chore: remove legacy startup paths

.gitignore

@@ -51,11 +51,17 @@ node_modules
 outputs/
 /production/
 /smoke_test/
+/frontend/dist/
+/frontend/test-results/
 
 # Local tooling state
 .omc/
 /.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

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/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/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/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/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/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

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/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/`

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/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
-