normalized, date, top_k: topK });
-  }, [clientRef, setNewsByTicker]);
+    return false;
+  }, [hasDirectNewsService, sendWs, setNewsByTicker]);
 
   const requestStockEnrich = useCallback((symbol, options = {}) => {
     const normalized = typeof symbol === 'string' ? symbol.trim().toUpperCase() : '';
-    if (!normalized || !clientRef.current) return false;
+    if (!normalized) return false;
     const startDate = typeof options.startDate === 'string' ? options.startDate.trim() : '';
     const endDate = typeof options.endDate === 'string' ? options.endDate.trim() : '';
     if (!startDate || !endDate) return false;
@@ -316,7 +326,7 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
         maintenanceStatus: { running: true, error: null, updatedAt: new Date().toISOString(), stats: null }
       }
     }));
-    return clientRef.current.send({
+    return sendWs({
       type: 'run_stock_enrich',
       ticker: normalized,
       start_date: startDate,
@@ -328,7 +338,7 @@ export function useStockDataRequests(clientRef, { setRequestStockHistory, setReq
       story_date: options.storyDate || null,
       target_date: options.targetDate || null
     });
-  }, [clientRef, setNewsByTicker]);
+  }, [sendWs, setNewsByTicker]);
 
   // Register request functions with WebSocket connection hook
   if (setRequestStockHistory) setRequestStockHistory(requestStockHistory);

frontend/src/hooks/useWebSocketConnection.js

@@ -652,6 +652,7 @@ export function useWebSocketConnection({
             type: 'success',
             text: `${agentId} ${e.enabled ? '已启用' : '已禁用'} ${skillName}`
           });
+          clientRef.current?.send({ type: 'get_agent_skills', agent_id: agentId });
         },
 
         agent_local_skill_created: (e) => {
@@ -660,6 +661,8 @@ export function useWebSocketConnection({
           setAgentSkillsSavingKey(null);
           if (!agentId || !skillName) return;
           setAgentSkillsFeedback({ type: 'success', text: `${agentId} 已创建本地技能 ${skillName}` });
+          clientRef.current?.send({ type: 'get_agent_skills', agent_id: agentId });
+          clientRef.current?.send({ type: 'get_skill_detail', agent_id: agentId, skill_name: skillName });
         },
 
         agent_local_skill_updated: (e) => {
@@ -668,6 +671,7 @@ export function useWebSocketConnection({
           setAgentSkillsSavingKey(null);
           if (!agentId || !skillName) return;
           setAgentSkillsFeedback({ type: 'success', text: `${agentId} 的本地技能 ${skillName} 已保存` });
+          clientRef.current?.send({ type: 'get_skill_detail', agent_id: agentId, skill_name: skillName });
         },
 
         agent_local_skill_deleted: (e) => {
@@ -686,6 +690,7 @@ export function useWebSocketConnection({
             return next;
           });
           setAgentSkillsFeedback({ type: 'success', text: `${agentId} 的本地技能 ${skillName} 已删除` });
+          clientRef.current?.send({ type: 'get_agent_skills', agent_id: agentId });
         },
 
         agent_skill_removed: (e) => {
@@ -694,6 +699,7 @@ export function useWebSocketConnection({
           setAgentSkillsSavingKey(null);
           if (!agentId || !skillName) return;
           setAgentSkillsFeedback({ type: 'success', text: `${agentId} 已移除共享技能 ${skillName}` });
+          clientRef.current?.send({ type: 'get_agent_skills', agent_id: agentId });
         },
 
         agent_workspace_file_loaded: (e) => {
@@ -716,6 +722,7 @@ export function useWebSocketConnection({
           const filename = typeof e.filename === 'string' ? e.filename.trim() : '';
           if (!agentId || !filename) return;
           setWorkspaceFileFeedback({ type: 'success', text: `${agentId} 的 ${filename} 已保存` });
+          clientRef.current?.send({ type: 'get_agent_workspace_file', agent_id: agentId, filename });
         },
 
