What Is the Check-Understanding Skill in AI Engineering from Scratch?

The check-understanding skill is an automated phase-quiz tool that tests learner comprehension by generating eight multiple-choice questions drawn directly from curriculum lesson documentation.

The check-understanding skill serves as a critical assessment component within the AI Engineering from Scratch repository by rohitg00. Designed to validate mastery of completed learning phases, this tool bridges the gap between passive content consumption and active knowledge recall by sourcing assessment questions exclusively from the actual lesson documentation stored in the repository.

Core Purpose of the Check-Understanding Skill

The primary function of this skill is to serve as a knowledge verification checkpoint at the end of each curriculum phase. When invoked, it generates a balanced assessment consisting of four conceptual and four practical multiple-choice questions specific to the requested phase. Unlike generic quizzes, every question is grounded in the actual lesson docs located in the phases/ directory, ensuring perfect alignment with the material covered in modules like 03-deep-learning-core or 02-ml-foundations.

According to the skill manifest in skills/check-understanding/SKILL.md (lines 4‑10), the tool not only tests knowledge but also guides learners through remediation, tracking scores and suggesting specific next steps based on performance.

How the Check-Understanding Skill Works

The skill follows a structured assessment pipeline defined in the SKILL.md specification.

Phase Resolution and Content Loading

First, the skill resolves the requested phase identifier—whether numeric like 3 or descriptive like deep-learning—to the corresponding directory (e.g., 03-deep-learning-core). It then dynamically loads the relevant lesson documentation from paths such as phases/03-deep-learning-core/03-backpropagation/docs/en.md to extract accurate, up-to-date content for question generation.

Quiz Generation and Interaction

As specified in SKILL.md (lines 64‑78), the skill generates exactly eight MCQs and presents them sequentially to the learner. Participants respond with single-letter answers (A|B|C|D). The system tracks responses in real-time, calculates the final score, and analyzes incorrect answers to identify specific knowledge gaps in conceptual versus practical understanding.

Feedback and Next Steps

Upon completion, the skill returns a detailed summary showing which questions were missed, explains the correct answers with references back to the source documentation, and suggests remediation actions—such as retaking the quiz, reviewing particular subtopics, or progressing to the next phase if mastery is demonstrated.

Invoking the Check-Understanding Skill

You can trigger this assessment through direct commands or natural language prompts, as defined in the skill's activation triggers.

Direct invocation uses the skill name followed by the phase identifier:

# Direct invocation (Codex host)

check-understanding 3

# or

/check-understanding deep-learning

Natural language triggers work in Claude Code or any host that parses the SKILL.md manifest:

"quiz me on phase 2"
"test phase 1"
"do I know phase 3"
"am I ready for the next phase"

Both methods cause the skill to resolve the phase, load the appropriate documentation, and initiate the eight-question assessment sequence.

Source Files and Implementation Details

The skill's behavior is fully specified in the repository's skill manifest system, with content dependencies spread across the curriculum structure.

SKILL.md Configuration

The complete specification—including activation triggers, phase mapping logic, quiz generation procedures, scoring algorithms, and post-quiz flow—resides in skills/check-understanding/SKILL.md. Lines 4‑10 define the skill's description and high-level procedure outline, while lines 64‑78 detail the specific steps for generating and scoring questions.

Content Dependencies

While the skill logic lives in the skills directory, it dynamically pulls content from phase-specific documentation. For example, questions about neural network backpropagation originate from phases/03-deep-learning-core/03-backpropagation/docs/en.md, ensuring the assessment remains synchronized with curriculum updates.

Repository Integration

The root README.md file lists the skill in the invocation table, serving as the high-level entry point for all available skills in the AI Engineering from Scratch repository. This integration ensures the check-understanding skill is discoverable alongside other learning tools in the ecosystem.

Summary

• The check-understanding skill functions as an automated phase-quiz assessment tool for the AI Engineering from Scratch curriculum.
• It generates eight multiple-choice questions (four conceptual, four practical) sourced directly from lesson documentation in the phases/ directory.
• Located in skills/check-understanding/SKILL.md, the skill supports both direct command invocation and natural language triggers parsed by compatible hosts.
• It provides immediate scoring, detailed feedback on missed concepts with references to source material, and personalized next-step recommendations.
• Content is dynamically extracted from phase directories like phases/03-deep-learning-core/ to ensure alignment with current course material rather than static question banks.

Frequently Asked Questions

How many questions does the check-understanding skill generate per phase?

The skill generates exactly eight multiple-choice questions for each assessment: four conceptual questions testing theoretical knowledge and four practical questions testing applied implementation skills. This balanced 50/50 split ensures comprehensive evaluation of both understanding and hands-on ability, as defined in the generation procedures within SKILL.md.

Can I use natural language to invoke the check-understanding skill?

Yes. While you can use direct commands like check-understanding 3, the skill also recognizes natural language triggers such as "quiz me on phase 2" or "am I ready for the next phase." These triggers are parsed by hosts like Claude Code that read the activation patterns defined in the SKILL.md manifest, making the tool accessible without memorizing specific syntax.

Where does the check-understanding skill get its quiz content?

All questions are generated from the actual lesson documentation stored in the repository. The skill loads content from specific phase subdirectories—for example, phases/03-deep-learning-core/03-backpropagation/docs/en.md—ensuring that every question accurately reflects the material taught in that phase rather than relying on external or outdated question banks.

What happens if I answer questions incorrectly in the check-understanding skill quiz?

The skill tracks your score throughout the session and provides specific feedback on missed items after completion. Rather than simply showing a percentage, it identifies which concepts need review and suggests concrete remediation actions, such as retaking the quiz, revisiting specific lesson docs, or proceeding to the next phase only if you demonstrate sufficient mastery of the current material.

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