GitHub Repository

Self-guided workshop to build your own CLI coding agent like Claude Code from scratch in Python.

13 starsPython

A Self-Paced Exercise to Build a CLI Coding Agent from Scratch

Name: A Self-Paced Exercise to Build a CLI Coding Agent from Scratch
Availability: InStock
Author: primaprashant

by primaprashant·Feb 20, 2026·2 points·1 comment

Visit Project View on HN

AI Analysis

●●SolidRabbit HoleCozy

Seven-phase hands-on build of a CLI coding agent, but it's a learning workshop, not a product tool.

Strengths

•Pedagogical gold: forces you to implement LLM integration, tool use, and agent loop yourself—deep learning by doing
•Reference commits + hints prevent frustration without hand-holding; good scaffolding design
•Self-paced format replicates in-person Tokyo workshop to wider audience; accessible

Weaknesses

•Educational content, not a shipping product—won't replace Claude Code or Continue.dev
•Limited scope: basic file/bash tools only; agents built won't match production-grade systems

Post Description

I ran a hands-on workshop in Tokyo where ~50 engineers built a CLI coding agent from scratch in Python. I've converted it into a self-paced exercise.

GitHub Repo: https://github.com/primaprashant/alduin

We start from a skeleton repo with a basic input loop and no LLM. Over seven phases, we add the Anthropic API, implement tools (read file, edit file, bash), and build the core agent loop.

Each phase has hints and a reference commit if you get stuck.