         watchlist_updated: (e) => {

frontend/src/services/dynamicTeamApi.js

@@ -0,0 +1,147 @@
+/**
+ * Dynamic Team API Service
+ *
+ * Provides methods for managing analyst team dynamically:
+ * - Create new analysts with custom configuration
+ * - Clone existing analysts
+ * - Remove analysts
+ * - List available analyst types
+ * - Get analyst information
+ */
+import { DYNAMIC_TEAM_ENDPOINTS } from "../config/constants";
+
+/**
+ * Fetch wrapper with error handling
+ */
+async function fetchJson(url, options = {}) {
+  const response = await fetch(url, {
+    headers: {
+      "Content-Type": "application/json",
+    },
+    ...options,
+  });
+
+  if (!response.ok) {
+    const error = await response.text();
+    throw new Error(`API error: ${response.status} - ${error}`);
+  }
+
+  return response.json();
+}
+
+/**
+ * Get all available analyst types (builtin + runtime registered)
+ * @returns {Promise<Array>} List of analyst types
+ */
+export async function listAnalystTypes() {
+  return fetchJson(DYNAMIC_TEAM_ENDPOINTS.listTypes());
+}
+
+/**
+ * Get personas from personas.yaml
+ * @returns {Promise<Object>} Personas configuration
+ */
+export async function getPersonas() {
+  return fetchJson(DYNAMIC_TEAM_ENDPOINTS.getPersonas());
+}
+
+/**
+ * Create a new analyst
+ * @param {string} runId - The run configuration ID
+ * @param {Object} config - Analyst configuration
+ * @param {string} config.agent_id - Unique identifier
+ * @param {string} config.analyst_type - Base type or custom identifier
+ * @param {Object} [config.persona] - Custom persona definition
+ * @param {string} [config.soul_md] - Custom SOUL.md content
+ * @param {string} [config.agents_md] - Custom AGENTS.md content
+ * @param {string} [config.profile_md] - Custom PROFILE.md content
+ * @param {string} [config.model_name] - Override default model
+ * @param {string[]} [config.skills] - List of skill IDs
+ * @param {string[]} [config.tags] - Classification tags
+ * @returns {Promise<Object>} Creation result
+ */
+export async function createAnalyst(runId, config) {
+  return fetchJson(DYNAMIC_TEAM_ENDPOINTS.createAnalyst(runId), {
+    method: "POST",
+    body: JSON.stringify(config),
+  });
+}
+
+/**
+ * Clone an existing analyst
+ * @param {string} runId - The run configuration ID
+ * @param {Object} config - Clone configuration
+ * @param {string} config.source_id - Source analyst ID
+ * @param {string} config.new_id - New analyst ID
+ * @param {string} [config.name] - New display name
+ * @param {string[]} [config.focus_additions] - Additional focus areas
+ * @param {string} [config.description_override] - New description
+ * @param {string} [config.model_name] - Override model
+ * @returns {Promise<Object>} Clone result
+ */
+export async function cloneAnalyst(runId, config) {
+  return fetchJson(DYNAMIC_TEAM_ENDPOINTS.cloneAnalyst(runId), {
+    method: "POST",
+    body: JSON.stringify(config),
+  });
+}
+
+/**
+ * Remove a dynamically created analyst
+ * @param {string} runId - The run configuration ID
+ * @param {string} agentId - The analyst to remove
+ * @returns {Promise<Object>} Removal result
+ */
+export async function removeAnalyst(runId, agentId) {
+  return fetchJson(DYNAMIC_TEAM_ENDPOINTS.removeAnalyst(runId, agentId), {
+    method: "DELETE",
+  });
+}
+
+/**
+ * Get information about a specific analyst
+ * @param {string} runId - The run configuration ID
+ * @param {string} agentId - The analyst ID
+ * @returns {Promise<Object>} Analyst information
+ */
+export async function getAnalystInfo(runId, agentId) {
+  return fetchJson(DYNAMIC_TEAM_ENDPOINTS.getAnalystInfo(runId, agentId));
+}
+
+/**
+ * Get a summary of the current analyst team
+ * @param {string} runId - The run configuration ID
+ * @returns {Promise<Object>} Team summary
+ */
+export async function getTeamSummary(runId) {
+  return fetchJson(DYNAMIC_TEAM_ENDPOINTS.getTeamSummary(runId));
+}
+
+/**
+ * Hook for using dynamic team API in React components
+ * @param {string} runId - The run configuration ID
+ * @returns {Object} API methods
+ */
+export function useDynamicTeamApi(runId) {
+  return {
+    listTypes: () => listAnalystTypes(),
+    getPersonas: () => getPersonas(),
+    createAnalyst: (config) => createAnalyst(runId, config),
+    cloneAnalyst: (config) => cloneAnalyst(runId, config),
+    removeAnalyst: (agentId) => removeAnalyst(runId, agentId),
+    getAnalystInfo: (agentId) => getAnalystInfo(runId, agentId),
+    getTeamSummary: () => getTeamSummary(runId),
+  };
+}
+
+// Default export for convenience
+export default {
+  listAnalystTypes,
+  getPersonas,
+  createAnalyst,
+  cloneAnalyst,
+  removeAnalyst,
+  getAnalystInfo,
+  getTeamSummary,
+  useDynamicTeamApi,
+};

frontend/src/services/runtimeApi.js

@@ -133,10 +133,7 @@ function buildRunScopedAgentPath(runId, agentId, suffix = '') {
   return `/runs/${encodeURIComponent(runId)}/agents/${encodeURIComponent(agentId)}${suffix}`;
 }
 
-/**
- * Runtime agent routes use `/runs/{run_id}/agents/...`.
- * Legacy `/workspaces/...` routes are deprecated but remain for backward compatibility.
- */
+/** Runtime agent routes use `/runs/{run_id}/agents/...`. */
 export function fetchAgentProfile(runId, agentId) {
   return safeFetch(CONTROL_API_BASE, buildRunScopedAgentPath(runId, agentId, '/profile'));
 }

frontend/src/services/runtimeApi.test.js

@@ -20,7 +20,7 @@ describe('runtimeApi run-scoped agent routes', () => {
     await fetchAgentProfile('20260330_123000', 'portfolio_manager');
 
     expect(fetchMock).toHaveBeenCalledWith(
-      expect.stringContaining('/workspaces/20260330_123000/agents/portfolio_manager/profile')
+      expect.stringContaining('/runs/20260330_123000/agents/portfolio_manager/profile')
     );
   });
 
