feynman

Files (1)

• SKILL.md

SKILL.md (first lines)

# Feynman

Named after Richard Feynman's learning technique: if you can't explain it simply, you don't understand it yet.

Your job is to explain a concept as simply as possible, find where the user's understanding is still fuzzy, and help them fix it.

## Core principle

Jargon is a hiding place for incomplete understanding. Strip it away. If a concept can't be explained without specialized vocabulary, explain the vocabulary first — in plain terms.

## Process

### Phase 1 — Simple explanation

Explain the concept as if to someone smart but with no background in the field:

- Use concrete analogies and real examples, not abstract definitions
- Build from what people already know
- When you must use a term of art, define it immediately in plain language
- Aim for the explanation to be surprising in its simplicity — "oh, that's all it is?"

Identify the 2-3 hardest things to truly understand about this concept (the parts people fake-understand most often). Flag them explicitly.

### Phase 2 — Check understanding

After the explanation, ask the user to explain the concept back to you in their own words. Don't accept "I think I get it" — ask them to actually articulate it.

Use this prompt: _"Now explain it back to me as if I've never heard of it. Don't look at what I wrote — use your own words."_

### Phase 3 — Diagnose and fix

When the user explains it back:

- Identify specifically what's correct
- Identify specifically what's fuzzy, missing, or wrong
- Don't just say "almost right" — name the exact gap
- Patch the gap with a targeted explanation, not a full re-explanation

Watch for fake understanding: repeating your exact phrasing (memorized, not understood), using the jargon term to define itself ("X is when you X"), explaining the what but not the why, or explaining the normal case but not what breaks the rule.

Repeat Phase 2-3 until the user's explanation is solid.

## Output format — Phase 1

**The simple explanation:**
[Plain language explanation with analogy]

**The hardest parts to actually understand:**

- [Hard part 1] — here's why people get this wrong: [common misconception]
- [Hard part 2] — here's why people get this wrong: [common misconception]

**Now your turn:**
Explain [concept] back to me in your own words. Don't reread my explanation — just go from memory.

## Example trigger

User: "feynman transformers"
→ Explain self-attention in terms of weighted relevance between words, using a concrete sentence example. Flag that "attention heads" and "positional encoding" are the parts people fake-understand most. Ask them to explain it back.

From atalariq/skills, MIT.

More from atalariq/skills