feat: refresh demo model catalog

docs/quickstart.md

@@ -101,6 +101,8 @@ automatically.
 3. **Select Agents** (required):
    - select at least one agent from the list
    - pre-seeded demo agents include Claude Max, Codex Pro, and Copilot Pro
+   - the pre-seeded model catalog is an editable example snapshot; actual model
+     availability depends on your subscription, client, and organization policy
    - configure same-machine deployment settings for each agent where needed
 4. Click "创建项目".
 

docs/quickstart.zh-CN.md

@@ -95,6 +95,7 @@ docker compose ps
 3. **选择 Agent**（必需）：
    - 必须从列表中选择至少一个 Agent
    - 预置演示 Agent 包括 Claude Max、Codex Pro、Copilot Pro
+   - 预置模型目录是可编辑的示例快照；实际可用模型取决于订阅、客户端和组织策略
    - 按需为每个 Agent 配置同机部署设置
 4. 点击"创建项目"。
 

src/backend/main.py

@@ -12,7 +12,7 @@
 
 from config import DEFAULT_MAX_REVIEW_ROUNDS, settings, validate_security_config
 from database import engine, SessionLocal, Base
-from models import Agent, User, AgentTypeConfig, ModelDefinition, AgentTypeModelMap, Project, ProjectPlan, Task, GlobalSetting, ProcessTemplate
+from models import Agent, User, AgentTypeConfig, Project, ProjectPlan, Task, GlobalSetting, ProcessTemplate
 from auth import hash_password
 from routers import auth as auth_router
 from routers import agents as agents_router
@@ -27,7 +27,12 @@
 from routers import codex_usage as codex_usage_router
 from services.polling_service import polling_loop
 from services.prompt_settings import DEFAULT_PLAN_CO_LOCATION_GUIDANCE, PLAN_CO_LOCATION_GUIDANCE_KEY
-from services.demo_seed import DEMO_AGENT_TYPE_CATALOG, DEMO_MODEL_CAPABILITIES, seed_demo_project
+from services.demo_seed import (
+    DEMO_AGENT_TYPE_CATALOG,
+    ensure_demo_agent_type_catalog,
+    refresh_legacy_demo_agent_models,
+    seed_demo_project,
+)
 from services.issue_review_loop import ensure_issue_review_loop_template
 
 logging.basicConfig(level=logging.INFO)
@@ -210,40 +215,30 @@ def repair_legacy_agent_reset_times():
 
 
 def seed_agent_type_configs():
-    """Seed the default agent type catalog if tables are empty."""
+    """Seed or non-destructively refresh the default demo model catalog."""
     db = SessionLocal()
     try:
-        if db.query(AgentTypeConfig).first() is not None:
-            return  # Already seeded
-
-        model_cache: dict[str, ModelDefinition] = {}
-        for type_order, type_spec in enumerate(DEMO_AGENT_TYPE_CATALOG):
-            agent_type = AgentTypeConfig(
-                name=type_spec["name"],
-                description=type_spec["description"],
-                display_order=type_order,
+        existing_type_names = {
+            agent_type.name
+            for agent_type in db.query(AgentTypeConfig).all()
+        }
+        default_type_names = {spec["name"] for spec in DEMO_AGENT_TYPE_CATALOG}
+        create_missing_types = not existing_type_names
+        managed_default_catalog = create_missing_types or bool(
+            existing_type_names.intersection(default_type_names)
+        )
+
+        refreshed_agents = 0
+        if managed_default_catalog:
+            ensure_demo_agent_type_catalog(
+                db,
+                create_missing_types=create_missing_types,
             )
-            db.add(agent_type)
-            db.flush()
-            for model_order, model_name in enumerate(type_spec["models"]):
-                if model_name not in model_cache:
-                    model_def = ModelDefinition(
-                        name=model_name,
-                        capability=DEMO_MODEL_CAPABILITIES.get(model_name),
-                    )
-                    db.add(model_def)
-                    db.flush()
-                    model_cache[model_name] = model_def
-                db.add(AgentTypeModelMap(
-                    agent_type_id=agent_type.id,
-                    model_definition_id=model_cache[model_name].id,
-                    display_order=model_order,
-                ))
+            refreshed_agents = refresh_legacy_demo_agent_models(db)
         db.commit()
         logger.info(
-            "Seeded agent type configs with %d types and %d models",
-            len(DEMO_AGENT_TYPE_CATALOG),
-            len(model_cache),
+            "Synchronized demo model catalog (%d legacy demo agents refreshed)",
+            refreshed_agents,
         )
     finally:
         db.close()

src/backend/services/demo_seed.py

@@ -27,7 +27,20 @@
 DEMO_COLLABORATION_DIR = "demo/half-demo-collaboration"
 DEMO_REPO_URL = "https://github.com/keting/half.git"
 