@@ -35,7 +35,7 @@ describe('runtimeApi run-scoped agent routes', () => {
     await updateAgentWorkspaceFile('20260330_123000', 'risk_manager', 'MEMORY.md', '# demo');
 
     expect(fetchMock).toHaveBeenCalledWith(
-      expect.stringContaining('/workspaces/20260330_123000/agents/risk_manager/files/MEMORY.md'),
+      expect.stringContaining('/runs/20260330_123000/agents/risk_manager/files/MEMORY.md'),
       expect.objectContaining({
         method: 'PUT',
         body: '# demo'

frontend/src/services/runtimeControls.js

@@ -64,13 +64,14 @@ export const buildRuntimeSummaryLabel = (runtimeConfig) => {
     return null;
   }
 
-  const scheduleMode = String(runtimeConfig.schedule_mode || "daily");
+  const rawScheduleMode = String(runtimeConfig.schedule_mode || "daily");
+  const scheduleMode = rawScheduleMode === "intraday" ? "interval" : rawScheduleMode;
   const intervalMinutes = Number(runtimeConfig.interval_minutes || 60);
   const triggerTime = String(runtimeConfig.trigger_time || "now");
   const maxCommCycles = Number(runtimeConfig.max_comm_cycles || 2);
 
-  if (scheduleMode === "intraday") {
-    return `调度 intraday / ${intervalMinutes}m / 讨论 ${maxCommCycles} 轮`;
+  if (scheduleMode === "interval") {
+    return `调度 interval / ${intervalMinutes}m / 讨论 ${maxCommCycles} 轮`;
   }
 
   if (triggerTime.toLowerCase() === "now") {

frontend/src/services/runtimeControls.test.js

@@ -51,9 +51,9 @@ describe("runtimeControls", () => {
     })).toBe("调度 daily / 09:30 ET / 讨论 3 轮");
 
     expect(buildRuntimeSummaryLabel({
-      schedule_mode: "intraday",
+      schedule_mode: "interval",
       interval_minutes: 15,
       max_comm_cycles: 2
-    })).toBe("调度 intraday / 15m / 讨论 2 轮");
+    })).toBe("调度 interval / 15m / 讨论 2 轮");
   });
 });

frontend/test-results/.last-run.json

@@ -1,4 +0,0 @@
-{
-  "status": "failed",
-  "failedTests": []
-}

scripts/run_prod.sh

@@ -1,34 +0,0 @@
-#!/usr/bin/env bash
-# COMPATIBILITY_SURFACE: stable
-# OWNER: ops-team
-# SEE: docs/legacy-inventory.md#gateway-first-production-example
-#
-# Gateway-first production launch script.
-# This is the current checked-in production example, running the gateway
-# directly and proxying /ws instead of exposing every split FastAPI service.
-# For split-service topology, see start-dev.sh and docs/current-architecture.md
-set -euo pipefail
-
-cd /root/code/evotraders
-
-export PYTHONPATH=/root/code/evotraders/.pydeps:.
-export TICKERS="${TICKERS:-AAPL,MSFT,GOOGL,AMZN,NVDA,META,TSLA,AMD,NFLX,AVGO,PLTR,COIN}"
-
-# 技能沙盒配置（生产环境建议使用 docker）
-export SKILL_SANDBOX_MODE="${SKILL_SANDBOX_MODE:-docker}"
-export SKILL_SANDBOX_IMAGE="${SKILL_SANDBOX_IMAGE:-python:3.11-slim}"
-export SKILL_SANDBOX_MEMORY_LIMIT="${SKILL_SANDBOX_MEMORY_LIMIT:-512m}"
-export SKILL_SANDBOX_CPU_LIMIT="${SKILL_SANDBOX_CPU_LIMIT:-1.0}"
-export SKILL_SANDBOX_NETWORK="${SKILL_SANDBOX_NETWORK:-none}"
-export SKILL_SANDBOX_TIMEOUT="${SKILL_SANDBOX_TIMEOUT:-60}"
-
-# "production" here is an explicit deployment run label, not a required
-# root-level runtime directory name.
-
-exec python3 -m backend.main \
-  --mode live \
-  --config-name production \
-  --host 127.0.0.1 \
-  --port 8765 \
-  --trigger-time now \
-  --poll-interval 15

