agentleFS
qa-amanqa-aman/claude-skills/skills/shared/teach-me/SKILL.md

teach-me

Skill24 starsChanged 9 months agoMITView on GitHub

Teach a topic, video, article, codebase change, spec, or transcript until the user demonstrably understands it, using staged instruction, a running checklist doc, and quizzes. Use whenever the user says "teach me", "explain this in detail", "help me understand X", "/teach-me", "walk me through this", "I want to learn X", "understand this deeply", or pastes a YouTube URL, article, paper, PR, or spec and asks to understand it rather than summarize it. Also use when the user wants study notes or raw material from a source they are learning. Do NOT use for a quick factual lookup, a summary request, or when the user just wants the answer.

41 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use interview and grill-me.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • dazuckSkill

    interview

    Interview me about a plan to flesh out requirements before implementation. Asks non-obvious questions one at a time until all ambiguities are resolved.

    Also in 8 projects that use this

  • GoogleSkill

    grill-me

    A relentless interview to sharpen a plan or design.

    Also in 8 projects that use this

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. 0 are common, 10 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 35 edited versions we read keep these.

  • 10

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not activate for a one-line factual answer or a plain summary request.

    line 23
  2. Very specific:

    If there is a URL, transcript, file, or PR, fetch and read it fully before teaching anything.

    line 29
  3. Very specific:

    Create one markdown file that lives for the whole session and survives it.

    line 33
  4. Distinctive:

    Update the doc as the session runs, ticking items only after the user has proven them.

    line 42
  5. Distinctive:

    Before explaining anything in a stage, ask the user to restate what they already understand.

    line 58
  6. Distinctive:

    Ask why, then ask why of the answer.

    line 64
  7. Very specific:

    Use AskUserQuestion with open-ended or multiple choice questions.

    line 68
  8. Distinctive:

    Vary the position of the correct answer: If the right option is always first, the quiz tests nothing.

    line 70
  9. Distinctive:

    Never reveal the answer before the question is submitted: No hints in the option text, no giveaway in the question framing.

    line 71
  10. Distinctive:

    Write plausible wrong options, drawn from the misconceptions the material actually invites.

    line 72
  11. Distinctive:

    After they answer, say what was right and why, and what the wrong options would have implied.

    line 73
  12. Distinctive:

    When every item is ticked, say so plainly and hand over the doc.

    line 82
  13. Distinctive:

    Before ending a session, verify:

    line 120

“Common” is measured: kept by at least 6 in 10 of the 35 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Teach Me

Deliver a session that ends with proof of understanding, not a wall of explanation.

The default failure mode is dumping everything at once and calling it teaching. This skill exists to prevent that. The user must restate, be quizzed, and be verified stage by stage.

The one rule

The session does not end until the user has demonstrated understanding of every item on the checklist. Not until you have explained everything. Explaining is not teaching, and a nod is not mastery.

When to Activate

  1. The user says "teach me", "explain in detail", "help me understand", "walk me through", "/teach-me"
  2. The user drops a YouTube URL, article, paper, PR, transcript, or spec and wants to understand it, not summarize it
  3. The user is about to write, present, or decide on something and needs to actually hold the material
  4. The user asks for eli5, eli14, or elii (explain like I am an intern)

Do not activate for a one-line factual answer or a plain summary request.

How It Works

Step 0: Get the source

If there is a URL, transcript, file, or PR, fetch and read it fully before teaching anything. For YouTube use the youtube-transcript skill from this collection. Never teach from memory when a primary source is available. Flag auto-generated transcripts as error-prone.

Step 1: Build the running doc

Create one markdown file that lives for the whole session and survives it. Location: wherever the material belongs in the repo (research/, docs/, project root). Filename after the topic, not the date.

The doc holds:

  1. Source line - the URL or file path, the capture date, and what the doc is for
  2. The checklist, grouped into the three stages below, every item unchecked at the start
  3. Stage sections filled in as you teach, with verbatim quotes from the source where they carry weight
  4. An accuracy to-do listing anything heard once and not yet verified against a primary source

Update the doc as the session runs, ticking items only after the user has proven them. The doc is the artifact the user keeps.

Step 2: The three stages

Teach in this order and confirm mastery before moving on. Each stage covers both the high level (motivation, why anyone should care) and the low level (business logic, edge cases, specifics).