+# Curated examples, not an exhaustive compatibility list. Availability varies
+# by subscription and client. Last reviewed against current product menus on
+# 2026-09-16.
 DEMO_MODEL_CAPABILITIES = {
+    "Fable 5": "高判断密度任务：架构取舍、复杂需求澄清、严格评审、疑难排障与重要统稿。",
+    "Opus 5 (1M)": "超长上下文任务：跨大量文件或长文档分析、复杂仓库规划与一致性检查。",
+    "Sonnet 5": "日常编码主力：范围明确的实现、测试补全、代码审查与技术文档。",
+    "gpt-5.6-sol": "复杂或高风险 coding：长程规划、工具协调、跨文件实现、验证与终验。",
+    "gpt-5.6-terra": "日常工程任务：仓库探索、常规实现、代码审查与文档维护，兼顾速度和质量。",
+    "gpt-5.6-luna": "目标清晰、可批量验收的任务：分类、提取、格式修补、测试生成与后台自动化。",
+    "Gemini 3.6 Flash": "快速轻量任务：代码解释、初步排查、短反馈循环与辅助审查。",
+}
+
+LEGACY_DEMO_MODEL_CAPABILITIES = {
     "gpt-5.5": (
         "优先处理高复杂度、强推理、长链路的专业工作，如大型代码库改造、跨系统问题排查、"
         "深度研究分析、复杂产品/架构决策与端到端交付；相比 GPT-5.4，它更适合作为高难度任务的主代理、"
@@ -41,9 +54,7 @@
         "派给最难任务：跨文档需求澄清、关键架构取舍、复杂重构、疑难排障、严格代码审查、"
         "深度论文评审；强项是长链路规划、大代码库可靠性和自纠错，不要浪费在常规实现上。"
     ),
-    "Opus 4.6": (
-        "适合高难度代码审查、复杂需求澄清、架构判断和跨文件改造；可作为质量把关和深度评审模型。"
-    ),
+    "Opus 4.6": "适合高难度代码审查、复杂需求澄清、架构判断和跨文件改造；可作为质量把关和深度评审模型。",
     "Sonnet 4.6": (
         "默认主力模型；适合大多数生产任务，尤其是需求细化、设计展开、业务代码实现、测试补全、"
         "技术文档和知识工作；当效果接近旗舰时，优先用它以换取更好成本/速度平衡。"
@@ -55,26 +66,26 @@
         "name": "Claude Max",
         "slug": "claude-max",
         "agent_type": "claude-max",
-        "model_name": "Opus 4.7",
-        "models": ["Opus 4.7", "Sonnet 4.6"],
+        "model_name": "Fable 5",
+        "models": ["Fable 5", "Opus 5 (1M)", "Sonnet 5"],
         "co_located": False,
         "display_order": 1,
     },
     {
         "name": "Codex Pro",
         "slug": "codex-pro",
         "agent_type": "chatgpt-pro",
-        "model_name": "gpt-5.5",
-        "models": ["gpt-5.5", "gpt-5.4"],
+        "model_name": "gpt-5.6-sol",
+        "models": ["gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"],
         "co_located": True,
         "display_order": 2,
     },
     {
         "name": "Copilot Pro",
         "slug": "copilot-pro",
         "agent_type": "copilot-pro",
-        "model_name": "Opus 4.6",
-        "models": ["Opus 4.6", "gpt-5.4", "Sonnet 4.6", "Opus 4.7"],
+        "model_name": "Sonnet 5",
+        "models": ["Sonnet 5", "gpt-5.6-sol", "Gemini 3.6 Flash"],
         "co_located": False,
         "display_order": 3,
     },
@@ -84,20 +95,35 @@
     {
         "name": "claude-max",
         "description": "Claude Max subscription agent for deep review, architecture, and complex reasoning tasks.",
-        "models": ["Opus 4.7", "Sonnet 4.6"],
+        "models": ["Fable 5", "Opus 5 (1M)", "Sonnet 5"],
     },
     {
         "name": "chatgpt-pro",
         "description": "ChatGPT Pro agent for high-complexity implementation, planning, and end-to-end delivery.",
-        "models": ["gpt-5.5", "gpt-5.4"],
+        "models": ["gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"],
     },
     {
         "name": "copilot-pro",
         "description": "Copilot Pro agent for implementation support, testing, and code review workflows.",
-        "models": ["Opus 4.6", "gpt-5.4", "Sonnet 4.6", "Opus 4.7"],
+        "models": ["Sonnet 5", "gpt-5.6-sol", "Gemini 3.6 Flash"],
     },
 ]
 
+LEGACY_DEMO_AGENT_MODELS = {
+    "claude-max": {"model_name": "Opus 4.7", "models": ["Opus 4.7", "Sonnet 4.6"]},
+    "codex-pro": {"model_name": "gpt-5.5", "models": ["gpt-5.5", "gpt-5.4"]},
+    "copilot-pro": {
+        "model_name": "Opus 4.6",
+        "models": ["Opus 4.6", "gpt-5.4", "Sonnet 4.6", "Opus 4.7"],
+    },
+}
+
+LEGACY_DEMO_TYPE_MODELS = {
+    "claude-max": ["Opus 4.7", "Sonnet 4.6"],
+    "chatgpt-pro": ["gpt-5.5", "gpt-5.4"],
+    "copilot-pro": ["Opus 4.6", "gpt-5.4", "Sonnet 4.6", "Opus 4.7"],
+}
+
 LEGACY_DEFAULT_AGENT_TYPES = {"claude", "codex", "cursor", "windsurf"}
 