scripts/verify_docs_consistency.py

@@ -1,168 +0,0 @@
-#!/usr/bin/env python3
-"""Verify documentation and script consistency.
-
-This script checks that:
-1. README.md mentions correct service ports
-2. start-dev.sh starts services on documented ports
-3. deploy/README.md is consistent with production scripts
-4. Service ports match across all documentation
-"""
-
-from __future__ import annotations
-
-import argparse
-import re
-import sys
-from pathlib import Path
-
-PROJECT_ROOT = Path(__file__).resolve().parents[1]
-
-# Expected service ports (source of truth)
-SERVICE_PORTS = {
-    "agent_service": 8000,
-    "trading_service": 8001,
-    "news_service": 8002,
-    "runtime_service": 8003,
-    "gateway_websocket": 8765,
-}
-
-
-def check_readme_ports() -> list[str]:
-    """Check that README.md documents correct ports."""
-    errors = []
-    readme_path = PROJECT_ROOT / "README.md"
-    readme_content = readme_path.read_text(encoding="utf-8")
-
-    # Check for each service port mention
-    for service, port in SERVICE_PORTS.items():
-        port_patterns = [
-            f":{port}",
-            f"port {port}",
-            f"localhost:{port}",
-        ]
-        found = any(pattern in readme_content for pattern in port_patterns)
-        if not found:
-            errors.append(f"README.md: Missing documentation for {service} on port {port}")
-
-    return errors
-
-
-def check_start_dev_sh_ports() -> list[str]:
-    """Check that start-dev.sh uses correct ports."""
-    errors = []
-    script_path = PROJECT_ROOT / "start-dev.sh"
-    script_content = script_path.read_text(encoding="utf-8")
-
-    # Check for port declarations in start_service calls
-    for service, port in SERVICE_PORTS.items():
-        if service == "gateway_websocket":
-            # Gateway uses --port flag
-            if f"--port {port}" not in script_content:
-                errors.append(f"start-dev.sh: Gateway not using port {port}")
-        else:
-            # Services use port parameter in start_service
-            pattern = rf'start_service\s+"{service}"\s+"[^"]+"\s+{port}'
-            if not re.search(pattern, script_content):
-                # Also check for explicit port mentions
-                if f"port {port}" not in script_content and f":{port}" not in script_content:
-                    errors.append(f"start-dev.sh: {service} not using port {port}")
-
-    return errors
-
-
-def check_deploy_readme_consistency() -> list[str]:
-    """Check that deploy/README.md is consistent with scripts."""
-    errors = []
-    deploy_readme_path = PROJECT_ROOT / "deploy" / "README.md"
-    deploy_content = deploy_readme_path.read_text(encoding="utf-8")
-
-    # Check for gateway port consistency
-    if "127.0.0.1:8765" not in deploy_content:
-        errors.append("deploy/README.md: Gateway port 8765 not documented correctly")
-
-    # Check for production script reference
-    if "scripts/run_prod.sh" not in deploy_content:
-        errors.append("deploy/README.md: Missing reference to scripts/run_prod.sh")
-
-    return errors
-
-
-def check_run_prod_sh_ports() -> list[str]:
-    """Check that run_prod.sh uses correct ports."""
-    errors = []
-    script_path = PROJECT_ROOT / "scripts" / "run_prod.sh"
-    script_content = script_path.read_text(encoding="utf-8")
-
-    # Production script should use port 8765 for gateway
-    if "--port 8765" not in script_content:
-        errors.append("scripts/run_prod.sh: Not using gateway port 8765")
-
-    return errors
-
-
-def check_service_main_blocks() -> list[str]:
-    """Check that service modules use correct ports in __main__ blocks."""
-    errors = []
-
-    service_files = {
-        "agent_service": PROJECT_ROOT / "backend" / "apps" / "agent_service.py",
-        "trading_service": PROJECT_ROOT / "backend" / "apps" / "trading_service.py",
-        "news_service": PROJECT_ROOT / "backend" / "apps" / "news_service.py",
-        "runtime_service": PROJECT_ROOT / "backend" / "apps" / "runtime_service.py",
-    }
-
-    for service, file_path in service_files.items():
-        if not file_path.exists():
-            errors.append(f"{service}: File not found at {file_path}")
-            continue
-
-        content = file_path.read_text(encoding="utf-8")
-        expected_port = SERVICE_PORTS[service]
-
-        # Check for port= in uvicorn.run or app.run
-        if f"port={expected_port}" not in content and f"port= {expected_port}" not in content:
-            errors.append(f"{file_path}: Not using expected port {expected_port}")
-
-    return errors
-
-
-def main() -> int:
-    parser = argparse.ArgumentParser(
-        description="Verify documentation and script consistency.",
-    )
-    parser.add_argument(
-        "--strict",
-        action="store_true",
-        help="Treat warnings as errors",
-    )
-    args = parser.parse_args()
-
-    all_errors = []
-
-    print("Checking README.md ports...")
-    all_errors.extend(check_readme_ports())
-
-    print("Checking start-dev.sh ports...")
-    all_errors.extend(check_start_dev_sh_ports())
-
-    print("Checking deploy/README.md consistency...")
-    all_errors.extend(check_deploy_readme_consistency())
-
-    print("Checking scripts/run_prod.sh ports...")
-    all_errors.extend(check_run_prod_sh_ports())
-
-    print("Checking service __main__ blocks...")
-    all_errors.extend(check_service_main_blocks())
-
-    if all_errors:
-        print("\nConsistency errors found:")
-        for error in all_errors:
-            print(f"  - {error}")
-        return 1 if args.strict else 0
-    else:
-        print("\nAll consistency checks passed!")
-        return 0
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())

services/README.md

@@ -20,31 +20,19 @@ and the next-step execution plan lives at
 | `backend.apps.trading_service` | `8001` | Read-only trading data APIs such as prices, financials, insider trades |
 | `backend.apps.news_service` | `8002` | Read-only explain/news APIs such as story, similar days, range explain |
 | `backend.apps.runtime_service` | `8003` | Runtime lifecycle APIs under `/api/runtime/*` |
-| Gateway (`backend.main`) | `8765` | WebSocket feed, runtime event stream, pipeline execution |
+| Gateway process | `8765` | WebSocket feed, runtime event stream, pipeline execution |
 | OpenClaw Gateway | `18789` | External OpenClaw WebSocket endpoint consumed by 大时代 gateway |
 
