<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>AI on 正能量</title><link>https://www.znl.net/tags/ai/</link><description>Recent content in AI on 正能量</description><generator>Hugo -- gohugo.io</generator><language>zh-CN</language><copyright>© 2026 正能量</copyright><lastBuildDate>Tue, 20 Oct 2026 12:27:42 +0800</lastBuildDate><atom:link href="https://www.znl.net/tags/ai/index.xml" rel="self" type="application/rss+xml"/><item><title>让 AI 编程任务持续推进：执行环境、状态交接与恢复设计</title><link>https://www.znl.net/posts/long-running-agent-harness/</link><pubDate>Tue, 24 Mar 2026 00:00:00 +0800</pubDate><guid>https://www.znl.net/posts/long-running-agent-harness/</guid><description>长任务不能只依赖一段越来越长的聊天记录。把任务拆成可验收的切片，保存代码版本与验证证据，才能在中断后找到可靠的下一步。</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://www.znl.net/posts/long-running-agent-harness/featured.png"/></item><item><title>AI Agent 评估怎么做：从一次演示到可靠完成任务</title><link>https://www.znl.net/posts/agent-evaluation-guide/</link><pubDate>Fri, 09 Jan 2026 00:00:00 +0800</pubDate><guid>https://www.znl.net/posts/agent-evaluation-guide/</guid><description>演示成功一次，不能说明 Agent 可以稳定交付。通过固定任务题集、核对最终状态、重复试验和隔离环境，才能发现退步并解释失败原因。</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://www.znl.net/posts/agent-evaluation-guide/featured.png"/></item><item><title>MCP 工具太多、结果太大怎么办：让代码处理中间数据</title><link>https://www.znl.net/posts/mcp-code-execution/</link><pubDate>Tue, 04 Nov 2025 00:00:00 +0800</pubDate><guid>https://www.znl.net/posts/mcp-code-execution/</guid><description>用只读构建统计流程说明如何把确定性处理交给代码，并让模型得到精简、完整且能追溯的结果。</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://www.znl.net/posts/mcp-code-execution/featured.png"/></item><item><title>把重复工作教给 AI：设计一份真正有用的 Agent Skill</title><link>https://www.znl.net/posts/agent-skills-guide/</link><pubDate>Thu, 16 Oct 2025 00:00:00 +0800</pubDate><guid>https://www.znl.net/posts/agent-skills-guide/</guid><description>将接口检查经验拆成入口说明、检查表和案例，让 AI 按需获取工作方法，并通过正反案例验证技能是否有效。</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://www.znl.net/posts/agent-skills-guide/featured.jpg"/></item><item><title>AI 为什么越聊越乱：用一次后端排障理解上下文工程</title><link>https://www.znl.net/posts/ai-context-engineering/</link><pubDate>Mon, 29 Sep 2025 00:00:00 +0800</pubDate><guid>https://www.znl.net/posts/ai-context-engineering/</guid><description>把 AI 的上下文当作一份需要维护的工作资料，设计可追溯的排障记录，并用证据检查回答质量。</description><media:content xmlns:media="http://search.yahoo.com/mrss/" url="https://www.znl.net/posts/ai-context-engineering/featured.png"/></item></channel></rss>