What is 14days-build-claude-code-cli?
A 14-day hands-on tutorial that walks you through building a Claude Code-style coding agent CLI from scratch in Python. Instead of just explaining concepts, each day delivers a runnable snapshot of a CLI that grows incrementally: starting with a basic REPL, then adding real model calls, file tools, safe editing, bash permissions, session memory, and eventually hooks, skills, and subagents. The course also ships a polished web version at buildcc.dev with interactive diagrams, terminal replays, and code diffs built into the lessons. The default setup uses DeepSeek's Anthropic-compatible API, but any provider following the Anthropic Messages protocol works.
Why is it gaining traction?
The hook is clear: if you want to understand how Claude Code actually works under the hood, this rewrites those ideas in Python with teaching-friendly simplification rather than just translating TypeScript. The "harness" framing resonates with developers frustrated by prompt tutorials -- it explains that reliability comes from outside the model: permission gates, context management, diff previews, and feedback loops. The daily snapshot approach lets you verify each step with a single command, and the web tutorial's visual components (Agent Loop players, Harness Movies) make the agent lifecycle inspectable rather than a black box.
Who should use this?
Backend and full-stack developers who want to build or extend AI coding tools will get the most value. If you've been copying prompts without understanding how tool calling, permissions, or session state actually work in production, this gives you the mental model and working code. Researchers studying Agent architecture will appreciate the explicit focus on "harness engineering" versus model behavior. Python familiarity is required; understanding of CLI design and basic async concepts helps. This is not for beginners -- it's for engineers ready to build on top of LLMs rather than just prompting them.
Verdict
With only 30 stars and a 1.0% credibility score, this is an early-stage educational project: the first 7 days are solid and complete, while days 8-14 are locked. The web tutorial and daily runnable snapshots show thoughtful design, but test coverage and polish vary by day. Worth starting if you want to understand Code Agent internals hands-on, but treat it as structured self-study rather than production-ready code to deploy.