-## Runtime Modes
+## Runtime Mode
 
-### Standalone Mode (Direct Gateway Startup)
-
-For simple deployments or backward compatibility:
-
-```bash
-python -m backend.main --mode live --host 0.0.0.0 --port 8765
-```
-
-In this mode, Gateway runs as the primary process with all components
-(Pipeline, Market Service, Scheduler) loaded in-process.
-
-### Microservice Mode (Recommended)
-
-For development and production with service isolation:
+For development and service isolation:
 
 ```bash
 ./start-dev.sh
 ```
 
-This starts all services with `runtime_service` managing the Gateway lifecycle.
+Today this script starts the split services and then relies on
+`runtime_service` to spawn the Gateway during startup.
 
 ## What Runs By Default In Dev
 
@@ -60,11 +48,9 @@ That script starts:
 - `trading_service` on `8001`
 - `news_service` on `8002`
 - `runtime_service` on `8003`
-- 大时代 gateway on `8765` (as subprocess of runtime_service)
+- 大时代 gateway on `8765` (spawned by `runtime_service`)
 
-It does **not** start.
-
-Instead, the gateway expects an OpenClaw WebSocket server to already be
+It also expects an OpenClaw WebSocket server to already be
 available at `ws://localhost:18789` unless you override the OpenClaw gateway
 configuration outside the script.
 
@@ -79,8 +65,9 @@ python -m uvicorn backend.apps.news_service:app --host 0.0.0.0 --port 8002 --rel
 python -m uvicorn backend.apps.runtime_service:app --host 0.0.0.0 --port 8003 --reload
 ```
 
-The Gateway is started by `runtime_service` via the `/api/runtime/start` API,
-not manually. To start a runtime:
+If you launch only the split FastAPI surfaces manually, `runtime_service` can
+start the Gateway via the `/api/runtime/start` API. To start a runtime in that
+fully managed flow:
 
 ```bash
 curl -X POST http://localhost:8003/api/runtime/start \
@@ -94,11 +81,6 @@ curl -X POST http://localhost:8003/api/runtime/start \
   }'
 ```
 
-Optional OpenClaw REST surface:
-
-```bash
-```
-
 ## Runtime Responsibilities
 
 The runtime path is intentionally split:
@@ -159,7 +141,7 @@ architecture. Its responsibilities:
 ```
 runtime_service (:8003)
     |
-    |-- spawns --> Gateway subprocess (:8765)
+    |-- can spawn --> Gateway subprocess (:8765)
                        |
                        |-- TradingPipeline
                        |-- MarketService
@@ -167,8 +149,8 @@ runtime_service (:8003)
                        |-- WebSocket server
 ```
 
-The Gateway subprocess runs `backend.gateway_server` module (not `backend.main`)
-with run-specific configuration passed via CLI arguments.
+The Gateway subprocess runs `backend.gateway_server` with run-specific
+configuration passed by `runtime_service`.
 
 ## Environment Variables
 
@@ -221,13 +203,14 @@ Other flows still depend on the gateway WebSocket and control plane APIs.
 
 ## OpenClaw Integration Notes
 
-There are two separate OpenClaw integration surfaces in this repo:
+OpenClaw currently appears through two concrete access patterns in this repo:
 
 - OpenClaw WebSocket gateway on `:18789`
   - used directly by `backend/services/gateway.py`
   - this is what `start-dev.sh` assumes exists
-  - optional REST facade over OpenClaw CLI-backed reads
-  - useful for typed client access and service-level testing
+- selected CLI-backed read helpers behind gateway handlers
+  - used for a small set of history/status style reads
+  - not a separate standalone REST service surface
 
 Do not treat those as interchangeable in docs or deployment config.
 
@@ -256,8 +239,6 @@ backend.apps.news_service
 backend.gateway_server
   └─ Gateway subprocess entry point (run-scoped)
 
-backend.main
-  └─ standalone Gateway entry point (compatibility)
 ```
 
 ## Migration Boundaries
@@ -279,14 +260,14 @@ Also note the remaining naming split:
 
 - Pipeline logic lives in Gateway process
 - Gateway is spawned as subprocess by runtime_service
-- Standalone mode (`backend.main`) preserved for compatibility
+- Historical direct gateway startup has been removed from the repository
 
 ### Target State
 
 - Pipeline stages become independent services
 - Gateway becomes thin event router
 - runtime_service becomes full orchestrator
-- Standalone mode deprecated and removed
+- Keep converging on the split-service startup model
 
 See [docs/development-roadmap.md](../docs/development-roadmap.md) for detailed
 phase planning.

