
  <rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
      <title>ApFramework</title>
      <link>https://apframework.com/blog</link>
      <description>Shoukai 技术博客，软件架构、AI、领域驱动、开源软件、阅读笔记</description>
      <language>zh</language>
      <managingEditor>huangshoukai@yeah.net (Shoukai Huang)</managingEditor>
      <webMaster>huangshoukai@yeah.net (Shoukai Huang)</webMaster>
      <lastBuildDate>Thu, 04 Jun 2026 00:00:00 GMT</lastBuildDate>
      <atom:link href="https://apframework.com/tags/function-calling/feed.xml" rel="self" type="application/rss+xml"/>
      
  <item>
    <guid>https://apframework.com/blog/essay/2026-06-04-tool-call-not-function-call</guid>
    <title>Agent 的 Tool Call 不是 Function Call：它是一条生产权限请求</title>
    <link>https://apframework.com/blog/essay/2026-06-04-tool-call-not-function-call</link>
    <description>整个行业用 &quot;function calling&quot; 来描述模型调用工具，让所有开发者都下意识地把它当函数调用来处理。但工具不是函数——工具背后是数据库、文件系统、邮件服务、生产 API、Shell 和 MCP Server。每一次 Agent 的 tool call，本质上是一次资源授权请求，应该进入授权、审批、审计链路，而不是被 SDK 自动执行。本文是「AI Agent 安全最佳实践 2026」的续篇，专门拆解这个核心判断。</description>
    <pubDate>Thu, 04 Jun 2026 00:00:00 GMT</pubDate>
    <author>huangshoukai@yeah.net (Shoukai Huang)</author>
    <category>AI Agent</category><category>Security</category><category>Tool Call</category><category>Function Calling</category><category>Agent Architecture</category><category>Authorization</category>
  </item>

    </channel>
  </rss>
