cc-self-train

skill
Guvenlik Denetimi
Basarisiz
Health Uyari
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 6 GitHub stars
Code Basarisiz
  • child_process — Shell command execution capability in .claude/scripts/check-updates.js
  • execSync — Synchronous shell command execution in .claude/scripts/check-updates.js
  • process.env — Environment variable access in .claude/scripts/check-updates.js
  • fs module — File system access in .claude/scripts/learner-context.js
  • fs module — File system access in .claude/scripts/learner-streak-check.js
  • fs.rmSync — Destructive file system operation in .claude/scripts/observe-interaction.js
  • fs module — File system access in .claude/scripts/observe-interaction.js
  • execSync — Synchronous shell command execution in tests/smoke/onboarding.mjs
  • process.env — Environment variable access in tests/smoke/onboarding.mjs
  • rm -rf — Recursive force deletion command in tests/smoke/run-smoke.sh
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

Learn features by building projects hands-on with Claude Code--no boring videos, outdated blog posts, or stale courses.

README.md

Agentic Education:
Using Claude Code to Teach Claude Code

Paper v2.22.0

TL;DR

Developers face a widening gap between the availability of powerful AI coding agents and the learning pathways needed to use them effectively. cc-self-train is a modular interactive curriculum for learning Claude Code through hands-on project construction. The system introduces five contributions: (1) a persona progression model that adapts the AI instructor's tone across four stages (Guide → Collaborator → Peer → Launcher); (2) an adaptive learning system that observes engagement quality through hook-based heuristics and adjusts scaffolding at two timescales — streak detection for mid-module intervention and aggregate metrics for module-boundary persona changes; (3) a cross-domain unified curriculum in which five distinct project domains share identical feature sequencing; (4) a step-pacing mechanism with explicit pause primitives to manage information overload; and (5) an auto-updating curriculum design in which the onboarding agent detects upstream tool changes and updates teaching materials before instruction begins. A pilot evaluation with 27 participants shows statistically significant self-efficacy gains across all 10 assessed skill areas (p < 0.001), with the largest effects on advanced features such as hooks and custom skills.


The best way to learn Claude Code is to build something real with it. Pick a project, pick your language, and work through 10 hands-on modules that teach you every major Claude Code feature — not by reading about them, but by using them to ship actual code.

Quick Start

# 1. Install Claude Code (if you haven't already)
npm install -g @anthropic-ai/claude-code

# 2. Clone this repo
git clone https://github.com/suspicious-cow/cc-self-train.git
cd cc-self-train

# 3. Launch Claude Code
claude

# 4. When prompted to trust project hooks, approve them
#    (they show a welcome banner and check GitHub for
#     CC updates, nothing else — /hooks to review if curious)

# 5. Type this when Claude starts:
/start

That's it. Claude will walk you through picking a project, checking your dev environment, and scaffolding everything. You'll need an Anthropic API key or a Claude subscription (Pro, Max, or Team).

What are those hooks? This repo includes two small SessionStart hooks. The first shows the welcome banner above. The second pings GitHub to check if a newer version of Claude Code is available — if so, it tells you to run claude update. Both are read-only, open source, and visible in .claude/scripts/.

Who This Is For

You've installed Claude Code and maybe run /init. Now what? This is your answer. Pick one of 4 tutorial projects (or bring your own) and work through 10 progressive modules that take you from "first session" to "multi-agent orchestration." By the end, you won't just know what these features do — you'll have used every one of them to build something you're proud of.

No specific language required. Forge, Nexus, and Sentinel describe what to build, not how. You choose Python, TypeScript, Go, Rust, or whatever you're comfortable with. Canvas uses plain HTML/CSS/JS — no build tools needed.

Manual Setup (if you prefer)

Click to expand manual setup instructions

Install Claude Code

npm install -g @anthropic-ai/claude-code
claude --version

Set Up Your Dev Environment

Make sure your language toolchain is ready:

Language What you need Quick check
Python Python 3.10+, a package manager (conda, venv, uv, or pip) python --version
TypeScript/Node Node.js 18+, npm/pnpm/yarn node --version
Go Go 1.21+ go version
Rust Rust via rustup, cargo rustc --version
Other Whatever your language needs — Claude can help you set it up

Docker users: Any project can be done inside a container. Bring your own Dockerfile or ask Claude to generate one for your language.

Install Git

git --version

If you don't have it: git-scm.com/downloads

Pick a Project and Go

Each project covers all 10 modules. Pick based on what sounds fun to build.