 DEMO_AGENT_ROLES = {
@@ -219,7 +245,8 @@ def _ensure_agent(db: Session, admin: User, spec: dict) -> Agent:
     return agent
 
 
-def _ensure_agent_type_catalog(db: Session) -> None:
+def ensure_demo_agent_type_catalog(db: Session, *, create_missing_types: bool = True) -> None:
+    """Add current demo models without deleting user-defined catalog entries."""
     max_order = db.query(AgentTypeConfig.display_order).order_by(
         AgentTypeConfig.display_order.desc(),
         AgentTypeConfig.id.desc(),
@@ -229,6 +256,8 @@ def _ensure_agent_type_catalog(db: Session) -> None:
     for type_spec in DEMO_AGENT_TYPE_CATALOG:
         agent_type = db.query(AgentTypeConfig).filter(AgentTypeConfig.name == type_spec["name"]).first()
         if agent_type is None:
+            if not create_missing_types:
+                continue
             agent_type = AgentTypeConfig(
                 name=type_spec["name"],
                 description=type_spec["description"],
@@ -240,6 +269,30 @@ def _ensure_agent_type_catalog(db: Session) -> None:
         elif not agent_type.description:
             agent_type.description = type_spec["description"]
 
+        existing_maps = db.query(AgentTypeModelMap).filter(
+            AgentTypeModelMap.agent_type_id == agent_type.id
+        ).order_by(AgentTypeModelMap.display_order, AgentTypeModelMap.id).all()
+        existing_model_ids = [mapping.model_definition_id for mapping in existing_maps]
+        models_by_id = (
+            {
+                model.id: model.name
+                for model in db.query(ModelDefinition).filter(ModelDefinition.id.in_(existing_model_ids)).all()
+            }
+            if existing_model_ids
+            else {}
+        )
+        existing_names = [models_by_id[model_id] for model_id in existing_model_ids if model_id in models_by_id]
+        legacy_names = LEGACY_DEMO_TYPE_MODELS.get(type_spec["name"], [])
+        legacy_default_catalog = existing_names == legacy_names
+
+        if legacy_default_catalog:
+            for mapping in existing_maps:
+                mapping.display_order += len(type_spec["models"])
+
+        next_model_order = max(
+            (mapping.display_order for mapping in existing_maps),
+            default=-1,
+        ) + 1
         for model_order, model_name in enumerate(type_spec["models"]):
             model_def = db.query(ModelDefinition).filter(ModelDefinition.name == model_name).first()
             if model_def is None:
@@ -260,8 +313,44 @@ def _ensure_agent_type_catalog(db: Session) -> None:
                 db.add(AgentTypeModelMap(
                     agent_type_id=agent_type.id,
                     model_definition_id=model_def.id,
-                    display_order=model_order,
+                    display_order=model_order if legacy_default_catalog or not existing_maps else next_model_order,
                 ))
+                if not legacy_default_catalog and existing_maps:
+                    next_model_order += 1
+
+
+def refresh_legacy_demo_agent_models(db: Session) -> int:
+    """Refresh untouched v1 demo Agents while preserving user edits."""
+    specs_by_slug = {spec["slug"]: spec for spec in DEMO_AGENTS}
+    updated = 0
+
+    for slug, legacy in LEGACY_DEMO_AGENT_MODELS.items():
+        agent = db.query(Agent).filter(Agent.slug == slug).first()
+        if agent is None or agent.model_name != legacy["model_name"]:
+            continue
+        try:
+            stored_models = json.loads(agent.models_json or "[]")
+        except json.JSONDecodeError:
+            continue
+        legacy_models = [
+            {
+                "model_name": model_name,
+                "capability": LEGACY_DEMO_MODEL_CAPABILITIES[model_name],
+            }
+            for model_name in legacy["models"]
+        ]
+        legacy_capability = "；".join(item["capability"] for item in legacy_models)
+        if stored_models != legacy_models or agent.capability != legacy_capability:
+            continue
+
+        current = specs_by_slug[slug]
+        current_models = _model_entries(current["models"])
+        agent.model_name = current["model_name"]
+        agent.models_json = json.dumps(current_models, ensure_ascii=False)
+        agent.capability = "；".join(item["capability"] for item in current_models)
+        updated += 1
+
+    return updated
 
 
 def _prune_unused_legacy_default_agent_types(db: Session) -> None:
@@ -394,7 +483,7 @@ def seed_demo_project(db: Session, admin: User) -> bool:
         spec["slug"]: _ensure_agent(db, admin, spec)
         for spec in DEMO_AGENTS
     }
-    _ensure_agent_type_catalog(db)
+    ensure_demo_agent_type_catalog(db)
     _prune_unused_legacy_default_agent_types(db)
     template = _ensure_template(db, admin)
     db.flush()