Flashcard Generator

The Minimum Information Principle is the rule that does most of the work: each card tests one piece of information. Not two. Not a concept plus its example. A card that asks "What is TCP and how does it differ from UDP?" looks reasonable and is functionally broken — there are four testable things inside it, and the spaced-repetition algorithm can't tell which one you actually remembered. The skill decomposes paragraphs into atomic facts before writing anything, and the resulting cards short-circuit the most common reason flashcard decks fail.

Card types match the kind of knowledge being tested. Basic cards for definitions and facts. Reversed pairs for vocabulary and terminology — passive recognition and active recall are different skills. Cloze deletions for formulas, lists, and process steps where the precise wording matters. Application cards for scenarios — the hardest to write, the most valuable to study, because they test transfer rather than memorisation. Comparison cards for distinguishing concepts that get confused. The skill picks the right type for each piece of source material rather than defaulting to question-and-answer for everything.

There's a hard gate at the start. If you're asking for cards on a topic you haven't actually learned — vague request, no source material, your own explanation reveals fundamental confusion — the skill refuses. Memorising what you don't understand doesn't stick, and creating cards under those conditions is actively harmful because it gives the illusion of studying. The skill redirects you to learn the topic first, then come back with source material.

Output is Anki-ready: tab-separated import format, proper cloze syntax, hierarchical tags for targeted study (cs::networking::tcp), and a study schedule calibrated to deck size — never more than 20 new cards a day, because review backlog compounds and 30/day creates an unsustainable load within two weeks. Plus warnings about the common frustration points: cards 30 to 50 will feel harder because they test similar concepts, that's normal, push through; if a card keeps failing, rewrite the card rather than brute-forcing memorisation. The card is probably the problem.