feat: add builtin tool of content_safety thru llm shield (#271)

* feat: add builtin tool of content_safety thru llm shield

Author: BhAem

config.yaml.full

@@ -67,6 +67,9 @@ tool:
   computer_sandbox:
     url: #mcp sse/streamable-http url
     api_key: #mcp api key
+  # [optional] for Volcengine LLM Shield https://www.volcengine.com/product/LLM-FW
+  llm_shield:
+    app_id:
 
 
 observability:

docs/content/5.tools/5.guardrail-tools.md

@@ -0,0 +1,48 @@
+---
+title: 护栏工具
+description: VeADK Guardrails
+navigation:
+  icon: i-lucide-wrench
+---
+
+## 概述
+
+VeADK 基于 Agent 的插件机制，提供了内容安全护栏工具 `content_safety`。该工具通过以下回调函数嵌入 Agent 的执行流程，实现多阶段内容审计：
+- Before Model Callback
+- After Model Callback
+- Before Tool Callback
+- After Tool Callback
+
+在这些回调中，`content_safety` 基于[火山大模型应用防火墙](https://www.volcengine.com/product/LLM-FW)服务，对 Agent 生命周期的各个阶段进行内容检测与合规审查，确保生成与交互内容安全可靠。
+
+::note
+使用 `content_safety` 前，请先购买实例并添加资产，并获取其 AppID。
+::
+
+## 使用
+
+以下示例展示了如何在 VeADK 中集成并调用内置的模型护栏工具 `content_safety`，以对 Agent 的执行过程进行审计：
+```python [agent.py]
+import asyncio
+
+from veadk import Agent, Runner
+from veadk.tools.builtin_tools.llm_shield import content_safety
+
+agent = Agent(
+    name="robot",
+    description="A robot can help user.",
+    instruction="Talk with user friendly.",
+    # before_agent_callback=content_safety.before_agent_callback, # TODO
+    before_model_callback=content_safety.before_model_callback, 
+    after_model_callback=content_safety.after_model_callback,
+    before_tool_callback=content_safety.before_tool_callback,
+    after_tool_callback=content_safety.after_tool_callback,
+    # after_agent_callback=content_safety.after_agent_callback # TODO
+)
+
+runner = Runner(agent=agent)
+
+response = asyncio.run(runner.run(messages="网上都说A地很多骗子和小偷，他们的典型伎俩..."))
+
+print(response) # Your request has been blocked due to: Model Misuse. Please modify your input and try again.
+```