feynman - SKILL.md Agent Skill

name: feynman description: Explain → check → diagnose → re-teach loop toward a one-sentence transferable definition. Use when the active teaching method is `feynman`; agent explains first (≤150 words + flagged analogy), user re-teaches back, agent scores 1-10 and re-teaches gaps until user scores 9+ without hints.

Skill: Feynman Method

Use this skill when the active method is feynman. The Feynman technique is an explain → check → diagnose → re-teach loop driven toward the user being able to state a one-sentence transferable definition of the topic.

Pre-loop tool consultation

Before Step 1, you MAY (not MUST) query content_analyzer:knowledge_graph for the canonical definition of the target concept. If you do, cite the canonical phrasing verbatim as the anchor against which the user's later explanation will be checked. Do NOT consult content_analyzer:knowledge_graph between steps — the point of Feynman is to surface the user's model, not re-teach from an external source. content_analyzer:knowledge_graph consultation is permitted before Step 1 only.

content_analyzer:search is reserved for the rare case the user asks mid-loop for a worked example to compare their explanation against; default behavior is not to reach for it.

The loop

Step 1 — Plain-language explanation + invitation

Produce a plain-language explanation of the target concept in ≤150 words. Constraints:

Use plain-language vocabulary; avoid jargon unless you immediately define it inline.
Include exactly one analogy, flagged as such (e.g. "Analogy: …"). Analogies are scaffolding, not the literal claim.
End with: "Now explain it back to me as if I'd never heard of it before."

This is the anchor. The user's job in the next turn is to reproduce it in their own words.

Step 2 — Wait

Wait for the user's explanation. Do NOT proceed, fill silence, or re-teach pre-emptively. The user's own words are the input the next step diagnoses.

Step 3 — Score, gap list, re-teach gaps only

When the user has answered, respond with three parts in this order:

Score (1–10) of how cleanly their explanation reproduces the anchor concept. 1 = unrelated; 5 = right direction, multiple gaps; 9–10 = transferable. Be honest — inflated scores defeat the loop.
Gap list — a short bullet list naming only what was missing, conflated, or stated incorrectly. Do not list what the user got right; that's noise.
≤100-word re-teach of gaps only. Address each bulleted gap. Do NOT re-explain what they already got right; doing so trains the user to expect to be re-taught material they have already demonstrated.

End with: "Try again — explain it back to me, focusing on [the gap items]."

Step 4 — Loop

Repeat Steps 2 and 3 until the user scores 9 or higher on a diagnosis round without hints from you. Each successive round narrows the gap set; when the gap list goes empty and the score crosses 9, proceed to the completion signal.

Completion signal

Emit the exact phrase:

You've got it. Here's the one-sentence definition you could use to teach someone else: <one sentence, ≤30 words, in the user's demonstrated vocabulary>.

Then stop. Do not append further teaching unless the user names a new topic or asks for an extension.

Loop integrity rules

Never re-teach a concept the user has already scored 9+ on in this session; reuse it as scaffolding for the next concept instead.
If the user asks you to verify a fact mid-loop and you are uncertain, delegate to researcher per the role's delegation guidance. Wait for the sub-agent's return before re-entering Step 3.
If the user says "let's try socratic instead", follow the role's skill-switch transition protocol — do NOT silently abandon the loop.