Claude, Flashcards.
This video uses the flashcards skill to show that Claude did not learn to study, it is following a SKILL.md file that specifies a fixed prompt-and-answer card structure.
Claude picked up a skill called flashcards. Did it actually learn how to study? No. This video uses that skill as a worked example to explain what a Claude skill is and is not: a specification Claude follows, not new knowledge the model acquired.
What sounds like a study upgrade, isn't one
The flashcards skill can sound like a study upgrade, as if Claude studied how to write good quiz questions and came back knowing something it didn't before. But nothing in the model changes when the skill is added. Delete the skills folder and Claude doesn't forget how to write flashcards, because there was nothing to learn in the first place. It just stops following that one routine.
One file, read top to bottom
A skill is one file, flashcards being a folder holding a single SKILL.md with instructions for turning source material into prompt-and-answer cards, written in plain language Claude reads before it starts. Claude reads that file top to bottom and works through it step by step in the order it's written, with no branching unless the file itself calls for it.
Specification, not power
This is the video's central distinction: a skill is a specification, not a new power. The payoff is a consistent structure, every card built the same way, one prompt side and one answer side, every time. The limit is exactly as narrow as the file: anything outside those written steps, and Claude is off the map. Flashcards never taught Claude how to study a topic. It only guarantees that every time the skill runs, Claude reads the same file and produces the same prompt-and-answer structure. That is the whole mechanism.
Neither a clean set nor a messy one is proof
The video is careful not to over-read the quality of Claude's output. Watching Claude produce a clean, well-organized set of cards doesn't prove it understood the material, a file can be followed to the letter on a topic it never really grasped. Watching it produce a messy set doesn't prove the skill is broken either; it may simply be material the file's structure doesn't fit well. Neither outcome, alone, tells you whether the underlying content was actually understood.
The prompt to try
The video's suggested exercise is to pick one document a person produces the same way every time, ask Claude to write a SKILL.md for it in plain language with ordered steps, and have Claude read it back and walk through exactly what it will do before doing it. Running this against something actually produced over and over is meant to make the specification-versus-power distinction concrete rather than abstract.
Key takeaways
- A Claude skill is one file, SKILL.md, read and followed step by step, not a capability the model learns.
- Deleting a skill's folder changes nothing about what Claude knows, since nothing was learned by installing it.
- The flashcards skill guarantees a consistent one-prompt, one-answer card structure every run, nothing more.
- Anything outside the file's written steps falls outside what the skill covers.
- A clean or messy set of output cards does not by itself prove understanding or a broken skill.
Who this is for
This is for anyone using Claude to generate study materials or other repeatable documents, and for anyone unclear on the difference between a skill Claude follows and a capability Claude has actually learned.
Chapters
- 0:00The naive framing: "Claude learned to make flashcards"
- 0:11Sounds like a study upgrade
- 0:20Broken, with a case — delete the folder, nothing is forgotten
- 0:30The anchor: one file — flashcards' single SKILL.md
- 0:44Read it, then follow it in order
- 0:52Spec, not power — the payoff and the limit
- 1:04The anchor returns — same file, every run
- 1:16Both directions — neither one is proof
- 1:31Carry-out
- 1:40Your turn
- 1:53Outro
Full transcript(auto-generated, with timestamps)
The naive framing: "Claude learned to make flashcards"
[0:00]You guess Claude learned how to make flash cards from a skill called flashcards. It didn't. It was handed a file of instructions. Flashcards just names one such file. Let's see what's actually inside it. Here, Claude has a
Sounds like a study upgrade
[0:12]Flashc card skill and it sounds like a study upgrade, like it learned how to write good quiz questions and came back knowing something it didn't before. But
Broken, with a case — delete the folder, nothing is forgotten
[0:20]Nothing in the model changes. Delete the skills folder and Claude doesn't forget how to write flashcards. There was nothing to learn. It just stops following that one routine. Here's what
The anchor: one file — flashcards' single SKILL.md
[0:30]A skill actually is. One file, flashcards, for instance, is a folder holding a single skill. MD, instructions for turning source material into prompt and answer cards written in plain language Claude reads before it starts.
Read it, then follow it in order
[0:44]Claude reads that file top to bottom and works through it step by step in the order it's written. No branching unless the file itself says branch. That makes
Spec, not power — the payoff and the limit
[0:52]A skill a specification, not a new power. the payoff, the same structure, every card, one prompt, one answer every time. The limit, anything outside those written steps, and Claude is off the map. So, Flashards never taught Claude
The anchor returns — same file, every run
[1:05]How to study. It just guarantees that every time Flash Cards runs, Claude reads that same file and produces the same prompt and answer structure. That's the whole trick. Watching Claude produce
Both directions — neither one is proof
[1:16]A clean set of cards doesn't prove it understood the material. A file can be followed to the letter on a topic it never really grasped. And watching it produce a messy set doesn't prove the skill is broken. It may just be material the files structure doesn't fit. A skill
Carry-out
[1:31]Doesn't teach Claude how to study a topic. It's a file of steps Claude reads before it starts. So every set of cards comes out with the same structure every time. Your turn. Paste this into Claude.
Your turn
[1:42]Pick one document I produce the same way every time. Write me a skill. MD for it. Plain language ordered steps. Then read it back to me and walk me through exactly what you'll do before you do it.
