explain git:20260817.0cc07d9A

Immutable. This exact content is served forever at /api/v1/blob/fed99ff6c6638022.

---
name: explain
description: Put one settled thing in the user's head, plainly. Use when the user does not hold something already settled.
---

## Extract

From the conversation/context/user:

- The **thing** — the settled design, term, behavior, or number to put across.
- The **evidence** it's missing — where the user said or showed they don't hold it.

## Gate

Proceed only when: the thing is settled and the user has said or shown they don't hold it.

Anything else, say which in one plain line — never explain anyway:

- The thing is not settled — still being weighed, designed, or decided → there is nothing settled to explain yet.
- The user already holds it → there is nothing missing to put across.
- The user holds it and you're the one missing it — a direct question would get it said → ask it and proceed with what it settles.
- The user holds it but cannot yet put it into words → use **elicit** skill.
- The user wants a verdict on it, not an understanding of it → use **judge** skill.

## Retrieve

Never explain from a guess. If the thing is not in front of you, retrieve it — read, search, look — before saying anything about it. Only when it cannot be retrieved, name what's missing in one plain line, ask the user for it, and wait — the explaining continues once they provide it.

## State

State the thing in one plain line — no jargon, no preamble, no build-up. Stop there: never push the full picture unpulled. Name the two or three directions the user could pull next, and wait.

## Answer

The user drives the loop: each iteration is one pull from them — a follow-up question, a "tell me more", an "I still don't get X".

- Answer each pull with one plain line first, depth only where pulled.
- "I still don't get it" means the line was wrong — find a different plain line, never add more words to the old one.
- After each line, wait for the next pull.

The loop ends when the thing has landed — the user says it back in their own words or acts on it — or when they move the work on without doing either.

## Output

The landed thing is the result — it lives in the user's head, not in anything handed back. When the loop ended by the user moving the work on, nothing verified it landed: the plain line stands as stated, never claimed as held.

Then check whether anything still unresolved would change what gets built:

- Nothing would → return to the work the explaining interrupted.
- The user holds something they cannot yet put into words → use **elicit** skill.
- Words in play admit two readings that would build different things → use **clarify** skill.
- A proposal is committed but its decisions stand undefended → use **interrogate** skill.
- A claim about something that exists needs a committed verdict → use **judge** skill.

The explaining is spent once the loop ends — anything further runs under the branch picked here, never as another round of explaining.