How Claude Agent Skills Find Your Level and Check Understanding in AI Engineering From Scratch

The find-your-level and check-understanding skills in the rohitg00/ai-engineering-from-scratch repository operate as automated assessment tools that map learners to personalized curriculum paths through interactive quizzes.

The Claude agent skills embedded in this open-source curriculum provide intelligent tutoring capabilities that eliminate the guesswork from self-paced learning. According to the source code in .claude/skills/, these tools leverage structured quiz logic and content sampling to deliver precise placement recommendations and mastery verification across a 20-phase, 260-lesson learning track.

How the find-your-level Skill Operates

The find-your-level skill functions as a diagnostic placement engine that maps total quiz scores to specific entry points in the curriculum.

Ten-Question Placement Quiz Structure

The skill administers a ten-question assessment covering five knowledge domains: Math & Statistics, Classical ML, Deep Learning, NLP & Transformers, and Applied AI. Each domain contains exactly two questions, presented in sequential rounds. Learners score 0–2 points per area, yielding a composite score between 0 and 10. This granular scoring allows the system to distinguish between surface familiarity and deep competency across the AI engineering stack.

Score-to-Phase Mapping Logic

Inside .claude/skills/find-your-level/SKILL.md, lines 72–80 define the mapping table that converts raw scores into curriculum phases. For example, a score of 0–3 routes the learner to Phase 1: Math Foundations, while a perfect 10 advances them directly to Phase 14: Agent Engineering. The mapping logic uses inclusive ranges to ensure learners start at the appropriate difficulty threshold without redundant review of mastered material.

Personalized Learning Path Generation

After calculating the entry phase, the skill constructs a markdown compatibility table spanning all 20 phases. As implemented in lines 84–122 of the skill definition, the logic marks each phase as Skip, Review, or Do, and injects hour estimates pulled dynamically from ROADMAP.md. The output includes the completed table, a total-hour estimate for the remaining curriculum, and a concise recommendation on the optimal starting point.

How the check-understanding Skill Operates

The check-understanding skill provides phase-level mastery verification through adaptive content sampling and interactive questioning.

Activation and Phase Resolution

Users trigger the skill via /check-understanding <phase> or natural language equivalents like "quiz me on phase 2". The command parser, defined in lines 13–19 of .claude/skills/check-understanding/SKILL.md, accepts phase numbers (0–19) or canonical names. A resolution table in lines 31–50 validates inputs against the phases/<phase-dir>/ directory structure, returning specific error messages for invalid entries like "quantum" or out-of-range indices.

Content Sampling and Question Generation

The skill ingests lesson content using a glob pattern to locate docs/en.md files within the target phase directory. For large phases, it implements strategic sampling—reading the first, middle, and last lessons to maintain quiz brevity while covering the full scope (lines 58–63).

Question generation occurs in lines 64–84, producing exactly eight multiple-choice items: four conceptual (What/Why) and four practical (How/Build). Each question includes 3–4 distractors, a single correct answer, and a metadata tag linking back to the originating lesson file for remediation purposes.

Interactive Delivery and Scoring

Delivery leverages the AskUserQuestion tool (lines 86–101), presenting questions one-by-one and awaiting learner responses before proceeding. The scoring engine (lines 111–138) assigns categorical grades based on accuracy:

• Mastered (7–8 correct)
• Almost (5–6 correct)
• Developing (3–4 correct)
• Start Over (0–2 correct)

Feedback includes a breakdown of missed items with explanations and direct links to the relevant lesson files. Lines 141–148 handle next-step logic, offering options to retake the quiz, attempt a different phase, or request targeted explanations.

Usage Examples

# Placement quiz invocation

User: /find-your-level
Claude: Welcome! Let's find the right starting point for you.
... (Claude administers 10 questions across 5 domains) ...

# Phase understanding check

User: /check-understanding 3
Claude: Great! We'll test your knowledge of Deep Learning Core.
1/8 (Conceptual) – What does back-propagation compute?
A) Learning rate   B) Gradient of loss   C) Number of layers   D) Batch size
> User: B
... (Claude continues with remaining questions) ...

# Error handling for invalid phases

User: /check-understanding quantum
Claude: Unknown phase 'quantum'. Pick from the list below:
0 Setup & Tooling, 1 Math Foundations, ..., 19 Capstone Projects

Key Implementation Files

• .claude/skills/find-your-level/SKILL.md – Contains the placement quiz logic, score mapping (lines 72–80), and personalized table generation (lines 84–122)
• .claude/skills/check-understanding/SKILL.md – Houses the phase resolution table, content sampling algorithm, and interactive questioning system
• README.md – Lists available skill commands and quick-start references
• ROADMAP.md – Provides hour estimates consumed by the find-your-level skill for learning path calculations
• phases/<phase-dir>/docs/en.md – Source lesson files sampled by the check-understanding skill

Summary

• The find-your-level skill uses a 10-question, 5-domain assessment to generate scores between 0–10, mapping results to 20 possible curriculum phases via a defined table in lines 72–80 of its SKILL.md file.
• The check-understanding skill validates phase inputs against a directory mapping table (lines 31–50), samples lesson content strategically, and delivers 8 adaptive questions (4 conceptual, 4 practical) using the AskUserQuestion tool.
• Both skills output structured markdown tables, provide specific file-path references for remediation, and integrate with the repository's ROADMAP.md for hour estimation and curriculum navigation.

Frequently Asked Questions

How does the find-your-level skill determine which curriculum phase to recommend?

The skill calculates a 0–10 point total across five knowledge areas, then consults a hardcoded mapping table in lines 72–80 of .claude/skills/find-your-level/SKILL.md. Score ranges correspond to specific phases—for instance, 4–5 points route to Phase 3 (Deep Learning Core), while a perfect 10 advances the learner to Phase 14 (Agent Engineering).

What happens if I request a quiz for a phase that doesn't exist in check-understanding?

The skill parses your input against a resolution table defined in lines 31–50 of .claude/skills/check-understanding/SKILL.md. If the phase name or number is invalid, it returns a helpful error message listing all valid options from 0 (Setup & Tooling) through 19 (Capstone Projects).

How many questions does each skill generate, and what types are they?

The find-your-level skill generates exactly 10 questions—two per knowledge domain (Math & Statistics, Classical ML, Deep Learning, NLP & Transformers, Applied AI). The check-understanding skill generates 8 questions per phase: four conceptual (What/Why) and four practical (How/Build), each with 3–4 multiple-choice options.

Can the check-understanding skill handle phases with many lessons?

Yes. According to lines 58–63 of the skill definition, when processing large phases, the skill implements content sampling—it reads the first, middle, and last lesson files within the phase directory to generate questions without overwhelming the context window or the learner.

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