feat: add Prompt template library, agent_call inter-agent tool, and RAG memory

- New PromptTemplatePicker component for browsing 13 preset prompt templates
- AgentConfig.vue: "Load from library" button for system prompt
- Agents.vue: "Create from Prompt template" entry with agent node + RAG memory
- seed_prompt_templates.py: 13 preset templates (客服/研发/教育/内容/分析/创意/健康医疗)
- agent_call tool: agents can delegate tasks to other agents (19th builtin tool)
- Created 全能助手 (general orchestrator) and 家庭医生助手 agents
- Switch template-created agents from type:llm to type:agent for full ReAct + RAG

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

backend/app/services/builtin_tools.py

@@ -2144,3 +2144,174 @@ SCHEDULE_DELETE_SCHEMA = {
         },
     },
 }
+
+
+# ── agent_call ─────────────────────────────────────────────────
+
+async def agent_call_tool(
+    agent_name: str,
+    query: str,
+    max_iterations: int = 10,
+) -> str:
+    """
+    调用另一个 Agent 处理任务并返回结果。
+
+    在数据库中按名称模糊匹配 Agent，用其 workflow_config 中 agent/llm 节点的
+    配置执行一次 ReAct 推理，然后将结果返回给调用方（如全能助手）。
+
+    Args:
+        agent_name: 目标 Agent 名称（支持模糊匹配，匹配到多个时取最接近的一个）
+        query: 发给目标 Agent 的用户消息
+        max_iterations: 最大推理步数（默认 10）
+    """
+    import asyncio as _asyncio
+
+    try:
+        from app.core.database import SessionLocal
+        from app.models.agent import Agent
+        from app.agent_runtime.core import AgentRuntime
+        from app.agent_runtime.schemas import (
+            AgentConfig,
+            AgentLLMConfig,
+            AgentToolConfig,
+        )
+
+        # 1. 查 DB，模糊匹配 Agent
+        db = SessionLocal()
+        try:
+            candidates = (
+                db.query(Agent)
+                .filter(Agent.name.like(f"%{agent_name}%"))
+                .limit(5)
+                .all()
+            )
+            if not candidates:
+                return json.dumps(
+                    {
+                        "error": "agent_not_found",
+                        "message": (
+                            f"未找到匹配「{agent_name}」的 Agent。"
+                            "请确认 Agent 名称是否正确，或在 Agent 管理中先创建目标 Agent。"
+                        ),
+                    },
+                    ensure_ascii=False,
+                )
+
+            # 精确匹配优先，否则取第一个
+            target = next(
+                (a for a in candidates if a.name == agent_name),
+                candidates[0],
+            )
+        finally:
+            db.close()
+
+        # 2. 从 workflow_config 提取 agent/llm 节点配置
+        wf = target.workflow_config or {}
+        nodes = wf.get("nodes", [])
+        agent_node = next(
+            (n for n in nodes if n.get("type") in ("agent", "llm")),
+            None,
+        )
+        if not agent_node:
+            return json.dumps(
+                {
+                    "error": "no_agent_node",
+                    "message": f"Agent「{target.name}」的工作流中未找到 Agent/LLM 节点，无法执行",
+                },
+                ensure_ascii=False,
+            )
+
+        nd = agent_node.get("data", {}) or {}
+
+        system_prompt = nd.get("system_prompt") or nd.get("prompt") or (
+            "你是一个有用的AI助手。"
+        )
+        model = nd.get("model", "deepseek-v4-flash")
+        provider = nd.get("provider", "deepseek")
+        temperature = float(nd.get("temperature", 0.7))
+        node_max_iter = int(nd.get("max_iterations") or 0)
+        if node_max_iter > 0:
+            max_iterations = min(max_iterations, node_max_iter)
+
+        # 3. 构建配置并执行
+        config = AgentConfig(
+            name=target.name or "sub_agent",
+            system_prompt=system_prompt,
+            llm=AgentLLMConfig(
+                provider=provider,
+                model=model,
+                temperature=temperature,
+                max_iterations=max_iterations,
+            ),
+            tools=AgentToolConfig(
+                include_tools=nd.get("tools") or [],
+                exclude_tools=nd.get("exclude_tools") or [],
+            ),
+            memory={
+                "enabled": nd.get("memory", True),
+                "persist_to_db": nd.get("memory", True),
+            },
+        )
+
+        runtime = AgentRuntime(config=config)
+        result = await runtime.run(query)
+
+        if result.success:
+            out = {
+                "agent": target.name,
+                "status": "success",
+                "iterations": result.iterations_used,
+                "tool_calls": result.tool_calls_made,
+                "reply": result.content,
+            }
+        else:
+            out = {
+                "agent": target.name,
+                "status": "error",
+                "error": result.error,
+                "reply": result.content or f"Agent 执行失败: {result.error}",
+            }
+
+        return json.dumps(out, ensure_ascii=False)
+
+    except Exception as e:
+        logger.error(f"agent_call 工具执行失败: {e}", exc_info=True)
+        return json.dumps(
+            {
+                "error": "execution_failed",
+                "message": f"调用 Agent 时出错: {e}",
+            },
+            ensure_ascii=False,
+        )
+
+
+AGENT_CALL_SCHEMA = {
+    "type": "function",
+    "function": {
+        "name": "agent_call",
+        "description": (
+            "调用另一个已注册的 Agent 来处理任务并返回结果。适合用来将子任务委托给"
+            "具备特定专长的 Agent（如家庭医生助手、代码助手等）。会话不会被接管，"
+            "结果会以文本形式返回给调用方用于整合回复。"
+        ),
+        "parameters": {
+            "type": "object",
+            "properties": {
+                "agent_name": {
+                    "type": "string",
+                    "description": "目标 Agent 名称，支持模糊匹配（如「家庭医生」「代码助手」）",
+                },
+                "query": {
+                    "type": "string",
+                    "description": "发给目标 Agent 的查询内容，可以包含上下文信息",
+                },
+                "max_iterations": {
+                    "type": "integer",
+                    "description": "最大推理步数（默认 10），控制 Agent 的思考深度",
+                    "default": 10,
+                },
+            },
+            "required": ["agent_name", "query"],
+        },
+    },
+}