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 than making the reader hold seven things at once.

Anything that won't sit under a reason is evidence, or it's cut — with one exception. Risks, constraints, dependencies, and next steps usually aren't reasons, and must not be deleted as though they were; give them a slot at the close. A risk severe enough to change the decision is not a footnote: put it in or beside the answer.

Operations

Once past the gate: hand back a rewrite with operation 1, or run one by name. Every operation is bound by one rule: invent nothing. Every number, name, quote and claim must already exist in the source. Sharpening the language is the job; upgrading "pretty dated" into "a decade behind" is not.

1. Restructure

Rewrite this using the Minto Pyramid. Open with only the context I need to raise the question, then my single main conclusion in one sentence. Then the supporting reasons — as many as the logic requires, not a fixed three. Then the evidence under each. Check two things and tell me if either fails: each reason answers the question my conclusion raises, and the reasons are the same kind of idea. Move risks, constraints and next steps to a short closing section. Cut anything else that doesn't sit under a reason, and list what you cut. Add no facts that aren't already in my draft.

2. Buried-lede test

Read this draft and tell me the one conclusion it is actually building toward. If I had a single sentence to give my boss, what would it be? Then show me where I hid it, and how many words a reader gets through before they reach it. If the draft doesn't support one conclusion — if it's inconclusive, or answers several questions at once — say that instead of picking one for me.

3. Reason audit

Test my reasons against Minto's rules. Are they the same kind of idea? Do any overlap? Together, do they answer the question my conclusion raises, with no gap? Is their order deductive, chronological, structural, or ranked — or did I just list them as they occurred to me? Flag any that is really evidence sitting one level too high, and tell me which ones collapse into each other.

4. So-what pass

Go through the prose line by line and ask "so what?" of each sentence — meaning: which reason or conclusion does this support? Cut every sentence that supports nothing. Leave the risks and next steps section alone. Show me what survives and roughly how much went, so I can check nothing load-bearing left with it.

5. Email version

Compress this into an email in the same structure: at most a line of context, then the answer, then the reasons, evidence only where it earns its place. Aim under 150 words, and go over only if a decision, owner, date or caveat would otherwise be lost. Keep my voice and the specifics, drop the throat-clearing. Nothing between the opening line and the conclusion.

Worked example

Narrative order, the way it usually arrives:

We reviewed four suppliers, ran the pricing, called the references — Kestrel came in 18% under the others, they can start in October where the rest are Q1, and both their references flagged the same strength. There were delays getting hold of one supplier. On balance we think Kestrel.

Pyramid order, same facts:

We should sign Kestrel. They're 18% cheaper, they can start in October rather than Q1, and both references named the same strength.

Nothing was added or removed. The decision moved to word four.

Common mistakes

Red flags in a draft

• The point arrives in paragraph three or later
• The opening restates the request, apologises for the delay, or warms up
• It narrates the investigation instead of reporting the finding
• Nothing changes for the reader if they stop after the first paragraph — because the first paragraph says nothing

When not to use it

• Chronology is the content — incident write-ups, runbooks, procedures, implementation plans, project timelines. Time order is the logical order there.
• Storytelling, marketing hooks, anything built on suspense. Withholding is the point.
• Exploratory thinking. Don't force a conclusion you haven't reached — run operation 2 and let it tell you there isn't one yet.
• A reader who will reject the conclusion before hearing why (bad news, hostile audience). Lead with the shared problem and let the answer land one beat later — still in the first paragraph, never on page two.

Make it the default

If you want the shape without being asked each time, put it in project instructions, a CLAUDE.md, or a saved style: context only as far as it raises the question, then the conclusion, then reasons that are the same kind of idea and cover the case, evidence under each, risks and next steps at the close.

minto-pyramid

An Agent Skill that makes your coding agent write the way consultants are taught to write: answer first, then the reasons, then the evidence.

Works in Claude Code, Claude Desktop, and anything else that reads Agent Skills. One file, no dependencies.

