🎩 You're Invited:Meet the Socket team at Black Hat in Las Vegas, August 3-6.RSVP
Sign In

codojo

Package Overview
Dependencies
Maintainers
1
Versions
4
Alerts
File Explorer

Advanced tools

Socket logo

Install Socket

Detect and block malicious and high-risk dependencies

Install

codojo

AI-powered coding dojo for developers learning new languages — concept mapping, guided practice, and quizzing

latest
Source
npmnpm
Version
0.3.1
Version published
Weekly downloads
8
166.67%
Maintainers
1
Weekly downloads
 
Created
Source

codojo

codojo is an AI-powered coding dojo — a mentor for developers learning a new programming language or framework. Instead of dumping code on you, it teaches the way a great senior engineer would: it maps new concepts onto the languages you already know, asks questions to guide you to the answer, checks your understanding before moving on, and quizzes you interview-style as you go. It runs inside Claude Code, scaffolding a workspace whose CLAUDE.md turns Claude into your personal mentor.

Quickstart

npx codojo@latest init
cd ~/workspace/codojo   # or wherever you chose
claude

codojo init only scaffolds the workspace. The first time you run claude in it, the mentor walks you through a short onboarding interview — your background, the languages you already know, and what you want to learn — then gets to work.

If you use the GitHub CLI, codojo init --allow-gh-cli opts into a read-only set of gh lookups — see Permission model.

Screenshots of onboarding process

Get to know each other

Get to know each other

Provide 4-5 ways of learning your background

Provide 4-5 ways of learning your background

Parse your resume to learn your background

Parse your resume to learn your background

What do you want to learn or practice?

What do you want to learn or practice?

End with a summary and recording to profile.md and goals.md

End with a summary and recording to profile.md and goals.md

Screenshots of mentoring session

Where were we?

Where were we?

What should we work on in this session?

What should we work on in this session?

Learn about React components using existing knowledge as a guide

Learn about React components using existing knowledge as a guide

How it works

  • Mentor mode, not autocomplete. The generated CLAUDE.md instructs Claude to guide you to solutions and write code only rarely, and only after explaining the concept behind it.
  • Concept mapping. New ideas are anchored to what you already know — e.g. Rails routing in config/routes.rb becomes the bridge to how routing works in Go, Next.js, or PHP.
  • Your notes vs. the mentor's notes. notes/ and projects/ are yours and are read-only to the mentor; mentor_notes/ is where it logs sessions, tracks a concept map, and records quiz history to space your practice.
  • Interview-style quizzing. The mentor periodically checks understanding and quizzes you in a job-interview style, always letting you say "let's move on."

Workspace layout

<workspace>/
├── CLAUDE.md            # puts Claude into mentor mode
├── .claude/
│   └── settings.json    # file-permission boundaries (see below)
├── profile.md           # who you are (mentor edits only with your OK)
├── goals.md             # what you want to learn
├── notes/               # your notes — mentor read-only
├── projects/            # your code — mentor read-only
└── mentor_notes/        # mentor read/write
    ├── sessions/        # per-session summaries
    ├── topics/          # per-topic progress
    ├── quiz_history.md
    └── concept_map.md

Permission model

A generated workspace is confined two ways:

  • OS-level sandbox (the real boundary). .claude/settings.json declares a sandbox block that Claude Code enforces at the OS level (Seatbelt on macOS, bubblewrap on Linux/WSL2): the mentor's shell commands can only read inside the workspace, and all shell writes are denied. This holds regardless of what the mentor runs — it isn't just a guardrail on Claude's own tools.
  • Permission rules (tool-level). The same file keeps notes/ and projects/ read-only to the mentor and lets it write mentor_notes/, profile.md, and goals.md — but these rules bind only Claude's own file tools, not the shell subprocesses it spawns, so they are backed by the sandbox rather than relied on alone.

Because shell writes are denied workspace-wide, the mentor changes its own files through its Edit/Write tools and hands any write-needing shell command to you to run. Network tools are blocked too, with one opt-in exception:

  • codojo init --allow-gh-cli enables a closed, read-only set of GitHub CLI lookups (viewing/listing PRs, issues, and runs; searching; repo and status views). gh runs outside the sandbox (it can't complete TLS inside it), so the flag is a deliberate, opt-in widening of the boundary; mutating and gh auth commands still require your approval. Without the flag, gh is blocked entirely.

Requirements

Contributing

codojo is built with Spec-Driven Development.

Note that .claude/ (the spec-kit slash-command skills) is git-ignored, so after cloning you'll need to regenerate it locally with specify init . --integration claude --force. See CONTRIBUTING.md for the full setup, including an important caveat about not overwriting the committed .specify/ config.

License

MIT © 2026 Jason Noble

Keywords

ai

FAQs

Package last updated on 08 Jun 2026

Did you know?

Socket

Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.

Install

Related posts