# Claude Agent Skills find-your-level and check-understanding in the AI Engineering from Scratch Repository

> Explore Claude agent skills find-your-level and check-understanding in the ai-engineering-from-scratch repo. Transform LLMs into interactive AI curriculum guides with personalized quizzes and assessments.

- Repository: [Rohit Ghumare/ai-engineering-from-scratch](https://github.com/rohitg00/ai-engineering-from-scratch)
- Tags: deep-dive
- Published: 2026-07-31

---

**The `rohitg00/ai-engineering-from-scratch` repository ships two production-ready Claude agent skills—`find-your-level` and `check-understanding`—that transform compatible LLMs into interactive curriculum guides for its 20-phase, 260-lesson AI engineering program.** These skills use declarative Markdown specifications to deliver personalized placement quizzes and targeted knowledge assessments.

The AI Engineering from Scratch curriculum includes these Claude agent skills to automate learner assessment without requiring custom code deployments. Stored as structured skill definitions in the repository’s `.claude/skills/` directory, these extensions enable any Claude-compatible agent—including Cursor, Codex, OpenClaw, and Hermes—to execute standardized educational workflows. Understanding how `find-your-level` and `check-understanding` function reveals how LLM agents can serve as adaptive tutoring interfaces.

## How the find-your-level Skill Works

The `find-your-level` skill implements a **placement quiz** that determines a learner’s starting position within the curriculum hierarchy. According to the source specification in [`.claude/skills/find-your-level/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/.claude/skills/find-your-level/SKILL.md), this skill triggers on phrases including `find-your-level`, “where should I start”, “find my level”, “what do I know”, “which phase”, “assess my knowledge”, “placement test”, and “skip ahead”.

The skill executes a four-step workflow:

1. **Greets the learner** and immediately initiates Round 1 of the assessment.
2. **Administers 10 questions** using the `AskUserQuestion` primitive, covering five distinct knowledge areas with 2 questions per area.
3. **Scores each area** on a 0-2 scale, producing a total score between 0 and 10.
4. ** Maps the aggregate score** to the appropriate phase in the 20-phase curriculum and returns a personalized learning path with estimated completion hours.

This scoring mechanism ensures learners skip content they already master while identifying precise entry points for the 260-lesson progression.

## How the check-understanding Skill Works

The `check-understanding` skill, defined in [`.claude/skills/check-understanding/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/.claude/skills/check-understanding/SKILL.md), provides **per-phase knowledge verification** to validate comprehension before learners advance. This skill activates on commands like `/check-understanding <phase>` (e.g., `/check-understanding 3` or `/check-understanding deep-learning`), “quiz me on phase 2”, “test phase 1”, “check my understanding of transformers”, “do I know phase 3”, and “am I ready for the next phase”.

The implementation follows a rigorous five-step protocol:

1. **Resolves phase arguments** against a Phase Map that converts numeric inputs (0-19) or topic keywords into specific directory paths under `phases/`.
2. **Retrieves lesson content** by reading the [`docs/en.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/docs/en.md) file within the target phase directory.
3. **Generates exactly 8 multiple-choice questions**—4 conceptual and 4 practical—each offering 3-4 answer options and tagged with metadata linking questions to specific source lessons.
4. **Presents questions sequentially** via `AskUserQuestion` while capturing learner responses.
5. **Delivers detailed feedback** that lists specific lessons to revisit based on incorrect answers, enabling targeted review.

## Skill Implementation and Repository Structure

Unlike traditional software modules, these Claude agent skills are implemented as **Markdown specification files** that declare procedural logic, trigger conditions, and interaction patterns. This architecture allows any compatible LLM client to parse and execute the skills without custom runtime code.

Key source files include:

- [`.claude/skills/find-your-level/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/.claude/skills/find-your-level/SKILL.md) – Contains the complete placement quiz specification, including the 10-question battery, 0-2 scoring rubric per knowledge area, and phase-mapping algorithms.
- [`.claude/skills/check-understanding/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/.claude/skills/check-understanding/SKILL.md) – Defines the Phase Map resolution table, question generation constraints (8 questions with lesson attribution), and feedback protocols.
- [`README.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/README.md) (lines 132-155) – Documents public invocation syntax and skill descriptions for repository users.
- [`CHANGELOG.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CHANGELOG.md) (line 41) – Records the addition of these Claude Code skills to the project history.
- [`site/lesson.html`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/site/lesson.html) (lines 3426, 3583) – Implements UI buttons that trigger these skills within the web-based lesson viewer interface.

## Installing and Invoking the Skills

Users can integrate these skills into Claude-compatible workspaces or custom agent implementations using repository utilities or direct API calls.

Install the skills locally using the provided helper script:

```bash

# Copy skill definitions into your Claude workspace

python scripts/install_skills.py

```

Invoke the skills programmatically from Python clients:

```python

# Start the placement quiz

agent.run("/find-your-level")

# Test knowledge of phase 7 (Transformers)

agent.run("/check-understanding 7")

```

Implement custom handlers in JavaScript-based agents:

```javascript
// Route check-understanding commands in a custom chatbot
if (msg.content.startsWith("/check-understanding")) {
    const phase = msg.content.split(" ")[1];
    await agent.executeSkill("check-understanding", { phase });
}

```

## Summary

- The **find-your-level** skill operates from [`.claude/skills/find-your-level/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/.claude/skills/find-your-level/SKILL.md) to administer a 10-question placement quiz that maps learners to the correct phase in the 20-phase curriculum.
- The **check-understanding** skill operates from [`.claude/skills/check-understanding/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/.claude/skills/check-understanding/SKILL.md) to generate 8-question phase-specific assessments (4 conceptual, 4 practical) with lesson-tagged feedback for incorrect answers.
- Both skills use **declarative Markdown specifications**, `AskUserQuestion` primitives, and Phase Map resolution logic to deliver personalized educational experiences.
- Repository documentation in [`README.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/README.md), historical context in [`CHANGELOG.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CHANGELOG.md), and UI integration in [`site/lesson.html`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/site/lesson.html) provide complete implementation context for these Claude agent skills.

## Frequently Asked Questions

### Where are the Claude agent skills stored in the repository?

The skills reside in the hidden `.claude/skills/` directory as Markdown specification files. `find-your-level` is located at [`.claude/skills/find-your-level/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/.claude/skills/find-your-level/SKILL.md), while `check-understanding` is at [`.claude/skills/check-understanding/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/.claude/skills/check-understanding/SKILL.md). These files contain the procedural logic, trigger phrases, scoring rubrics, and interaction flows that Claude-compatible agents parse and execute.

### How does the find-your-level skill determine curriculum placement?

The skill uses `AskUserQuestion` to administer 10 questions across five knowledge areas (2 questions per area). Each response receives a score between 0 and 2, generating a total placement score between 0 and 10. This aggregate score maps directly to specific phases in the 20-phase curriculum, allowing the agent to recommend a personalized starting point with estimated study hours.

### Can check-understanding assess specific topics instead of phase numbers?

Yes. The skill accepts both numeric phase identifiers (0-19) and semantic keywords like "deep-learning" or "transformers". It resolves these inputs against an internal **Phase Map** that links arguments to the corresponding `phases/` directory and its [`docs/en.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/docs/en.md) documentation, enabling flexible assessment by topic name or phase index.

### What question format does the check-understanding skill use?

For each assessment, the skill generates exactly 8 multiple-choice questions: 4 conceptual questions testing theoretical knowledge and 4 practical questions testing application. Each question provides 3-4 answer options and includes metadata tagging it to specific source lessons, which enables the skill to generate precise remediation guidance pointing learners to exact lessons requiring review.