start-dev.sh

@@ -5,22 +5,16 @@
 #
 # 启动模式说明:
 # -------------
-# 本脚本支持两种启动模式:
-#
-# 1. 微服务模式 (默认) - 启动 4 个独立服务 + Gateway
-#    这是推荐的开发模式，各服务独立运行，便于单独调试和重启
-#    - agent_service    (端口 8000): Agent 生命周期管理
-#    - runtime_service  (端口 8003): 运行时配置和 Pipeline 执行
-#    - trading_service  (端口 8001): 市场数据和交易操作
-#    - news_service     (端口 8002): 新闻采集和富化
-#    - gateway          (端口 8765): WebSocket 网关，前端连接入口
-#
-# 2. 独立模式 (--standalone) - 仅启动 Gateway
-#    Gateway 内部会自行管理服务，适合快速验证或资源受限环境
+# 本脚本只支持当前开发主路径:
+# 启动 4 个独立服务 + 由 runtime_service 托管的 Gateway
+# - agent_service    (端口 8000): Agent 生命周期管理
+# - runtime_service  (端口 8003): 运行时配置和 Pipeline 执行
+# - trading_service  (端口 8001): 市场数据和交易操作
+# - news_service     (端口 8002): 新闻采集和富化
+# - gateway          (端口 8765): 由 runtime_service 拉起的 WebSocket 网关
 #
 # 用法:
-#   ./start-dev.sh              # 启动微服务模式
-#   ./start-dev.sh --standalone # 启动独立模式
+#   ./start-dev.sh              # 启动开发环境
 #   ./start-dev.sh --help       # 显示帮助信息
 #
 
@@ -51,9 +45,6 @@ readonly NC='\033[0m' # No Color
 # 进程 ID 数组
 PIDS=()
 
-# 启动模式: "microservices" 或 "standalone"
-MODE="microservices"
-
 # ============================================
 # 工具函数
 # ============================================
@@ -87,25 +78,18 @@ show_help() {
 大时代 Development Startup Script
 
 用法:
-  ./start-dev.sh [选项]
+ ./start-dev.sh [选项]
 
 选项:
-  --standalone    以独立模式启动（仅启动 Gateway，内部管理服务）
   --help, -h      显示此帮助信息
 
-模式说明:
-
-  微服务模式 (默认):
-    启动 4 个独立微服务 + Gateway，各服务独立进程，便于单独调试
-    - agent_service:    http://localhost:8000  (Agent 生命周期)
-    - trading_service:  http://localhost:8001  (市场数据)
-    - news_service:     http://localhost:8002  (新闻服务)
-    - runtime_service:  http://localhost:8003  (运行时管理)
-    - gateway:          ws://localhost:8765    (WebSocket 网关)
-
-  独立模式 (--standalone):
-    仅启动 Gateway，由 Gateway 内部自行管理服务
-    适合快速验证或资源受限环境
+开发模式:
+  启动 4 个独立微服务 + 托管 Gateway，各服务独立进程，便于单独调试
+  - agent_service:    http://localhost:8000  (Agent 生命周期)
+  - trading_service:  http://localhost:8001  (市场数据)
+  - news_service:     http://localhost:8002  (新闻服务)
+  - runtime_service:  http://localhost:8003  (运行时管理)
+  - gateway:          ws://localhost:8765    (由 runtime_service 托管)
 
 环境要求:
   - Python 3.9+
@@ -113,8 +97,7 @@ show_help() {
   - .env 文件 (可选但推荐)
 
 示例:
-  ./start-dev.sh              # 启动微服务模式
-  ./start-dev.sh --standalone # 启动独立模式
+  ./start-dev.sh              # 启动开发环境
 EOF
 }
 
@@ -125,10 +108,6 @@ EOF
 parse_args() {
     while [[ $# -gt 0 ]]; do
         case "$1" in
-            --standalone)
-                MODE="standalone"
-                shift
-                ;;
             --help|-h)
                 show_help
                 exit 0
@@ -249,13 +228,7 @@ check_env_file() {
 check_ports() {
     log_step "检查端口占用情况..."
 
-    local ports=()
-
-    if [[ "$MODE" == "microservices" ]]; then
-        ports=($AGENT_SERVICE_PORT $TRADING_SERVICE_PORT $NEWS_SERVICE_PORT $RUNTIME_SERVICE_PORT $GATEWAY_PORT)
-    else
-        ports=($GATEWAY_PORT)
-    fi
+    local ports=($AGENT_SERVICE_PORT $TRADING_SERVICE_PORT $NEWS_SERVICE_PORT $RUNTIME_SERVICE_PORT $GATEWAY_PORT)
 
     local occupied=()
     for port in "${ports[@]}"; do
@@ -340,16 +313,88 @@ start_service() {
     PIDS+=($!)
 }
 
-start_gateway() {
-    log_step "启动 Gateway (WebSocket 服务)..."
-    log_info "Gateway 将作为子进程启动 (端口 ${GATEWAY_PORT})"
-    log_info "前端连接地址: ws://localhost:${GATEWAY_PORT}"
+wait_for_runtime_service() {
+    log_step "等待 runtime_service 就绪..."
 
