AI Tutor Skills in ai-engineering-from-scratch: Complete Guide to 8 Modular Learning Agents

The ai-engineering-from-scratch repository provides eight portable AI tutor skills—start-learning, learn, learn-mcp, learn-agent-skills, find-your-level, check-understanding, course-guide, and claude-certification—that form a modular, state-driven tutoring architecture powered by markdown-based skill contracts.

This open-source project by rohitg00 implements a portable skill system where each AI tutor capability lives in its own directory under skills/<skill-name>/ with a standardized SKILL.md contract. These skills enable interactive, resumable learning experiences for AI engineering topics without requiring external platforms or databases.

Core Architecture of AI Tutor Skills

Every skill in the repository follows a common portable-skill contract defined in its SKILL.md file. This contract specifies:

• Name and version – explicit identifier for host routing
• Description – concise purpose statement
• Host-invocation contract – syntax for Codex, Claude-Code, and natural-language triggers
• State files – markdown files that persist learner progress
• Routing logic – how resume/continue/start requests delegate between skills
• Content sources – local phases/ directory or raw GitHub URLs for lesson material

This architecture allows learners to switch routes dynamically, resume any progress, and receive consistent tutoring regardless of which AI host they use.

Complete List of AI Tutor Skills

start-learning – Personalized Onboarding

Primary purpose: Interviews the learner, runs placement assessment, and creates the master learning plan.

Attribute	Details
State file	LEARNING.md
Definition	[skills/start-learning/SKILL.md](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/start-learning/SKILL.md)

When invoked, this skill:

1. Gathers learner background and goals
2. Executes find-your-level for placement
3. Writes LEARNING.md with mission, recommended entry phase, and selected learning path

As implemented in rohitg00/ai-engineering-from-scratch, the skill includes resume routing logic (lines 43-53) that checks for existing state files and hands off to the appropriate specialized skill.

learn – Full Curriculum Interactive Tutor

Primary purpose: Reads the learning plan, fetches next lessons, teaches interactively, and logs progress.

Attribute	Details
State file	LEARNING.md
Definition	[skills/learn/SKILL.md](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/learn/SKILL.md)

The learn skill implements the core tutoring loop:

• Reads LEARNING.md to identify the current lesson
• Sources content from local phases/ directory or raw GitHub URLs (lines 35-42)
• Presents material interactively with inline quizzes
• Appends progress rows to LEARNING.md (lines 64-71)
• Adds low-scoring lessons to a review queue

# Direct invocation

learn

learn-mcp – Model-Context-Protocol Focused Track

Primary purpose: Dedicated tutor for the Model-Context-Protocol specialization with isolated state management.

Attribute	Details
State file	MCP-LEARNING.md
Manifest	[learning-paths/model-context-protocol.json](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/learning-paths/model-context-protocol.json)
Definition	[skills/learn-mcp/SKILL.md](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/learn-mcp/SKILL.md)

This skill maintains its own evidence-recording scheme (lines 30-33) separate from the main curriculum. It's designed for learners who want to focus exclusively on MCP concepts without the broader AI engineering curriculum.

# Claude-Code syntax

/learn-mcp

learn-agent-skills – Agent Skills Specialization

Primary purpose: Tutor for the five-lesson Agent-Skills track.

Attribute	Details
State file	AGENT-SKILLS-LEARNING.md
Manifest	[learning-paths/agent-skills.json](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/learning-paths/agent-skills.json)
Definition	[skills/learn-agent-skills/SKILL.md](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/learn-agent-skills/SKILL.md)

Similar to learn-mcp, this skill operates in isolated state for focused learning on agent construction patterns.

find-your-level – Placement Assessment

Primary purpose: 10-question quiz that maps learner knowledge to appropriate entry phase.

Attribute	Details
State file	None (stateless)
Definition	[skills/find-your-level/SKILL.md](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/find-your-level/SKILL.md)

The quiz structure (lines 19-23) presents questions in 5 rounds, isolates answer keys to prevent leakage, and applies scoring logic (lines 25-28) to recommend entry points from Phase 1 through advanced tracks.

# Invoke directly

find-your-level

check-understanding – Phase-Level Assessment

Primary purpose: Post-lesson comprehensive quiz for an entire phase.

