AI Learning Skills in ai-engineering-from-scratch: 8 Interactive Tutors Explained

The repository ships eight portable AI learning skills—self-contained SKILL.md modules that handle everything from onboarding and placement quizzes to specialized tracks like MCP and Agent Skills.

The ai-engineering-from-scratch curriculum by rohitg00 implements a modular, host-agnostic tutoring system. Each skill is a standalone markdown contract stored under skills/<skill-name>/SKILL.md, enabling AI hosts like Codex and Claude to deliver interactive lessons, run assessments, and manage persistent learner state across multiple specialized tracks.

Core Curriculum Skills

These foundational skills manage the primary learning loop and initial learner onboarding.

start-learning: Onboarding and Placement

Located at skills/start-learning/SKILL.md, this skill performs one-time learner initialization. It interviews the user, executes the placement quiz via find-your-level, and generates a persistent LEARNING.md study plan that records the learner’s current phase and progress.

Primary commands:

• Codex: start-learning
• Claude: /start-learning

learn: Interactive Lesson Tutor

The central tutoring engine lives in skills/learn/SKILL.md. This skill reads the learner’s LEARNING.md state, fetches the next lesson from phases/**/docs/en.md, teaches content step-by-step, administers quizzes, and updates progress markers.

Primary commands:

• Codex: learn
• Claude: /learn

Specialized Learning Tracks

These skills bypass generic onboarding to launch dedicated certification or protocol-specific curricula.

learn-mcp: Model Context Protocol Path

Defined in skills/learn-mcp/SKILL.md, this skill drives a 17-lesson route focused on Model Context Protocol (MCP) implementation. It maintains separate state in MCP-LEARNING.md and consumes the manifest at learning-paths/model-context-protocol.json to sequence lessons.

Primary commands:

• Codex: learn-mcp
• Claude: /learn-mcp

learn-agent-skills: Agent Competency Track

Situated at skills/learn-agent-skills/SKILL.md, this skill manages a 5-lesson Agent Skills curriculum. It tracks progress via AGENT-SKILLS-LEARNING.md and references learning-paths/agent-skills.json for lesson ordering.

Primary commands:

• Codex: learn-agent-skills
• Claude: /learn-agent-skills

claude-certification: Official Exam Preparation

Found in skills/claude-certification/SKILL.md, this skill orchestrates onboarding and progress tracking for Claude’s official certification exams. It maintains state in CLAUDE-CERTIFICATION.md and provides structured preparation workflows aligned with Anthropic’s exam objectives.

Primary commands:

• Codex: claude-certification
• Claude: /claude-certification

Utility and Assessment Skills

These supporting skills provide topic lookup, placement testing, and phase-specific quizzing.

course-guide: Topic Navigation

Implemented in skills/course-guide/SKILL.md, this skill maps arbitrary topic names to exact lesson files. Given a concept like "attention", it directs the learner to the precise phase and document covering that material.

Primary commands:

• Codex: course-guide <topic>
• Claude: /course-guide <topic>

find-your-level: Placement Quiz Engine

Stored in skills/find-your-level/SKILL.md, this skill administers a 10-question assessment across five knowledge areas. It calculates a score, maps it to an entry phase (1-18), and writes a placement section to the learner’s state file.

Primary commands:

• Codex: find-your-level
• Claude: /find-your-level

check-understanding: Phase-Level Assessment

Located at skills/check-understanding/SKILL.md, this skill executes comprehensive quizzes for specific phases. It reads phases/<phase>/quiz.json files and evaluates mastery before permitting progression.

Primary commands:

• Codex: check-understanding <phase>
• Claude: /check-understanding <phase>

How to Invoke AI Learning Skills

All skills follow a portable contract, allowing identical invocation across CLI, chat interfaces, and natural-language contexts.

Codex CLI Style

# Initialize learner profile and study plan

start-learning

# Resume standard curriculum

learn

# Access specialized tracks

learn-mcp
learn-agent-skills
claude-certification

# Navigation and assessment

course-guide transformers
find-your-level
check-understanding 13

Claude Code Slash Commands

/start-learning
/learn
/learn-mcp
/learn-agent-skills
/claude-certification
/course-guide attention
/find-your-level
/check-understanding 13

Natural Language Invocation

Use start-learning to create my study plan.
Use learn to continue the next lesson.
Use course-guide to show me the RAG lesson.
Use find-your-level to place me in the right phase.
Use check-understanding 13 to quiz me on Phase 13.
Use learn-mcp to start the Model Context Protocol path.

Skill Architecture and State Management

The curriculum engine operates on a state-first resolution pattern. When invoked, each skill first checks for existing state files in the working directory:

• LEARNING.md → Resume via learn
• MCP-LEARNING.md → Resume via learn-mcp
• AGENT-SKILLS-LEARNING.md → Resume via learn-agent-skills
• CLAUDE-CERTIFICATION.md → Resume via claude-certification

If no state exists, start-learning executes automatically, running the placement quiz from find-your-level before handing control to the generic learn skill. Specialized routes (learn-mcp, learn-agent-skills) skip generic onboarding and immediately initialize their dedicated state files and JSON manifests.

Summary

• Eight portable skills drive the entire ai-engineering-from-scratch tutoring experience, each defined in isolated SKILL.md files.
• Core skills (start-learning, learn) handle onboarding and daily lesson delivery, reading from LEARNING.md and phases/**/docs/en.md.
• Specialized tracks (learn-mcp, learn-agent-skills, claude-certification) maintain separate state files and JSON manifests for protocol-specific or certification curricula.
• Utility skills (course-guide, find-your-level, check-understanding) provide topic lookup, placement testing, and phase-level assessment by reading quiz.json files.
• Host-agnostic design allows identical skill invocation across Codex CLI, Claude Code, and natural-language interfaces without code changes.

Frequently Asked Questions

How does the repository determine which AI learning skill to run first?

The skill engine checks for existing state files (LEARNING.md, MCP-LEARNING.md, etc.) in the working directory. If a state file exists, the corresponding skill resumes automatically. If no state exists, start-learning runs the onboarding interview and placement quiz before transferring control to the generic learn skill.

Can I switch between the general curriculum and specialized tracks like MCP or Agent Skills?

Yes. Each track maintains its own state file (e.g., MCP-LEARNING.md for the MCP path). Invoking learn-mcp or learn-agent-skills initializes the dedicated track and its separate progression log. You can resume the general curriculum later by invoking learn, which reads the original LEARNING.md.

What file format stores the lesson content and quizzes consumed by these skills?

Lesson content resides in phases/<phase-number>/docs/en.md as markdown documents. Quizzes are stored as structured JSON in phases/<phase-number>/quiz.json. The learn and check-understanding skills parse these files to deliver content and evaluate responses.

Are these AI learning skills compatible with AI hosts other than Codex and Claude?

Yes. The skills are implemented as portable SKILL.md contracts that any host supporting the "skill" pattern can execute. The repository demonstrates usage for Codex (CLI commands) and Claude (slash commands), but the same markdown definitions work with any compatible tutoring engine that respects the skill contract.

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