The 5 Options

Canvas — Personal Portfolio Site ⭐ Recommended for first-timers

Build a multi-page portfolio site with responsive design, a blog, and a contact form using plain HTML, CSS, and JavaScript.

Why this project? Every developer needs a portfolio site but never gets around to building one. Canvas uses zero build tools — no npm, no frameworks, no compilers. Just open index.html in your browser. That means you spend 100% of your time learning Claude Code instead of fighting your toolchain. You'll walk away with a real, deployable site and mastery of every CC feature.

Best for: First-timers, anyone who wants the fastest path to learning CC


Forge — Personal Dev Toolkit

Build a command-line tool (a program you run from your terminal) for notes, snippets, bookmarks, and templates that grows into a searchable, pluggable knowledge base with API.

Why this project? Most tutorials build throwaway apps you'll never open again. Forge builds something you'll actually use every day — a personal tool that organizes your dev life. Notes from meetings, code snippets you keep re-Googling, project templates you set up the same way every time. By the end, you'll have a tool that saves you time and deep expertise in Claude Code.

Best for: Tool builders, "I want something I'll actually use after this course"


Nexus — Local API Gateway

Build a local server that sits between apps and manages their traffic — routing requests, limiting how fast clients can call, caching responses, and checking if services are healthy.

Why this project? Every production system has a gateway that manages traffic between services, but most developers treat it as a black box. By building one from scratch — routing, rate limiting, caching, health checks — you'll understand how services actually talk to each other at a level most developers never reach. If you've ever wondered what sits between a user's request and the server that handles it, this is your chance to find out by building it yourself.

Best for: Backend devs, anyone curious about infrastructure and how services connect


Sentinel — Code Analyzer & Test Generator

Build a tool that scans code for problems, generates tests automatically, and tracks how well-tested your code is — growing into a full quality dashboard.

Why this project? Sentinel is a tool that makes your other code better. It finds bugs before they ship, writes tests so you don't have to start from scratch, and tracks quality over time. If you care about code quality, this project teaches you how to enforce it automatically. It's the "meta-tool" — a program that improves every other program you write.

Best for: Quality-focused devs, "I want to build a tool that levels up everything else I write"


Bring Your Own Project -- For Experienced Developers

Already have a project you're working on? Learn every CC feature by applying it to YOUR existing codebase. Same 10 modules, but every exercise targets your real code instead of a tutorial project.

Best for: Developers with an existing project who want practical CC skills, not another tutorial


All five options teach the same CC features through the same 10 modules. Canvas, Forge, Nexus, and Sentinel are local-only and genuinely useful. Canvas uses HTML/CSS/JS (no setup needed); Forge, Nexus, and Sentinel are language-agnostic — you choose. BYOP works with any existing project. Pick based on interest, not difficulty.

The 10 Modules

Every project follows this same progression:

# Module CC Features Taught
1 Setup & First Contact CLAUDE.md, /init, /memory, interactive mode, keyboard shortcuts, /color, /effort, session naming
2 Blueprint & Build Plan mode, git integration, basic prompting, /branch, /plan with descriptions, includeGitInstructions
3 Rules, Memory & Context .claude/rules/, CLAUDE.local.md, @imports, /context, /compact, memory hierarchy, HTML comment hiding, autoMemoryDirectory
4 Skills & Commands SKILL.md, frontmatter, custom commands, hot-reload, argument substitution, effort frontmatter, ${CLAUDE_SKILL_DIR}, /claude-api
5 Hooks SessionStart, PostToolUse, Stop hooks, matchers, hook scripting, StopFailure, PostCompact, InstructionsLoaded, Elicitation hooks
6 MCP Servers MCP servers, .mcp.json, scopes, skills+MCP integration, elicitation, channels
7 Guard Rails PreToolUse, hook decision control, prompt-based hooks, allowRead, sandbox network isolation
8 Subagents .claude/agents/, subagent frontmatter, chaining, parallel, background, SendMessage, agent frontmatter fields
9 Tasks & TDD Tasks system, dependencies, cross-session persistence, TDD loops, /loop, cron scheduling
10 Parallel Dev, Plugins & Evaluation Worktrees, agent teams (experimental), plugins, eval, PermissionRequest hooks, continuous learning, plugin ecosystem, /remote-control, ExitWorktree

Feature Coverage Matrix

Every major CC feature is taught in all 5 options:

