# How Focused Routes (learn-mcp and learn-agent-skills) Differ Functionally from the Main learn Skill > Understand the functional differences between the main learn skill and focused routes like learn-mcp and learn-agent-skills. Discover how each route manages learner state and curriculum delivery for targeted learning experiences. - Repository: [Rohit Ghumare/ai-engineering-from-scratch](https://github.com/rohitg00/ai-engineering-from-scratch) - Tags: deep-dive - Published: 2026-08-29 --- **The generic `learn` route acts as a catch-all entry point that inspects learner state and delegates to specialized tracks, while `learn-mcp` and `learn-agent-skills` are dedicated tutors that bypass the generic onboarding to immediately start their specific curricula.** The **rohitg00/ai-engineering-from-scratch** repository implements a portable-skill system where learning paths are defined by **SKILL.md** files under the `skills/` directory. Understanding the functional difference between the umbrella `learn` route and the focused alternatives is essential for navigating the AI engineering curriculum efficiently. ## The Generic learn Route (skills/learn/SKILL.md) The **generic `learn` skill** serves as the primary entry point for the entire learning ecosystem. Defined in [`skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/learn/SKILL.md), this route functions as an intelligent router that inspects the learner’s current state and determines which curriculum to initiate. When invoked, the generic route can: - Start the general onboarding flow (e.g., `start-learning`, `find-your-level`) - Delegate to focused skills when users explicitly request specific tracks - Resume existing learning paths across different curricula According to the source, the portable skill `learn` "hands off to the portable skill `learn-mcp`" and "hands off to the portable skill `learn-agent-skills`" when specific tracks are requested. This delegation logic allows the generic route to remain agnostic of specific curriculum details while providing a unified entry point. ## The Focused Routes Unlike the generic router, the focused routes provide dedicated, streamlined experiences for specific technical tracks. They eliminate the decision-tree overhead of the general `learn` route and jump directly into specialized curricula. ### learn-mcp (Focused Model Context Protocol Tutor) The **`learn-mcp`** route, defined in [`skills/learn-mcp/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/learn-mcp/SKILL.md), operates as a dedicated tutor for the Model Context Protocol (MCP) curriculum. When invoked via `/learn-mcp` (Claude Code) or `learn-mcp` (Codex), this route immediately creates **[`MCP-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/MCP-LEARNING.md)** and follows a strict 17-lesson MCP manifest. Key characteristics of this focused route include: - **Direct curriculum access**: Bypasses the generic onboarding flow entirely - **Evidence collection**: Records progress across wire, security, reliability, and conformance categories - **Portable fallback**: Functions as a standalone skill when explicitly selected from `/skills` ### learn-agent-skills (Focused Agent Skills Engineering Tutor) Similarly, the **`learn-agent-skills`** route, defined in [`skills/learn-agent-skills/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/learn-agent-skills/SKILL.md), provides a focused learning path for the Agent-Skills Engineering curriculum. Upon invocation, it creates **[`AGENT-SKILLS-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENT-SKILLS-LEARNING.md)** and follows its own dedicated manifest. This route mirrors the MCP structure but targets agent-specific competencies: - **Dedicated evidence tracking**: Collects the same four categories of evidence (wire, security, reliability, conformance) but within the Agent-Skills context - **No delegation overhead**: Directly starts or resumes the Agent-Skills track without consulting the generic router’s decision logic - **Curriculum isolation**: Maintains separate progress tracking from the MCP track ## Key Functional Differences **Entry Point Syntax** - **Generic `learn`**: `/learn` (Claude Code) or `learn` (Codex) - **Focused `learn-mcp`**: `/learn-mcp` (Claude Code) or `learn-mcp` (Codex) - **Focused `learn-agent-skills`**: `/learn-agent-skills` (Claude Code) or `learn-agent-skills` (Codex) **Scope and Routing Logic** - **`learn` (generic)**: Handles multiple learning flows and determines which curriculum to start based on learner state; can resume any path - **`learn-mcp`**: Solely manages the 17-lesson MCP curriculum; directly starts or resumes without extra checks - **`learn-agent-skills`**: Exclusively manages the Agent-Skills curriculum with direct entry logic **File Generation** - The generic route may create generic learning files or delegate file creation to sub-skills - `learn-mcp` specifically creates and manages [`MCP-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/MCP-LEARNING.md) - `learn-agent-skills` specifically creates and manages [`AGENT-SKILLS-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENT-SKILLS-LEARNING.md) **Evidence Tracking** Both focused routes collect structured evidence across four categories (wire, security, reliability, conformance), but the generic `learn` route only coordinates this tracking indirectly through delegation. ## Invoking the Routes Across Host Environments The **[`site/data.js`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/site/data.js)** file maps these routes for different host environments (Codex, Claude Code, and generic hosts). User-facing tables in the multilingual README files (`i18n/*/README.md`) illustrate the exact invocation syntax: ```markdown # Generic entry (defaults to start-learning or resumes last track) /learn # Claude Code learn # Codex # Direct MCP track entry /learn-mcp # Claude Code learn-mcp # Codex # Direct Agent-Skills track entry /learn-agent-skills # Claude Code learn-agent-skills # Codex ``` When using the focused routes, the system bypasses the generic skill’s routing logic entirely, creating the specific learning files immediately and loading the relevant manifest. ## Summary - The **generic `learn` route** acts as an intelligent router that delegates to focused skills based on learner state and explicit requests. - **Focused routes** (`learn-mcp` and `learn-agent-skills`) provide streamlined, curriculum-specific experiences that skip the generic onboarding flow. - **File isolation**: Each focused route creates distinct tracking files ([`MCP-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/MCP-LEARNING.md) vs [`AGENT-SKILLS-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENT-SKILLS-LEARNING.md)). - **Evidence parity**: Both focused tracks collect the same four categories of evidence (wire, security, reliability, conformance) but within their specific technical contexts. - **Invocation syntax varies** by host environment, with Claude Code using slash commands and Codex using space-separated commands. ## Frequently Asked Questions ### Can the generic learn route resume a focused track? Yes. According to [`skills/learn/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/learn/SKILL.md), the generic `learn` skill can resume any learning path, including those started via the focused routes. When resuming, it recognizes the existing [`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) files and continues from the appropriate checkpoint without requiring the user to remember which specific route they initially invoked. ### What specific evidence categories do the focused routes track? Both `learn-mcp` and `learn-agent-skills` track evidence across four standardized dimensions: **wire** (protocol implementation), **security** (authentication and authorization patterns), **reliability** (error handling and resilience), and **conformance** (specification compliance). This consistent taxonomy allows for portable skill assessment across different technical tracks within the repository. ### How do I switch between learn-mcp and learn-agent-skills tracks? To switch tracks, invoke the specific focused route for your desired curriculum. The system maintains separate progress files ([`MCP-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/MCP-LEARNING.md) and [`AGENT-SKILLS-LEARNING.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/AGENT-SKILLS-LEARNING.md)), allowing you to pause one track and resume another without losing context. The generic `learn` route can also route you to either track if you request a specific curriculum by name during the session. ### Are the focused routes available in all host environments? Yes, according to [`site/data.js`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/site/data.js) and the i18n documentation tables, both focused routes are mapped across all supported host environments including Claude Code, Codex, and generic hosts. However, invocation syntax differs: Claude Code uses slash commands (`/learn-mcp`, `/learn-agent-skills`), while Codex uses space-separated commands (`learn-mcp`, `learn-agent-skills`).