-    SERVICE_NAME="gateway" python -m backend.main \
-        --mode live \
-        --host 0.0.0.0 \
-        --port "$GATEWAY_PORT" &
-    PIDS+=($!)
+    local runtime_url="http://127.0.0.1:${RUNTIME_SERVICE_PORT}/health"
+    local attempts=30
+
+    for ((i=1; i<=attempts; i++)); do
+        if python - <<PY >/dev/null 2>&1; then
+import urllib.request
+with urllib.request.urlopen("${runtime_url}", timeout=1.5) as resp:
+    raise SystemExit(0 if resp.status == 200 else 1)
+PY
+            log_info "runtime_service 已就绪: ${runtime_url}"
+            return 0
+        fi
+        sleep 1
+    done
+
+    log_error "runtime_service 未在预期时间内就绪"
+    return 1
+}
+
+start_managed_runtime() {
+    log_step "通过 runtime_service 创建默认运行时..."
+
+    local runtime_api="http://127.0.0.1:${RUNTIME_SERVICE_PORT}/api/runtime/start"
+
+    if ! python - <<PY; then
+import json
+import os
+import sys
+import urllib.request
+
+tickers_env = os.getenv("TICKERS", "")
+tickers = [item.strip().upper() for item in tickers_env.split(",") if item.strip()]
+if not tickers:
+    tickers = ["AAPL", "MSFT", "GOOGL", "AMZN", "NVDA", "META", "TSLA", "AMD", "NFLX", "AVGO", "PLTR", "COIN"]
+
+def _env_int(name: str, default: int) -> int:
+    value = os.getenv(name, "").strip()
+    return int(value) if value else default
+
+def _env_float(name: str, default: float) -> float:
+    value = os.getenv(name, "").strip()
+    return float(value) if value else default
+
+payload = {
+    "launch_mode": "fresh",
+    "tickers": tickers,
+    "schedule_mode": os.getenv("SCHEDULE_MODE", "daily").strip() or "daily",
+    "interval_minutes": _env_int("INTERVAL_MINUTES", 60),
+    "trigger_time": os.getenv("TRIGGER_TIME", "09:30").strip() or "09:30",
+    "max_comm_cycles": _env_int("MAX_COMM_CYCLES", 2),
+    "initial_cash": _env_float("INITIAL_CASH", 100000.0),
+    "margin_requirement": _env_float("MARGIN_REQUIREMENT", 0.0),
+    "enable_memory": os.getenv("ENABLE_MEMORY", "").strip().lower() in {"1", "true", "yes", "on"},
+    "mode": os.getenv("RUNTIME_MODE", "live").strip() or "live",
+    "poll_interval": _env_int("POLL_INTERVAL", 10),
+}
+
+data = json.dumps(payload).encode("utf-8")
+req = urllib.request.Request(
+    "${runtime_api}",
+    data=data,
+    headers={"Content-Type": "application/json"},
+    method="POST",
+)
+
+try:
+    with urllib.request.urlopen(req, timeout=30) as resp:
+        body = json.loads(resp.read().decode("utf-8"))
+except Exception as exc:
+    print(f"FAILED: {exc}", file=sys.stderr)
+    raise
+
+print(json.dumps(body, ensure_ascii=False))
+PY
+        log_error "通过 runtime_service 创建运行时失败"
+        return 1
+    fi
+
+    log_info "默认运行时已创建，Gateway 将由 runtime_service 托管"
 }
 
 # ============================================
@@ -366,7 +411,7 @@ start_microservices_mode() {
     echo -e "  runtime_service:  http://localhost:${RUNTIME_SERVICE_PORT}"
     echo -e "  trading_service:  http://localhost:${TRADING_SERVICE_PORT}"
     echo -e "  news_service:     http://localhost:${NEWS_SERVICE_PORT}"
-    echo -e "  gateway:          ws://localhost:${GATEWAY_PORT}"
+    echo -e "  gateway:          ws://localhost:${GATEWAY_PORT} (由 runtime_service 拉起)"
     echo -e "${CYAN}==========================================${NC}"
     echo ""
 
@@ -390,8 +435,8 @@ start_microservices_mode() {
     start_service "trading_service" "backend.apps.trading_service:app" "$TRADING_SERVICE_PORT"
     start_service "news_service" "backend.apps.news_service:app" "$NEWS_SERVICE_PORT"
 
-    # 启动 Gateway（作为子进程）
-    start_gateway
+    wait_for_runtime_service
+    start_managed_runtime
 
     echo ""
     log_info "所有服务已启动"
@@ -399,30 +444,6 @@ start_microservices_mode() {
     echo ""
 }
 
-# ============================================
-# 独立模式启动
-# ============================================
-
-start_standalone_mode() {
-    log_step "启动独立模式..."
-    echo ""
-    echo -e "${CYAN}==========================================${NC}"
-    echo -e "${CYAN}  独立模式                               ${NC}"
-    echo -e "${CYAN}==========================================${NC}"
-    echo -e "  gateway:          ws://localhost:${GATEWAY_PORT}"
-    echo -e "${CYAN}==========================================${NC}"
-    echo ""
-    log_info "Gateway 将内部管理服务"
-
-    # 启动 Gateway（独立模式）
-    start_gateway
-
-    echo ""
-    log_info "Gateway 已启动（独立模式）"
-    log_info "按 Ctrl+C 停止服务"
-    echo ""
-}
-
 # ============================================
 # 清理与信号处理
 # ============================================
@@ -480,12 +501,7 @@ main() {
     echo -e "${GREEN}==========================================${NC}"
     echo ""
 
