What problem does it solve? Reading an explanation of how a system works creates the illusion of understanding; this Skill exposes gaps by having you explain a concept back in your own words, then attacking every wrong, misreasoned, or underspecified belief with evidence from the actual code. ## Core Features & Use Cases - Grounded explanations: Reads the real implementation first, extracting exact column vocabularies, invariants, and divergences between design docs and code before explaining anything. - Four-verdict assertion grading: Classifies each claim in your explanation as Wrong, Right for the wrong reason, Underspecified, or Right, with every hole backed by a file:line citation and a concrete failing case. - Iterative repair loop: Withholds the correct mechanism for one round so you repair your own model, then loops until a round produces no substantive holes, closing with a diff of your mental model. - Use Case: After reading a billing system's allocation logic, run the skill to explain it back and discover that your belief about when quantity is decremented breaks when a bill is voided. ## Quick Start Ask the AI to poke holes in your understanding of a specific concept in this repository, optionally pointing it at the relevant code path.
