此文章同时还提供英文版本 / English Version Availble In Medium:When an Agent Says “I’m Done” — 4 Claude Code Skills Born from Failure 基于大模型和 Coding Agent 能力的突飞猛进,最近几个月,我的「古法编程」率成功的由 10% 变成了 0% —— 基本没有一行代码是我手写的。然而生产力的提升却暴露出了现有大模型在长程代码开发上的很多短板:在 Opus 和 Sol 代替我通宵达旦的写下一行又一行代码的同时,它们也为我写出了一个又一个的逻辑陷阱和 BUG。在与 AI 不断的斗智斗勇和不断的翻车中,我沉淀出了几个可面向所有代码项目开发 skill,它们都诞生于一个共同的短板:一套在大模型看来「看起来成功」的代码,其实满是漏洞和缺陷。 在日常开发中,
当 agent 说”我做完了” —— 四个从翻车里长出来的 Claude Code skill
Find Your Interest
- 304 - Agent Skills - when to use them and why they matter
- I Was Drowning in Claude Code Sessions - Here’s How I Hacked My Way Out
- Bear Blog 的新 API / A new Bear Blog API
- Run AI Agents in TeamCity
- Claude’s Duet agent is a little neurotic and surprisingly helpful
- Why Are Coding Agents So Dumb?
- The Joining Tax: A Year of Governing Claude Code
- Agent skills in Xcode: How to install and use them today
- Agent skills in Xcode: How to install and use them today
- SwiftUI Agent Skill - Write better code with Claude, Codex, and other AI tools





