# AI Tutoring Curriculum Skills Patterns: 8 Reusable Agent Modules in AI-Engineering-From-Scratch > Discover the 8 AI tutoring curriculum skills patterns from rohitg00/ai-engineering-from-scratch. Learn about reusable agent modules, deterministic workflows, and host-agnostic contracts for portable AI learning. - Repository: [Rohit Ghumare/ai-engineering-from-scratch](https://github.com/rohitg00/ai-engineering-from-scratch) - Tags: tutorial - Published: 2026-08-29 --- **The eight AI tutoring curriculum skills in rohitg00/ai-engineering-from-scratch follow a unified architectural template featuring YAML front-matter metadata, host-agnostic invocation contracts, state-file routing, and deterministic six-step workflows that enable portable, extensible learning agents.** The rohitg00/ai-engineering-from-scratch repository implements a modular AI tutoring system where eight curriculum skills act as reusable agents. These portable modules share standardized patterns for metadata declaration, invocation protocols, and progress tracking, creating a consistent learning experience across Codex, Claude Code, and compatible hosts. Understanding these AI tutoring curriculum skills patterns reveals how the system maintains architectural coherence while supporting diverse learning paths from Model Context Protocol to Agent Skills. ## The Eight Core Curriculum Skills The repository defines eight distinct tutoring agents in the `/skills/` directory: - **start-learning**: Initiates the onboarding flow and determines the appropriate learning path - **learn**: Delivers the full AI engineering curriculum through sequential lessons - **learn-mcp**: Focuses specifically on Model Context Protocol concepts and implementation - **learn-agent-skills**: Teaches autonomous agent development patterns - **find-your-level**: Runs placement assessments to determine learner entry points - **course-guide**: Provides quick reference documentation for specific topics - **claude-certification**: Guides learners through Claude-specific certification materials - **check-understanding**: Validates comprehension of completed phases through targeted quizzes Each skill module resides in its own subdirectory (e.g., [`/skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn/SKILL.md), [`/skills/learn-mcp/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn-mcp/SKILL.md)) and implements the same structural contract. ## Architectural Patterns in AI Tutoring Skills ### YAML Front-Matter Metadata Every skill file begins with a standardized YAML block that defines `name`, `version`, `description`, and `tags`. For example, the `learn` skill uses tags like `learn`, while `learn-mcp` uses `learn-mcp`. This metadata appears at the top of files like [`/skills/start-learning/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/start-learning/SKILL.md) and [`/skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn/SKILL.md), enabling the orchestration layer to discover and route to appropriate tutors without parsing the full document. ### Host-Agnostic Invocation Contracts Each skill specifies exactly how it should be invoked across three supported host types: **Codex** (plain command), **Claude Code** (slash-prefixed), and other compatible agents. The contract appears as a small table in the skill documentation, ensuring that `start-learning`, `/learn`, or `/learn-mcp` trigger the same behavior regardless of the underlying runtime. This pattern in [`/skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn/SKILL.md) guarantees portability across AI development environments. ### State-File Routing and Persistence Skills declare dedicated state files that serve as the single source of truth for learner progress: - [`LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/LEARNING.md) for general curriculum (`learn`, `start-learning`) - [`MCP-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/MCP-LEARNING.md) (legacy [`MCP-ENGINEERING-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/MCP-ENGINEERING-LEARNING.md)) for Model Context Protocol studies - [`AGENT-SKILLS-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENT-SKILLS-LEARNING.md) for agent development tracks - [`CLAUDE-CERTIFICATION.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CLAUDE-CERTIFICATION.md) for certification progress As documented in [`/skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn/SKILL.md), each skill explains how it locates these state files to resume interrupted sessions, ensuring continuous learning paths without data loss. ### Content Source Declaration Skills explicitly declare where lesson content resides, typically referencing both local paths (`phases/…/docs/en.md` and [`quiz.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/quiz.json)) and remote raw-GitHub URLs. When the repository isn't cloned locally, the skill automatically falls back to remote content sources. This dual-path resolution in [`/skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn/SKILL.md) ensures the tutoring system functions in both offline and online contexts. ### Deterministic Step-by-Step Workflows Every skill implements an identical six-phase pipeline documented in numbered steps: 1. **Step 0**: Locate and validate the state file 2. **Step 1**: Warm-up recall of previous concepts 3. **Step 2**: Teach the current lesson 4. **Step 3**: Administer the knowledge quiz 5. **Step 4**: Record progress to the state file 6. **Step 5**: Close the session and prepare next steps This deterministic workflow in [`/skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn/SKILL.md) makes the tutor's behavior predictable and debuggable across all eight skills. ### Evidence Modes and Security Gating Skills that execute runnable code, such as `learn-mcp` and `learn-agent-skills`, implement **evidence modes** distinguishing between Executable and Conceptual interactions. They also define **public deployment gates** in [`/skills/learn-mcp/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn-mcp/SKILL.md) that must be satisfied before performing network operations or file modifications, protecting learners from accidental execution of untrusted code. ### Explicit Prohibitions and Safety Rules Each skill file contains a uniform "Do-not-do" section listing actions strictly forbidden to the agent. These include never exposing secret values, never inferring host syntax globally, and never blocking lesson delivery while waiting for repository cloning. As