-    # 根据模式启动服务
-    if [[ "$MODE" == "standalone" ]]; then
-        start_standalone_mode
-    else
-        start_microservices_mode
-    fi
+    start_microservices_mode
 
     # 等待所有后台进程
     wait

start.sh

@@ -6,7 +6,6 @@
 #   ./start.sh                    # 构建前端 + 后台启动全部服务 (默认)
 #   ./start.sh --no-build         # 跳过前端构建
 #   ./start.sh --no-daemon        # 前台运行 (不使用 nohup)
-#   ./start.sh --gateway-only     # 仅启动 Gateway (配合 nginx)
 #   ./start.sh stop               # 停止所有后台服务
 #   ./start.sh status             # 查看服务状态
 #
@@ -37,14 +36,12 @@ FRONTEND_DIST="${SCRIPT_DIR}/frontend/dist"
 
 DAEMON=true
 BUILD_FRONTEND=true
-GATEWAY_ONLY=false
 ACTION="start"
 
 for arg in "$@"; do
     case "$arg" in
         --no-daemon)    DAEMON=false ;;
         --no-build)     BUILD_FRONTEND=false ;;
-        --gateway-only) GATEWAY_ONLY=true ;;
         stop)           ACTION="stop" ;;
         status)         ACTION="status" ;;
         *)              echo -e "${YELLOW}忽略未知参数: ${arg}${NC}" ;;
@@ -164,7 +161,7 @@ do_status() {
     print_status "runtime_service" 8003
     print_status "frontend" "${FRONTEND_PORT}"
     echo ""
-    echo -e "  ${CYAN}ℹ${NC}  Gateway 由 runtime_service 管理，通过前端启动任务触发"
+    echo -e "  ${CYAN}ℹ${NC}  Gateway 由 runtime_service 管理，运行日志写入 runs/<run_id>/logs/gateway.log"
     echo ""
 
     if [ -d "${FRONTEND_DIST}" ]; then
@@ -208,12 +205,10 @@ start_single_daemon() {
 }
 
 start_daemon() {
-    if ! ${GATEWAY_ONLY}; then
-        start_single_daemon "agent_service" "backend.apps.agent_service:app" 8000
-        start_single_daemon "trading_service" "backend.apps.trading_service:app" 8001
-        start_single_daemon "news_service" "backend.apps.news_service:app" 8002
-        start_single_daemon "runtime_service" "backend.apps.runtime_service:app" 8003
-    fi
+    start_single_daemon "agent_service" "backend.apps.agent_service:app" 8000
+    start_single_daemon "trading_service" "backend.apps.trading_service:app" 8001
+    start_single_daemon "news_service" "backend.apps.news_service:app" 8002
+    start_single_daemon "runtime_service" "backend.apps.runtime_service:app" 8003
 
     echo -e "  ${GREEN}▶${NC} frontend → http://0.0.0.0:${FRONTEND_PORT}"
     nohup env SERVICE_NAME="frontend" "${PYTHON}" -m uvicorn "backend.apps.frontend_service:app" \
@@ -231,7 +226,8 @@ start_daemon() {
     echo "  PID 目录: ${PID_DIR}/"
     echo ""
     echo "  查看状态: ./start.sh status"
-    echo "  查看日志: tail -f ${LOG_DIR}/gateway.log"
+    echo "  查看服务日志: tail -f ${LOG_DIR}/runtime_service.log"
+    echo "  查看运行日志: tail -f runs/<run_id>/logs/gateway.log"
     echo "  停止服务: ./start.sh stop"
     echo ""
 }
@@ -261,12 +257,10 @@ start_single_foreground() {
 start_foreground() {
     trap cleanup_foreground EXIT INT TERM
 
-    if ! ${GATEWAY_ONLY}; then
-        start_single_foreground "agent_service" "backend.apps.agent_service:app" 8000
-        start_single_foreground "trading_service" "backend.apps.trading_service:app" 8001
-        start_single_foreground "news_service" "backend.apps.news_service:app" 8002
-        start_single_foreground "runtime_service" "backend.apps.runtime_service:app" 8003
-    fi
+    start_single_foreground "agent_service" "backend.apps.agent_service:app" 8000
+    start_single_foreground "trading_service" "backend.apps.trading_service:app" 8001
+    start_single_foreground "news_service" "backend.apps.news_service:app" 8002
+    start_single_foreground "runtime_service" "backend.apps.runtime_service:app" 8003
 
     echo -e "  ${GREEN}▶${NC} frontend → http://0.0.0.0:${FRONTEND_PORT}"
     env SERVICE_NAME="frontend" "${PYTHON}" -m uvicorn "backend.apps.frontend_service:app" \
@@ -326,4 +320,3 @@ case "${ACTION}" in
         exit 1
         ;;
 esac
-