How to Use the start-learning Skill for Onboarding in AI Engineering from Scratch

The start-learning skill creates a personalized onboarding plan by conducting a brief interview, running a placement quiz, and generating a LEARNING.md file that serves as the single source of truth for your curriculum progress.

The AI Engineering from Scratch curriculum by rohitg00/ai-engineering-from-scratch provides a structured, self-paced learning environment through modular skills. The start-learning skill serves as the entry point for all new learners, establishing your learning profile and determining your optimal entry phase into the course material.

Host-Specific Invocation Syntax

The skill follows a host invocation contract defined in skills/start-learning/SKILL.md (lines 26-35) that adapts to different interfaces. The metadata header (lines 1-10) declares the trigger phrases that activate the skill across supported hosts.

Depending on your environment, invoke the skill using one of these patterns:

• Codex-compatible hosts: start-learning
• Claude Code-compatible hosts: /start-learning
• Plain-language interfaces: "Use start-learning to begin the course."

This contract ensures the skill works across any host—whether you are using a CLI, integrated development environment, or natural-language chat interface—without requiring code changes to the underlying skill logic.

Resume Routing and State Detection

Before initiating a fresh onboarding flow, the skill checks for existing learning state files through resume routing logic documented at lines 41-64 in SKILL.md. This prevents redundant onboarding for returning learners.

The skill searches for files such as LEARNING.md, MCP-LEARNING.md, or AGENT-SKILLS-LEARNING.md. If detected, the skill routes you directly to the appropriate downstream skill—learn, learn-mcp, or learn-agent-skills—rather than re-running the placement interview. This architecture ensures state-file ownership remains clean: start-learning only modifies LEARNING.md, while other learning routes maintain their own separate state files (lines 46-52).

The Onboarding Flow

The start-learning skill executes a four-stage pipeline to establish your learning environment.

1. The Placement Interview (Lines 105-115)

The skill prompts you with three structured questions to ground future explanations:

• Why are you learning AI engineering?
• How many hours can you devote each week?
• What do you most want to build by the end of the curriculum?

Your answers are stored verbatim in the generated state file to personalize subsequent lessons.

2. The find-your-level Placement Quiz

Immediately following the interview, the skill invokes the companion find-your-level skill (line 21) to administer a 5-area, 10-question assessment. This quiz determines your entry phase and populates the Path table in your learning plan.

3. Generating the LEARNING.md State File

Section 39-71 of SKILL.md defines a strict markdown template (lines 43-71) that creates your LEARNING.md with four mandatory sections:

• Mission: Your stated goals from the interview
• Placement: Results from the find-your-level quiz
• Path: Structured table of phases with estimated hours drawn from ROADMAP.md
• Progress log and Review queue: Empty containers for ongoing tracking

The template embeds a permanent link back to the repository (line 45) for reference.

4. Hand-off to Learning Mode

After writing LEARNING.md, the skill outputs three concise lines (lines 75-81):

1. A summary of your entry point and estimated total hours
2. The correct host-specific invocation for the learn skill
3. The invocation pattern for course-guide <topic> to jump to specific subjects

State-File Ownership and Persistence

Each learning route owns its own state file to prevent cross-route contamination (lines 46-52). The start-learning skill strictly manages only LEARNING.md; if you later switch to the MCP or Agent Skills tracks, those routes respectively create MCP-LEARNING.md or AGENT-SKILLS-LEARNING.md.

This separation ensures that your progress in one curriculum mode does not accidentally overwrite or interfere with another. The LEARNING.md file serves as the single source of truth for the standard learning path, readable by subsequent learn sessions to maintain continuity across disconnected study sessions.

Practical Usage Examples

Initialize onboarding on a Codex-compatible CLI:

start-learning

Begin the curriculum in Claude Code:

/claude-code
/start-learning

For hosts without slash-command support, use natural language:

Use start-learning to begin the course.

Typical interaction flow:

User: /start-learning
Assistant: Welcome! Why are you learning AI engineering?
User: To build a personal AI assistant.
Assistant: How many hours can you devote each week?
User: ~5 h.
Assistant: What do you most want to build by the end?
User: An autonomous chatbot.
Assistant: Running the placement quiz… (invokes find-your-level)
Assistant: Writing LEARNING.md…
Assistant: Done! Your entry point is Phase 3 – "Neural Networks". 
          Estimated total: 120 h.
          Next step: run `learn` (or `/learn`) to start the first lesson.
          You can also jump to a specific topic with `course-guide <topic>`.

Extensibility and Host Compatibility

Because the skill relies solely on plain-text I/O and a well-known file name (LEARNING.md), future hosts can adopt the same contract without code modifications. The natural-language fallback (line 35) guarantees graceful degradation on hosts that do not support structured slash commands.

The architecture separates host-specific rendering from core skill logic, meaning the interview questions, quiz invocation, and file generation remain consistent regardless of whether you are using Codex, Claude Code, or a custom interface.

Summary

• The start-learning skill is the mandatory entry point for the AI Engineering from Scratch curriculum, defined in skills/start-learning/SKILL.md.
• It supports multiple host syntaxes: bare commands for Codex, slash commands for Claude Code, and natural language for generic interfaces.
• The skill performs resume routing to detect existing LEARNING.md files and avoid redundant onboarding.
• The onboarding flow includes a 3-question interview, invocation of the find-your-level placement quiz, and generation of a structured LEARNING.md file.
• State-file ownership is strictly partitioned: start-learning only writes to LEARNING.md, while other learning modes use their own distinct state files.
• The skill hands off to the learn command with host-specific syntax and provides access to the course-guide utility for topic-specific navigation.

Frequently Asked Questions

How do I restart the onboarding process if I want to change my goals?

Delete the LEARNING.md file from your workspace root and invoke the start-learning skill again. Because the skill checks for existing state files before running the interview, removing LEARNING.md triggers a fresh onboarding flow. Your previous progress in other tracks (such as MCP-LEARNING.md) remains unaffected due to strict state-file separation.

What happens if I already have a LEARNING.md file from a different course?

The start-learning skill specifically checks the structure and metadata within LEARNING.md to determine if it belongs to the AI Engineering from Scratch curriculum. If the file indicates you are mid-progress, the skill routes you to the appropriate resume logic (lines 41-64) rather than overwriting your existing plan. To force a reset, manually archive or delete the existing file.

Can I use the start-learning skill without Codex or Claude Code?

Yes. The skill implements a natural-language fallback (line 35) that allows any host supporting plain text I/O to invoke the skill. Simply type "Use start-learning to begin the course" in compatible chat interfaces. The skill generates the same LEARNING.md structure regardless of host, ensuring consistent behavior across environments.

Where does the estimated hours information come from in my learning plan?

The Path table in your generated LEARNING.md pulls estimated hours from ROADMAP.md in the repository root. During file generation (lines 43-71), the skill maps your placement results—determined by the find-your-level quiz—to the corresponding phases in the roadmap, calculating your total estimated commitment based on the interview responses regarding weekly availability.

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