How to Install and Use the AI Engineering from Scratch Curriculum Skills: start-learning, learn, and course-guide

Install the curriculum skills with npx skills add rohitg00/ai-engineering-from-scratch, then invoke start-learning to onboard, learn to progress through lessons, and course-guide to navigate topics.

The AI Engineering from Scratch repository by rohitg00 distributes a 523-lesson curriculum through three portable skill packages. These skills implement a host-invocation contract that allows interactive onboarding, personalized tutoring, and topic-based navigation without requiring you to manually clone the repository or manage lesson files.

What Are the Curriculum Skills?

The curriculum ships three core skill definitions, each contained in a SKILL.md file that specifies its behavior, parameters, and host integration points:

Skill	Purpose	Source Definition
start-learning	One-time onboarding that conducts a placement quiz and generates a persistent study plan.	skills/start-learning/SKILL.md
learn	Interactive tutor that reads your study plan, teaches lessons step-by-step, and updates progress.	skills/learn/SKILL.md
course-guide	Topic router that maps concepts, errors, or questions to specific lessons.	skills/course-guide/SKILL.md

Each skill operates as a stateful agent that reads from and writes to a local LEARNING.md file, making your progress portable across sessions.

Installation Requirements and Setup

The skills are distributed as an npm-style skill package installer. Before adding the curriculum, verify your environment has the required runtimes:

node --version
npx --version
python3 --version

Install the skills directly from the repository using the npx command:

npx skills add rohitg00/ai-engineering-from-scratch

The installer detects your host environment (Node.js, Python, or compatible coding agents like Codex or Claude Code) and registers the skill definitions. It creates a writable scope for storing state files such as LEARNING.md. No manual clone of the repository is required; the installer pulls SKILL.md definitions directly from raw GitHub URLs.

Using the start-learning Skill for Onboarding

The start-learning skill initializes your learning journey by creating the LEARNING.md study plan. Invoke it using your host's specific syntax:

# Codex syntax

start-learning

# Claude Code syntax

/start-learning

According to lines 41-71 of skills/start-learning/SKILL.md, the skill executes the following workflow:

1. Interview Phase: Asks three mission-aligned questions about your goals, weekly time commitment, and desired build projects.
2. Placement Assessment: Runs the placement quiz from the find-your-level skill to determine your entry phase.
3. Plan Generation: Writes a LEARNING.md file containing Mission, Placement, Path, Progress log, and Review queue sections.
4. State Initialization: Creates the single source of truth that the learn skill reads for all subsequent sessions.

The generated LEARNING.md persists across restarts, allowing the curriculum to resume exactly where you left off.

Progressing Through Lessons with the learn Skill

Once LEARNING.md exists, the learn skill functions as an interactive tutor. Invoke it with:

# Codex syntax

learn

# Claude Code syntax

/learn

As implemented in skills/learn/SKILL.md (lines 35-46), the skill follows this teaching protocol:

• State Reading: Parses LEARNING.md to locate the next lesson marked Do or Review.
• Content Fetching: Retrieves the lesson's docs/en.md and quiz.json from host-agnostic URLs (e.g., phases/*/*/docs/en.md).
• Interactive Delivery: Teaches through problem framing → core concept → code walkthrough → production-library comparison → final quiz.
• Progress Recording: Appends a timestamped row to the Progress log, updates phase status, and suggests the next action (lines 63-73).

If LEARNING.md is missing, the skill automatically falls back to start-learning (see line 15 of learn/SKILL.md), ensuring new users always begin with onboarding.

Navigating Topics with the course-guide Skill

The course-guide skill acts as a semantic router when you need to find specific content without manually browsing the 523-lesson index. Trigger it with a natural language query:

# Codex examples

course-guide attention
course-guide "my loss keeps diverging"

# Claude Code examples

/course-guide attention
/course-guide "I'm stuck on backprop"

Per skills/course-guide/SKILL.md (lines 56-66), the skill interprets your request type (topic, struggle, meta-question, or certification intent) and performs a scoped search through the Contents tables in the top-level README.md (lines 38-41). It returns up to three lesson links with justification and the exact command to run next (typically learn).

For focused routes like the Model Context Protocol or Agent Skills, course-guide directly hands off to specialized skills such as learn-mcp or learn-agent-skills, referencing manifest files in learning-paths/*.json.

Complete Workflow Example

The following session demonstrates the three skills operating in sequence on a Codex host:

$ start-learning
> Why are you learning AI engineering?   # → "to build a personal assistant"

> How many hours per week?               # → "~5 h"

> What do you want to build?             # → "an autonomous chatbot"

# Placement quiz runs, selects Phase 2 as entry point.

# LEARNING.md is created with a Path table covering all 20 phases.

$ learn
> Next lesson: Phase 2 – Linear Regression (01-linear-regression)
... (interactive teaching flow) ...

$ course-guide "my loss keeps diverging"
> Suggested lesson: Phase 3 – Deep Learning Core → 02-gradient-descent
> Run: learn

Key Files and Architecture

Understanding the underlying structure helps debug skill behavior or extend the curriculum:

Path	Role
README.md	Central index listing every lesson with phase, number, title, type, language, and path.
skills/start-learning/SKILL.md	Onboarding logic, interview steps, and LEARNING.md template definition.
skills/learn/SKILL.md	Tutor implementation describing state-file handling, lesson fetching, and teaching flow.
skills/course-guide/SKILL.md	Topic router logic for interpreting user asks and scanning lesson tables.
learning-paths/*.json	Manifest files for focused routes (MCP, Agent Skills).
phases/*/*/docs/en.md	Individual lesson content fetched by the learn skill.
phases/*/*/quiz.json	Assessment data for each lesson.

Summary

• Install the curriculum skills using npx skills add rohitg00/ai-engineering-from-scratch without cloning the repository.
• Onboard with start-learning to generate your personalized LEARNING.md study plan through a placement quiz.
• Progress with learn, which reads your state file, teaches lessons interactively, and updates your progress log automatically.
• Navigate with course-guide to jump to specific topics based on concepts, errors, or questions.
• All skills follow a host-invocation contract that adapts syntax for Codex, Claude Code, or other compatible agents.

Frequently Asked Questions

Do I need to clone the repository to use these skills?

No. The npx skills add command pulls skill definitions directly from raw GitHub URLs and registers them with your host environment. The installer creates only the necessary local state files (like LEARNING.md) without requiring a full repository clone.

What happens if I run learn before start-learning?

The learn skill checks for the existence of LEARNING.md at initialization. If the file is missing, it automatically falls back to start-learning (see line 15 of skills/learn/SKILL.md). This ensures you cannot accidentally begin mid-curriculum without a placement assessment and study plan.

How does the course-guide skill determine which lesson to recommend?

The skill parses your natural language request to identify intent (topic lookup, debugging struggle, or meta-question). It then scans the Contents tables in the top-level README.md (lines 38-41) to locate matching lessons, verifying content for struggle-based queries. It returns up to three ranked suggestions with direct links and the next command to execute.

Can I use these skills with any AI coding agent?

Yes, provided the agent implements the host-invocation contract described in each SKILL.md. The skills are portable across hosts like Codex (using bare command syntax) and Claude Code (using slash-command syntax). The underlying logic in skills/learn/SKILL.md, skills/start-learning/SKILL.md, and skills/course-guide/SKILL.md remains host-agnostic, adapting only the invocation syntax.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}