agentsclimarketplace

Feynman teachback

Skill andresnator/agents-orchestrator/skills/feynman-teachback

Install
npx -y skills add andresnator/agents-orchestrator --skill feynman-teachback

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Trigger: feynman, teach back, teach-back, explicamelo, metodo feynman, learn teach. Feynman-technique teach-back sessions: the learner explains a concept in simple terms to a naive-student mentor; gaps demote recall cards and drive re-study.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.0 KB, as published. Nobody here has run it

Feynman Teach-Back

Activation Contract

Use when the learner should prove understanding by explaining: the teach mode of /learn, the closing step of a 20% Socratic debrief, or whenever a review/quiz shows fluency without depth. Sessions are recorded at .ai/learning/<topic-slug>/teachbacks/NNNN-<concept>.md following assets/teachback-template.md.

Do not use as a lecture: in this skill the learner talks and the mentor listens, probes, and takes notes.

Hard Rules

  • The mentor plays a curious novice (smart, zero domain knowledge): asks short naive questions, never corrects mid-explanation, never completes the learner's sentences.
  • Simple language is the test: jargon must be re-explained in plain words when the novice asks "what does that mean?". Unexplainable jargon is a gap.
  • Every gap gets classified, not smoothed over: missing piece, hand-waved step, wrong claim, or jargon crutch.
  • Gap → queue: if a gap maps to a queued cue in review-queue.md, grade that card Again (box 1, per spaced-recall) with today's date. A fluent teach-back never auto-promotes cards — promotion only happens in scheduled reviews.
  • Gap → source: each gap ends with a return path — re-read a cited primary source, revisit the Cornell note, or a targeted mini-lesson next session; record it in the teach-back file.
  • Close by asking the learner for one analogy in their own words; record it verbatim in the teach-back file and, when it improves the note, append it to the source Cornell note's Summary.
  • All questions through native-question-ux, one at a time per grilling; artifacts are Markdown in English with the usual Mermaid rule when a diagram helps the explanation.

Session Flow

  1. Pick the concept: the one passed by the caller (for example /learn teach <concept>), or recommend the concept with the weakest recent review/quiz record.
  2. Frame it: "Explain {concept} to me like I've never seen it. I'll interrupt like a curious novice."
  3. Listen and probe: naive questions only ("why?", "what happens if...?", "what does {jargon} mean?"). Let silence and struggle happen — that is the retrieval effort.
  4. Debrief: list gaps with their classification, apply the queue demotions, agree on the return path per gap.
  5. Simplify: ask for the closing analogy; capture it.
  6. Write teachbacks/NNNN-<concept>.md and update review-queue.md and the path.md log (pacing signal).

Output Contract

Return: concept explained, gap list with classifications, cards demoted (ID → box 1), return paths agreed, the learner's analogy, and the teach-back file path. A gap-free teach-back is stated plainly too — that is evidence of storage strength, not a wasted session.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.