Attribute	Details
State file	Reads LEARNING.md for context
Definition	[skills/check-understanding/SKILL.md](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/check-understanding/SKILL.md)

This skill aggregates all "post" quiz items from a phase's quiz.json file—such as [phases/13-tools-and-protocols/quiz.json](https://github.com/rohitg00/ai-engineering-from-scratch/tree/main/phases)—to validate mastery before progression.

# Natural-language trigger

Check my understanding of Phase 13

course-guide – Topic Navigation

Primary purpose: Jump directly to any lesson or concept regardless of current plan.

Attribute	Details
State file	Uses LEARNING.md if present (optional)
Definition	[skills/course-guide/SKILL.md](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/course-guide/SKILL.md)

The course-guide skill enables non-linear exploration—learners can query specific topics like "transformers" and receive direct navigation to relevant lessons.

# Codex-style invocation

course-guide transformers

claude-certification – Certification Orchestration

Primary purpose: Manages the complete Claude certification workflow including tracks, exams, and artifact validation.

Attribute	Details
State file	CLAUDE-CERTIFICATION.md
Definition	[skills/claude-certification/SKILL.md](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/claude-certification/SKILL.md)

This is the highest-level guidance skill, coordinating multiple certification tracks and maintaining persistent progress across exam attempts.

# Claude-Code slash command

/claude-certification

Skill Invocation Examples

The portable-skill contract supports multiple host syntaxes. Here are complete examples for each skill:

# 1. Start personalized learning plan

start-learning

# 2. Resume existing progress

Resume my learning plan

# 3. Run next lesson in full curriculum

learn

# 4. Switch to MCP specialization

/learn-mcp

# 5. Run placement quiz standalone

find-your-level

# 6. Assess phase mastery

Check my understanding of Phase 13

# 7. Navigate to specific topic

course-guide transformers

# 8. Begin certification workflow

/claude-certification

State File Routing Mechanism

A key architectural feature is automatic skill delegation based on state file presence. When the start-learning or learn skills receive a "resume" request, they check for these files in order:

State File Detected	Skill Delegated To
MCP-LEARNING.md	learn-mcp
AGENT-SKILLS-LEARNING.md	learn-agent-skills
CLAUDE-CERTIFICATION.md	claude-certification
LEARNING.md	learn (default)

This routing table (defined in start-learning SKILL.md lines 43-53) ensures learners always resume in the correct context without manual intervention.

Content Sourcing Strategy

Skills source lesson material through a fallback hierarchy:

1. Local clone: Read from phases/<phase-number>-<name>/ directory
2. Remote fetch: Retrieve from raw.githubusercontent.com/rohitg00/ai-engineering-from-scratch/... URLs

This dual-mode operation enables the tutor to function whether the repository is cloned locally or accessed through a web-based AI host.

Summary

• Eight portable AI tutor skills provide complete coverage: onboarding, full-curriculum tutoring, two specialized tracks (MCP and Agent Skills), placement testing, phase assessment, topic navigation, and certification management
• Markdown-based state files (LEARNING.md, MCP-LEARNING.md, AGENT-SKILLS-LEARNING.md, CLAUDE-CERTIFICATION.md) enable resumable, host-agnostic learning
• Portable-skill contract standardizes invocation across Codex, Claude-Code, and natural-language interfaces
• Automatic routing delegates resume requests to the appropriate skill based on detected state files
• Dual content sourcing works with local clones or remote GitHub URLs

Frequently Asked Questions

How do I switch between the main curriculum and a specialized track?

Use course-guide to navigate directly, or simply start the specialized skill (learn-mcp or learn-agent-skills). The system will create the appropriate state file and begin that track. To return, invoke learn—the routing logic in start-learning (lines 43-53) will detect your LEARNING.md and resume the main curriculum.

Can I run these AI tutor skills without cloning the repository?

Yes. The skills fetch lesson content from raw GitHub URLs when the local phases/ directory is unavailable. However, cloning rohitg00/ai-engineering-from-scratch provides faster access and enables offline learning.

What happens to my progress if I switch AI hosts?

Progress persists in markdown state files committed to your repository fork or saved locally. Since all hosts read the same SKILL.md contracts and state file formats, you can resume seamlessly across Codex, Claude-Code, or any other compatible environment.

How does the placement quiz determine my starting phase?

The find-your-level skill presents 10 questions across 5 rounds, maps your total score to phase thresholds defined in its SKILL.md (lines 25-28), and writes the recommended entry point into LEARNING.md. You can override this manually or retake the quiz at any time.

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