Feature Canvas Forge Nexus Sentinel BYOP Module
CLAUDE.md, /init, /memory x x x x x 1
Interactive mode (shortcuts, @, /, !) x x x x x 1
/color, /effort, session naming (-n) x x x x x 1
Plan mode x x x x x 2
Git integration, /branch x x x x x 2
.claude/rules/ (path-scoped) x x x x x 3
CLAUDE.local.md, memory hierarchy x x x x x 3
@imports, /context, /compact, HTML comments x x x x x 3
paths: frontmatter for rules x x x x x 3
Skills (SKILL.md, frontmatter, hot-reload) x x x x x 4
Custom slash commands, effort frontmatter x x x x x 4
paths: for skills, disableSkillShellExecution x x x x x 4
Hooks (SessionStart, PostToolUse, Stop) x x x x x 5
Hook scripting, new events (StopFailure, PostCompact, etc.) x x x x x 5
Hook if conditions, CwdChanged, FileChanged x x x x x 5
MCP servers (.mcp.json, scopes) x x x x x 6
Skills + MCP, elicitation, channels x x x x x 6
PreToolUse, hook decision control x x x x x 7
Prompt-based hooks, allowRead, sandbox settings x x x x x 7
PermissionDenied, defer, PreToolUse updatedInput x x x x x 7
Subagents (.claude/agents/) x x x x x 8
Subagent chaining, parallel, SendMessage x x x x x 8
Agent initialPrompt frontmatter x x x x x 8
Tasks system (dependencies, cross-session) x x x x x 9
TDD, /loop, cron scheduling x x x x x 9
Git worktrees, parallel dev, ExitWorktree x x x x x 10
Agent teams (experimental) x x x x x 10
Plugins, /remote-control, plugin ecosystem x x x x x 10
Plugin userConfig, sensitive storage x x x x x 10
Evaluation framework x x x x x 10
Continuous learning patterns x x x x x 10

Reference Docs

The context/ folder contains detailed reference documentation for every CC feature:

File Covers
context/claudemd.txt CLAUDE.md hierarchy, @imports, rules
context/skillsmd.txt Skills SKILL.md format, frontmatter, arguments
context/hooks.txt Hook lifecycle, events, scripting, decision control
context/configure-hooks.txt Practical hook configuration examples
context/subagents.txt Subagent creation, frontmatter, patterns
context/agent-teams.txt Agent teams (experimental), coordination
context/plugins.txt Plugin structure, manifest, bundling
context/tasks.txt Tasks system, dependencies, cross-session
context/mcp.txt MCP servers, transports, scopes
context/skills-plus-mcp.txt Combining skills with MCP tools
context/interactive-mode.txt Keyboard shortcuts, vim mode
context/common-workflows.txt Common workflows and patterns
context/when-to-use-features.txt Feature comparison and selection guide
context/boris-workflow.txt Real-world patterns, parallel Claude workflows
context/anthropic-basics.txt How Claude Code works (agentic loop, models, tools)
context/sl-guide.txt Real-world CC patterns from daily use and hackathon experience
context/changelog-cc.txt Claude Code changelog (v2.0.0 — v2.1.80)
context/auto-memory.txt Auto-memory system reference
context/models.txt Model tiers (Haiku/Sonnet/Opus), selection guidance, effort levels
context/ide-integration.txt VS Code/Cursor extension, IDE setup, CLI vs extension features

Design Principles

  • Learn by building. You won't read a single tutorial. Every feature is taught through a real task in your project — you use it, see the result, and move on.
  • Language-agnostic. Every project works in any language. You choose.
  • Local-only. No cloud services required (MCP connections are optional/local).
  • Same curriculum, different domains. All 5 options teach the same features in the same order. Pick based on interest.
  • Real tools, not toys. Every project produces something genuinely useful that you can keep using after you finish.

Always Current

You don't need to worry about the curriculum going stale. When you run /start, two things happen automatically:

  1. Repo freshness check — If the maintainers have pushed updates since you last pulled, Claude offers to git pull so you get the latest modules and docs.
  2. CC version sync — If your installed Claude Code is newer than what the curriculum covers, Claude fetches the changelog and updates the relevant lesson files and reference docs on the spot.

Returning users get a banner reminder at session start if updates are available. Everything works offline too — if either check fails, onboarding continues with the current materials.

Companion Resources

Changelog

See CHANGELOG.md for version history and updates.

License

This project is created by Zain Naboulsi and licensed under the MIT License. Co-authored with Claude by Anthropic — see DISCLAIMER.md for details on AI-generated content, warranty, and support.

Yorumlar (0)

Sonuc bulunamadi