Quartz 5

开源模型质变：Claude Code 超级小白入门指南

Properties1
tagstutorial, ai, claude-code

3 min read

Summary

A beginner’s guide to Claude Code (CC) for non-developers, framing it as a general-purpose agent — not just a coding tool — that can do Q&A, writing, web building, data analysis, and even splitting payroll spreadsheets in 29 seconds. It walks through installing Node.js and CC, then advocates using domestic Chinese open-source models (GLM 4.7, MiniMax M2.1, Kimi K2) via a Coding plan to escape Anthropic’s account bans and pricing. It covers three core concepts (working folders as context, CLAUDE.md as long-term memory/constitution, and dangerous-permissions mode) plus loading prebuilt Skills, ending on the theme that open-source models bring “technical equality.”

一份寫給非開發者小白的 Claude Code（CC）入門指南，將 CC 定位為通用 Agent 而不只是寫代碼工具——可做問答、寫作、寫網頁、資料分析，甚至 29 秒拆分工資表。文章帶你安裝 Node.js 與 CC，並主張改用國產開源模型（GLM 4.7、MiniMax M2.1、Kimi K2）的 Coding 套餐，以避開 Anthropic 的封號與定價。內容涵蓋三個核心概念（工作資料夾作為上下文、CLAUDE.md 作為長期記憶／宪法、危險權限模式）以及載入現成 Skills，最後以開源模型帶來「技術平權」收尾。

Key Points

  • CC is positioned as a true general-purpose agent that operates your computer via natural conversation — no coding required.
  • Install path: prepare network access, install Node.js (and Git for Windows), then npm install -g @anthropic-ai/claude-code or the curl/irm install scripts; skip onboarding via "hasCompletedOnboarding": true in ~/.claude.json.
  • Recommends domestic open-source models with Coding plans (GLM 4.7 example via 智谱 open platform) configured through the @z_ai/coding-helper tool, citing aggressive Anthropic account bans.
  • Three core concepts: the working folder is CC’s “playground”/context (one folder per task), CLAUDE.md is its long-term memory/“constitution,” and dangerous mode (claude --dangerously-skip-permissions) trades safety for speed.
  • Practical tips: paste images with Ctrl+V (not Cmd+V) in the terminal; the author’s own launcher “Claude Code Now” auto-loads folders and dangerous mode.
  • Skills are reusable, pre-validated workflows you can copy from sites or have CC generate (references the official anthropics/skills repo); demoed with a frontend-design skill.

Insights

The article’s real thesis is economic, not technical: the “质变” (qualitative change) is that domestic open-source coding models finally crossed the capability threshold to run CC usably, which the author frames as “技术平权” (technical equality) — cloud vendors can deploy and sell them at ~1% of the price. The payroll-splitting demo is a deliberately mundane, non-coding example chosen to show CC’s value to ordinary office workers, reinforcing the “general agent” reframing. The candid security caveat around dangerous mode (specify the folder, keep backups) is worth heeding for any automation that grants unattended file access.

Connections

Raw Excerpt

现在国产开源模型都跟上来了，这正是我们期待已久的技术平权。

Graph View