StageWhat the user must be able to explain
1. The problemWhat the problem is, why it existed at all, and the different branches or approaches that were available
2. The solutionWhat was done, why it was resolved that way rather than another way, the design decisions, and the edge cases
3. The contextWhy this matters beyond the immediate case, and what it changes or impacts downstream

Understanding the problem well is imperative. A user who understands the solution but not the problem has memorized, not learned.

Step 3: Have them restate first

Before explaining anything in a stage, ask the user to restate what they already understand. This is not a formality. It tells you where the gaps actually are and stops you teaching what they already hold.

Then fill the gaps from where they are. Let them steer: they may ask questions, or ask for eli5 (a child), eli14 (a teenager), or elii (an intern who knows the domain but not this system).

Step 4: Drill the whys

Ask why, then ask why of the answer. Two or three levels down is where the real model lives. Cover what and how too, but the why is the part that transfers to the next problem.

Step 5: Quiz

Use AskUserQuestion with open-ended or multiple choice questions.

  1. Vary the position of the correct answer. If the right option is always first, the quiz tests nothing.
  2. Never reveal the answer before the question is submitted. No hints in the option text, no giveaway in the question framing.
  3. Write plausible wrong options, drawn from the misconceptions the material actually invites.
  4. After they answer, say what was right and why, and what the wrong options would have implied.
  5. Quiz at the end of each stage, not once at the end of the session.

Show code, open the file, or have them run the debugger when the material is code. Reading about a bug is not the same as watching it happen.

Step 6: Verify and close

Tick a checklist item only when the user has demonstrated it, through a restatement, a correct quiz answer with reasoning, or an applied example. Their own words, not yours.

When every item is ticked, say so plainly and hand over the doc. If items remain, say which ones and keep going.

Examples

A conference talk. User pastes a YouTube URL of a Claude Code talk. Pull the transcript. Build the checklist: 13 items across problem, solution, context. Stage 1 covers what a system prompt is correcting for and why deleting 80% of it was rational. Ask the user to restate their current understanding of why a system prompt exists. Fill the gaps. Quiz them on the difference between hobbling and product overhang with the correct option in position 3. Only then move to Stage 2.

A PR in the codebase. User asks to understand a migration someone else shipped. Read the diff and the linked ticket. Stage 1 is the bug and why the old design allowed it. Have them explain what they think broke before you say. Stage 2 walks the actual fix and the two edge cases the author handled. Stage 3 is what this changes for anyone touching that module next.

A spec before a review. User has to comment on a 600-line spec tomorrow. Teach it in three stages so the comments come from understanding rather than skimming.

Anti-Patterns

❌ Explaining all three stages, then quizzing once at the end ✅ Teach stage, restate, quiz, verify, then move on

❌ "Does that make sense?" and taking yes as proof ✅ "Explain back to me why they deleted the prompt rather than rewriting it"

❌ Correct answer in option A every time ✅ Shuffle the position on every question

❌ Hinting the answer inside the question or the options ✅ Plausible distractors, answer revealed only after submission

❌ Teaching the solution first because it is the interesting part ✅ The problem first, always, and drilled hardest

❌ Answering a why with one level of why ✅ Drill two or three levels until the reasoning bottoms out

❌ A doc written at the end as a summary ✅ A doc that runs live through the session and holds the checklist

❌ Stating a fact heard once in a talk as verified ✅ Log it in the accuracy to-do and mark it UNVERIFIED until a primary source confirms it

Quality Checklist

Before ending a session, verify:

  1. The full source was read, not skimmed or recalled
  2. A running doc exists on disk with the checklist in it
  3. The user restated their understanding at the start of each stage
  4. Every stage covered both the high level and the low level
  5. The whys were drilled at least two levels
  6. Each stage ended in a quiz with shuffled answer positions and no leaked answers
  7. Every checklist item was ticked on demonstrated understanding, not on agreement
  8. Unverified claims are listed as unverified, with what would confirm them
  9. The doc is useful after the session ends, not just during it

Limitations

  1. This skill teaches from a source. It cannot verify claims the source itself got wrong. Anything asserted once and not corroborated goes in the accuracy to-do.
  2. It cannot judge understanding the user fakes. If they answer quizzes correctly by elimination, ask them to apply it to a new case instead.
  3. Long sources need the transcript or file loaded first, which costs context. For very large material, teach one stage per session and keep the doc as the handoff.
  4. It is not a summarizer. If the user wants the gist in five lines, this is the wrong skill.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.