seen in [`/skills/learn-agent-skills/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn-agent-skills/SKILL.md), these explicit prohibitions protect both the learner's environment and the integrity of the tutoring process. ## Cross-Host Invocation Examples The following examples demonstrate the portable invocation contract across different AI hosts: ```text # Start the onboarding flow (Codex) start-learning # Start the onboarding flow (Claude Code) /start-learning ``` ```text # Teach the next lesson (Codex) learn # Teach the next lesson (Claude Code) /learn ``` ```text # Focus on Model Context Protocol (Codex) learn-mcp # Focus on Model Context Protocol (Claude Code) /learn-mcp ``` ```text # Focus on Agent Skills (Codex) learn-agent-skills # Focus on Agent Skills (Claude Code) /learn-agent-skills ``` ```text # Run placement quiz (Codex) find-your-level # Run placement quiz (Claude Code) /find-your-level ``` ```text # Get quick reference (Codex) course-guide # Get quick reference (Claude Code) /course-guide ``` ```text # Claude certification route (Codex) claude-certification # Claude certification route (Claude Code) /claude-certification ``` ```text # Quiz on finished phase (Codex) check-understanding # Quiz on finished phase (Claude Code) /check-understanding ``` Because each skill respects the same contract, hosts can swap calls without code changes, and the tutor automatically locates the correct state file and lesson assets. ## Skill-to-State File Mapping The following table maps each curriculum skill to its implementation file and persistent state: | Skill | Primary File | State File | Manifest | |-------|-------------|------------|----------| | **start-learning** | [`/skills/start-learning/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/start-learning/SKILL.md) | [`LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/LEARNING.md) | — | | **learn** | [`/skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn/SKILL.md) | [`LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/LEARNING.md) | — | | **learn-mcp** | [`/skills/learn-mcp/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn-mcp/SKILL.md) | [`MCP-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/MCP-LEARNING.md) | [`learning-paths/model-context-protocol.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/learning-paths/model-context-protocol.json) | | **learn-agent-skills** | [`/skills/learn-agent-skills/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn-agent-skills/SKILL.md) | [`AGENT-SKILLS-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENT-SKILLS-LEARNING.md) | [`learning-paths/agent-skills.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/learning-paths/agent-skills.json) | | **find-your-level** | [`/skills/find-your-level/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/find-your-level/SKILL.md) | — (produces placement result) | — | | **course-guide** | [`/skills/course-guide/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/course-guide/SKILL.md) | — | — | | **claude-certification** | [`/skills/claude-certification/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/claude-certification/SKILL.md) | [`CLAUDE-CERTIFICATION.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CLAUDE-CERTIFICATION.md) | — | | **check-understanding** | [`/skills/check-understanding/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/check-understanding/SKILL.md) | — | — | ## Summary - **Eight portable skill modules** form the backbone of the AI tutoring system, each residing in `/skills//SKILL.md` - **YAML front-matter** standardizes metadata across all skills, enabling automatic discovery and routing - **Host-agnostic invocation contracts** support Codex, Claude Code, and compatible agents through consistent command patterns - **State-file routing** maintains learner progress in dedicated markdown files ([`LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/LEARNING.md), [`MCP-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/MCP-LEARNING.md), etc.) - **Six-step deterministic pipelines** (locate, recall, teach, quiz, record, close) ensure predictable tutor behavior - **Security gates and evidence modes** protect learners when executing code through explicit deployment checks and "do-not-do" rules ## Frequently Asked Questions ### What are the eight curriculum skills in the AI-Engineering-From-Scratch repository? The eight curriculum skills are **start-learning**, **learn**, **learn-mcp**, **learn-agent-skills**, **find-your-level**, **course-guide**, **claude-certification**, and **check-understanding**. Each functions as a reusable tutoring agent in the `/skills/` directory, following standardized patterns for metadata, invocation, and state management. ### How do the AI tutoring skills maintain state across sessions? Each skill writes progress to dedicated state files such as [`LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/LEARNING.md), [`MCP-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/MCP-LEARNING.md), or [`AGENT-SKILLS-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENT-SKILLS-LEARNING.md). According to [`/skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn/SKILL.md), Step 0 of every skill involves locating these files to resume interrupted sessions, creating a single source of truth for learner progress that persists between interactions. ### Can these curriculum skills work with AI hosts other than Codex and Claude Code? Yes. The skills implement **host-agnostic invocation contracts** that specify how to call the agent on "other compatible hosts" beyond Codex and Claude Code. As documented in [`/skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn/SKILL.md), the contract table and prose notes ensure any agent runtime supporting the basic command structure can execute the tutoring workflows. ### What security measures prevent unsafe code execution in the tutoring skills? Skills involving runnable code define **evidence modes** (Executable vs Conceptual) and **public deployment gates** that must be satisfied before network or file operations execute. Additionally, each skill includes explicit "Do-not-do" rules prohibiting exposure of secrets and blocking lessons during repository cloning, as specified in [`/skills/learn-agent-skills/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn-agent-skills/SKILL.md) and [`/skills/learn-mcp/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main//skills/learn-mcp/SKILL.md).