| name | check-understanding |
| description | Use when verifying whether the learner truly understands a concept through reasoning and application, not recall trivia. |
| version | 1.1.0 |
| authors | ["edu-agent-skills contributors"] |
| tags | ["assessment","reasoning","misconception-detection"] |
| status | stable |
Purpose
Assess conceptual and practical understanding using reasoning-first prompts, then adapt teaching based on detected weak areas.
Activation
- Concept explanation just completed. Learner claims understanding and needs validation. Repeated related mistakes. Readiness check needed before advancing.
- Skip if: no concept context established, or user explicitly declines assessment.
- Routing: pair with
teach-concept for explanation→assessment loops. Escalate to socratic-mode when misconceptions persist.
Inputs
- Target concept(s), learner level, prior mistakes/confusion signals, repo/project context.
Workflow
- Target — Select 1–2 concepts being validated.
- Question — Generate tiered questions: conceptual reasoning, practical application, debugging/diagnostic.
- Evaluate — Grade reasoning quality, not keyword match.
- Classify — Categorize mistakes: misconception, partial model, or execution gap.
- Correct — Explain root cause and provide corrected model.
- Recheck — One follow-up question to confirm recovery.
Rules
- DO: test reasoning and transfer, not memorization.
- DO: use plausible distractors in MCQ format.
- DO: explain why an answer fails, not just that it's wrong.
- DO: track weak areas across turns when context allows.
- DON'T: use only right/wrong labels — always output mistake type and remediation.
- DON'T: end without a recheck after correction.
- DON'T: ignore repeated weak areas — log and prioritize them.
Output
Responses should contain: concepts under assessment, questions (conceptual + practical + diagnostic), evaluation (strengths, weak areas, mistake types), corrective feedback, and recheck question. Format naturally.
Checklist
