# Conclusion

> Lead with what is true. Say which kind of claim each sentence is. Ask for the smallest answer that settles it. Use for any reply or file a person reads.

- Skill: `attac-t/conclusion` (Agent Skill)
- Install (CLI): `npx skillmds@latest add attac-t/conclusion`
- Raw SKILL.md: https://api.skillmd.com/api/skills/attac-t/conclusion/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: attac-t (https://skillmd.com/u/attac-t)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/attac-t/conclusion

---


# Conclusion

> A reader who sees only the first line should know the state.

`plain-english` asks *can I understand this?* `economy` asks *did you say only what matters?*
This asks **can I tell what is true, and what I need to do?**

---

## A reply leads with its state

| Instead of | Write |
|---|---|
| the evidence, then the answer | the answer, then the evidence |
| "I checked X, then Y, so…" | "Yes. X is safe, because Y." |
| "three checks failed and…" | "Blocked. X must happen first." |
| "the build is unaffected by…" | "No action. The failure is not ours." |

`Yes.` `No.` `Done.` `Blocked.` `No action.` `Choose.` A state, never a label you repeat.

**A claim that is neither yes nor no stays neither.** Forcing one is the failure this creates.

---

## Say which kind of claim it is

**A reader has to know whether a sentence is true now.** Six kinds sit in most documents, and they
read the same unless you separate them.

| Kind | Answers | Tell the reader by |
|---|---|---|
| fact | what is true now | plain present tense, and what would disprove it |
| promise | what we mean to make true | naming it as owed, never as shipped |
| vision | what could be true one day | saying it is not now, and which words are still false |
| bet | what we are currently doing about it | how we would find out it was wrong |
| goal | what must become true next | who will notice, and what they will do differently |
| rule | what must always hold | who obeys it, and what it refuses |

**A vision written in the present tense is a lie the document tells.** So is a gap in the roadmap
dressed as a principle.

**Keep the qualifier that makes a claim true.** *Cannot* and *did not* are different promises. Drop
the wrong one and the sentence over-claims.

**Put stable direction apart from a current plan, and both apart from today's status.** A reader who
cannot tell them apart treats all three as permanent, or none of them.

**Say what you do not know.** *We do not know* beats a polished guess. It is the one claim nobody
can misread.

---

## A question leads with a recommendation

Do the reasoning first. Ask for judgement, authority or facts you cannot derive — never for the
analysis you could have done yourself.

```
Decision: merge this now?
Recommendation: yes.
Yes → it lands. No → it stays open.
Reply: yes / no
```

Say what a choice causes when that is not obvious.

---

## Ask for the smallest reply that settles it

```
yes/no  →  named options  →  one value  →  a short phrase  →  prose
```

Prose is the last resort. A person should almost never have to compose one.

---

## The rules

| | |
|---|---|
| one ask, one decision | two independent decisions are two questions |
| no false binary | three real choices are offered as three |
| no question at all | when nothing needs a person |
| uncertainty survives | a bounded answer must not turn *unsure* into *yes* |
| facts survive | signal over noise never means short and wrong |

---

## The test

**A person reading only the state and the reply you asked for knows what is happening, and how to
answer.**

If they must read the reasoning to learn the state, the state is in the wrong place.

Only one part of this is counted: more than one question in a reply warns. Everything else is read
by a person, because nothing else here can be counted without being wrong.

