freemcplab / 目錄 / 演練場 / Learn Claude Code

● 社群 shareAI-lab ⚡ 即開即用

Learn Claude Code

Name: Learn Claude Code (Claude Skill)
Author: shareAI-lab

作者 shareAI-lab · shareAI-lab/learn-claude-code

Reverse-engineered, readable harness for Claude Code in bash — the 'nand2tetris' version of an AI coding agent.

Learn Claude Code is a teaching artifact: a working bash-implemented agent harness that mimics Claude Code's loop. The skill packages it so you can install, run, and read along inside Claude itself.

▶ 觀看演示 📦 安裝 💡 使用場景

為什麼要用

核心特性

From-scratch bash harness
Tool-loop in readable shell
Skill packaging
Educational comments
Hackable

即時演示

實際使用效果

就緒

安裝

作者儲存庫 · Claude Skills 官方文件

使用場景

實戰用法： Learn Claude Code

Understand the Claude Code loop by reading 200 lines of bash

👤 Claude Code users ⏱ ~15 min intermediate

何時使用： You want to know what really happens between user prompts and tool calls.

步驟

Install

Clone the repo.✓ 已複製

→ Files local
Read

Follow main.sh.✓ 已複製

→ Mental model
Run

Run the harness; trace executions.✓ 已複製

→ Aha moments

結果： Real understanding of agent loops.

組合

與其他 MCP 搭配，撬動十倍槓桿

learn-claude-code-skill + nanobot-mcp

Bash harness for learning; Nanobot for a Python version of the same idea

Combine learn-claude-code-skill with nanobot-mcp: Bash harness for learning; Nanobot for a Python version of the same idea✓ 已複製

工具

此 MCP 暴露的能力

工具	輸入參數	何時呼叫	成本
run / trace	(see docs)	Two flows	1 call

成本與限制

運行它的成本

API 配額: N/A
每次呼叫 Token 數: Small
費用: Free
提示: Skip to main.sh — the rest is examples

安全

權限、密鑰、影響範圍

憑證儲存： None

資料出站： Configured LLM

切勿授予： Run with API keys you don't want to lose if the harness misbehaves

故障排查

常見錯誤與修復

Bash agent loops infinitely

It's educational — add a step cap at the top of main.sh

替代方案