clarify · diff
git:20260817.0cc07d9 to git:20260817.f04083f
4 added, 32 removed. Audit A to A.
---
name: clarify
description: Turn words with two readings into one confirmed ask. Use when the user's words admit two readings that would build different things.
---
- ## Extract
-
- From the user's words or the conversation:
-
- - The **ask** as given in the context
- - Its **readings** — the two (or more) ways the words could be taken.
-
- ## Gate
-
- Proceed only when: the readings would build or do different things.
-
- Anything else, say which in one plain line — never clarify anyway:
-
- - Every reading leads to the same work → skip the pick and proceed with the work. Never stage a choice that changes nothing.
- - The ask is already concrete and the pull is to keep questioning it — that is challenging its decisions, not picking a reading → use **interrogate** skill.
- - The user holds something they can't yet put into words — no reading of the words captures it → use **elicit** skill.
-
- ## Stage
-
- 1. Name the readings apart — one line each, ending in what each would build or do differently.
- 2. Ask the user to pick.
-
- One exchange: readings out, pick back. Never ask a second question. If the user cannot pick, the words are not the problem — stop clarifying; use **elicit** skill.
-
- ## Output
-
- Take the pick as the confirmed ask. Restate it in one line.
+ Name the readings apart, each ending in what it would build or do differently. As short as they go with nothing that tells them apart lost — plain words. Then ask the user to pick.
- The pick resolved the words — now check whether the restated ask is clear enough to return to the work it interrupted: would anything still unresolved change what gets built?
+ One exchange: readings out, pick back. If the user cannot pick, the words are not the problem — stop.
- - Nothing would → proceed with the work.
- - Something would, and the user holds it but hasn't said it → use **elicit** skill.
- - Something would, and it's a decision inside the ask that needs defending → use **interrogate** skill.
+ Take the pick as the confirmed ask, restate it, and name what it leaves open. As short as it goes with nothing in the ask lost — plain words.
- Clarify's one exchange is spent — any further question runs under the named skill's procedure, never as another round of clarifying.
+ The ask is already concrete and the pull is to keep questioning its decisions → use **interrogate** skill. The user holds something no reading of their words captures → use **elicit** skill.