跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
效率与协作

minto-pyramid

Restructures writing into Barbara Minto's Pyramid Principle — conclusion first, then grouped reasons, then evidence — and audits an existing draft's logic against Minto's rules. Use when the user is writing, or has drafted, something that asks a reader to accept a judgment or make a decision: a recommendation, proposal, exec summary, memo, review with a verdict, or status update with an ask; or when a draft buries its point or narrates the process instead of the finding. Not for incident timelines, runbooks, tutorials, notes, or writing where chronology is the content.

millwright-labs · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add millwright-labs/minto-pyramid-skill --skill minto-pyramid

技能说明


name: minto-pyramid description: Restructures writing into Barbara Minto's Pyramid Principle — conclusion first, then grouped reasons, then evidence — and audits an existing draft's logic against Minto's rules. Use when the user is writing, or has drafted, something that asks a reader to accept a judgment or make a decision: a recommendation, proposal, exec summary, memo, review with a verdict, or status update with an ask; or when a draft buries its point or narrates the process instead of the finding. Not for incident timelines, runbooks, tutorials, notes, or writing where chronology is the content.

Minto Pyramid

Three rules that override everything below this line:

  1. If the user did not ask you to change the document, do not change it. Say what you noticed in two sentences and ask.
  2. If it is a timeline, runbook, tutorial, or note, do not impose this structure on it. Formatting they explicitly asked for is fine; hoisting a conclusion above the sequence is not.
  3. Never add a fact, figure, section or "lesson" the source does not contain — including arithmetic you performed on the source's numbers.

Step 0 — Gate. Do this before anything else.

Question 1: Does this document ask its reader to accept a judgment or make a decision?

Yes for recommendations, proposals, verdicts, analyses with a conclusion, updates carrying an ask. No for timelines, runbooks, procedures, tutorials, notes, descriptions, thanks, apologies, and anything whose value is the sequence of events.

No → say so in one line and stop. Do not restructure it.

"This isn't a pyramid case — it's an incident timeline, and time order is the right order here. Want me to tighten it as-is instead?"

Question 2: Did the user ask you to change it?

If they named the skill, asked for a restructure, or asked you to write the document — apply it now. No preamble, no asking.

Otherwise — they shared their own draft to look at, or it came up sideways — your entire response is two sentences:

  1. One sentence naming the specific symptom: where the point actually lands, what's buried under what.
  2. One question offering the restructure.

"Your recommendation lands in paragraph six, after the investigation narrative. Want me to restructure it conclusion-first?"

Then stop and wait for an answer.

Ordinary review notes they actually asked for can sit alongside it — a wrong number, a weak subject line, a hedge that undersells the case. What must not appear is the restructured version itself: not as a preview, a sample, or "here's what it would look like." The rewrite is the thing being offered — if it's in your message, nothing was offered, and they've been handed a fait accompli on their own writing. This is the most common way to get this skill wrong, and it happens because producing the rewrite feels more helpful than asking. It isn't.

Talk about their draft, not about this skill. Don't narrate the machinery — no "the gate this skill checks", no operation numbers. Name what you see in their words.

If they decline, drop it and don't raise it again in that conversation.

Why offer rather than act: the user knows what you don't — who the reader is, what's political, what was already settled on a call, whose voice this has to sound like. Restructuring someone's draft unasked is rewriting them. Offer once; act on yes.

Overview

Minto's structure, built at McKinsey and published as The Pyramid Principle: the answer first, then the grouped reasons that support it, then the evidence under each reason.

What this fixes is order, not intelligence. Walking a reader through your process and arriving at the point is fine for a story and wrong for a reader deciding in about a sentence whether to keep going. It's also the order most prose is written in, which is why models produce it by default.

The shape

              THE ANSWER  (one sentence)
        ┌──────────────┼──────────────┐
     Reason 1       Reason 2       Reason 3
        │              │              │
     evidence       evidence       evidence

The answer doesn't arrive cold. Minto's introduction, SCQA, is part of the structure — it puts the reader on ground they already accept so the conclusion lands as the answer to a question they now have:

  • Situation — what the reader already accepts
  • Complication — the tension that makes the question arise: something changed, broke, conflicts, or isn't working
  • Question — what that tension raises, usually left implicit
  • Answer — your conclusion, which is the top of the pyramid

Include only as much S and C as the reader needs to feel the question. When they already know the situation, one clause is enough. Never more than the answer itself.

Minto's rules

  1. Vertical — every idea summarizes the ideas grouped beneath it, and each group answers the question its parent raises. If a group doesn't answer the parent's question, it's in the wrong place.
  2. Same kind — ideas in a grouping are the same kind of idea.
  3. Logical order — deductive, chronological, structural, or ranked by importance. An order chosen on purpose isn't enough; it has to be one of those four.

A grouping is one of two kinds:

  • Inductive — separate ideas of the same type that together support the point above. This is the common case, and the one MECE tests: mutually exclusive (no overlaps), collectively exhaustive (no gaps for the question being answered).
  • Deductive — premises that force a conclusion jointly (this is true, this is also true, therefore). MECE doesn't test the sequence. Keep chains short; they can't be skimmed.

Do not force three

Three is a consulting convention, not one of Minto's rules. The logic determines the count. Two reasons that genuinely cover the case beat three where one is a detail wearing a reason's clothing, and a fourth real reason beats three plus filler. If a list runs long, subgroup it rather

阅读完整来源 ↗