AGENT SKILL

feynman-technique

Verify your understanding of a concept by attempting an undergrad-level explanation; surface exactly where the explanation cracks and study those.

License 许可
MIT
最近更新
2026-10-29
Stars
★ 0
Forks
0

01 前言 Front Matter

name
feynman-technique
description
Verify your understanding of a concept by attempting an undergrad-level explanation; surface exactly where the explanation cracks and study those.

03 SKILL.md 正文

feynman-technique

Richard Feynman: if you can't explain it simply, you don't understand it. The point isn't the explanation — it's the cracks that surface during the attempt.

Mode: analytical — judgment inside a fixed frame. Keep the output shape; every verdict cites its evidence; unknowns are marked [TBD] or [VERIFY], never filled in.

When to use

  • You think you understand a concept and want to verify.
  • A team member is about to teach something and wants to pressure-test their grasp.
  • Onboarding: you've read the docs, now try to teach without them.

Inputs

  • The concept (required).
  • Your current understanding, in your own words (required — don't fetch a definition; produce one).

Procedure

Who writes what. When coaching a person, they write Attempt 1 and Attempt 2 — you mark the cracks and point at sources. Writing their attempt for them defeats the technique. Write the attempts yourself only when the user asks you to pressure-test your understanding.

  1. Step 1 — Attempt the explanation. Write the concept in your own words, aimed at a curious undergrad who has not heard the term. Limit: ~250 words. Don't open the source material.
  2. Step 2 — Identify the cracks. Walk through the explanation and mark:
    • Vague: "somehow", "it just works", "magically".
    • Jargon-leak: a term you used without grounding it.
    • Hand-waved: a step you can't actually defend. Be honest. Cracks are the point; pretending they're not there defeats the technique.
  3. Step 3 — Hit the source for exactly the cracks. Don't re-read the whole thing. Find the answer to each crack specifically.
  4. Step 4 — Re-attempt the explanation. Same constraints. Now the cracks should be gone or honestly noted ("we don't fully know — the literature says X").
  5. Step 5 — Test on a real person. Whatever you couldn't explain to them is the next round.

Output

feynman: <concept>

Attempt 1:
  <your ~250-word explanation>

Cracks identified:
  - VAGUE: "somehow caches the result" — by what mechanism? when invalidated?
  - JARGON-LEAK: used "memoization" without grounding.
  - HAND-WAVED: claimed it's O(1); is it amortized?

Sources consulted (only for the cracks above):
  - <ref> — explained the invalidation mechanism.
  - <ref> — confirmed amortized O(1).

Attempt 2:
  <revised ~250-word explanation; cracks closed or explicitly admitted>

Honest residuals:
  - "Why this specific replacement policy?" — couldn't find a clean answer.
    Note this for follow-up.

Quality bar

  • Attempt 1 is yours, not a paste of the docs.
  • Cracks are specific (line/word level), not "I'm fuzzy on this".
  • Source consultation is targeted to the cracks.
  • Attempt 2 is materially better than Attempt 1; if it's not, the cracks weren't real.
  • Honest residuals are listed. Pretending mastery defeats the technique.

Anti-patterns to avoid

  • "Attempt 1" copied from the source — that's reading, not testing.
  • Vague crack list ("I think I'm missing something"). Be specific.
  • Skipping Step 5. Real audiences expose the cracks self-review can't.

04 同仓库更多 Skills