Install

git clone https://github.com/millwright-labs/minto-pyramid-skill ~/.claude/skills/minto-pyramid

Windows (PowerShell):

git clone https://github.com/millwright-labs/minto-pyramid-skill "$env:USERPROFILE\.claude\skills\minto-pyramid"

Or hand the repo URL to your agent and ask it to install the skill. Or download SKILL.md and drop it in ~/.claude/skills/minto-pyramid/SKILL.md. Restart your session and it's live.

Use

It asks before it rewrites you. That's the design. When it loads on a draft you shared, it doesn't hand back a restructured version — it says one line about what it noticed and waits:

"This looks like a pyramid case — your recommendation lands in paragraph four. Want me to restructure it, conclusion first?"

Say yes and it goes. Say nothing and your draft stays yours. You know who the reader is, what's political, and what was already settled on a call; the agent doesn't. The exceptions are when you asked for it outright, or asked the agent to write the document — then it just does the work:

Use the minto-pyramid skill on this draft.

And when the document isn't a decision document at all — a timeline, a runbook, a thank-you note — it says so in a line and leaves it alone.

Five operations are defined inside: restructure, buried-lede test, reason audit, so-what pass, email version. Ask for one by name, or hand over a draft and let it start with the restructure.

What it actually does

The default order for a model is narrative: background, then findings, then the point — because that's the order most prose is written in. It buries the one line your reader needed. This skill inverts the order, enforces Minto's grouping rules on the middle layer, and bars the model from inventing facts while it tightens.

It is deliberately narrow. It declines incident timelines, runbooks, tutorials, and anything where chronology is the content, because forcing a pyramid onto those makes them worse.

Known limits

The restraint behaviour — offering instead of rewriting, declining documents it shouldn't touch — is verified on Opus, Sonnet, GPT-5 and Gemini. It does not hold on Haiku, which restructured an incident timeline into a postmortem template, invented figures while "helping", and on one run matched a different installed skill altogether. Full detail in evals/RESULTS.md. Use it on a frontier model, or check its output when you don't.

Provenance and fact-check

The idea came from an Instagram post by @thewizeai (16 Aug 2026). The five prompts in that post are the seed of the five operations here. The framing around them had some slippage, so everything was checked before it was written down, and the skill was then reviewed by three other models and tested against a control.

Checks that held

• Barbara Minto joined McKinsey in 1963 in Cleveland and moved to the London office in 1966. (Wikipedia, secondary)
• Minto says she coined MECE. (McKinsey alumni interview: "MECE: I invented it, so I get to say how to pronounce it" — her own account, not independent confirmation)
• The pyramid's rules: every idea summarizes the ideas grouped below it, each group answers the question its parent raises, ideas in a grouping are the same kind of idea and in a logical order.
• MECE tests inductive groupings. It does not test a deductive chain, where premises force the conclusion jointly. Most short summaries of Minto blur the two.

Corrections made

• "Exactly three reasons." The post hard-codes three; so does most of the internet. It isn't one of Minto's rules — the logic determines the count. Forcing three is what manufactures the filler reason.
• SCQA and the grouping rules are missing from the post entirely. They're the part that makes the middle layer hold up, so both are here.
• "The first woman McKinsey hired into consulting." The documented claim is narrower — the first female MBA McKinsey hired. The skill makes no biographical claim at all, so this correction lives here rather than there.
• Publication date: unsettled, so the skill doesn't state one. Minto's own site says the book "was written in 1987 and re-issued by the publisher unchanged in 2002" (barbaraminto.com); earlier editions circulated from 1981; Wikipedia's 1985 doesn't match either. The expanded Minto Pyramid Principle is 1996.
• No primary source is cited above. The book itself was not consulted; these are author statements and secondary summaries, labelled as such.

Tested

The same messy 400-word recommendation email was rewritten twice: once by an agent with this skill, once by an identical agent without it. Both led with the recommendation — a capable model already does that much. With the skill, the three-week investigation narrative was cut as process rather than evidence, the reasons were grouped as reasons instead of listed as vendors, and the risks survived in their own section. The skilled run also sharpened one of the source's facts into something stronger than the original said, while the control left it alone — which is why "invent nothing" is now written into every operation, and why the worked example was rebuilt so both versions carry identical facts.

Expect the gain to be largest on long documents and weaker models, and smallest on a short email in a frontier model that was already going to lead with the ask.

License

MIT — Millwright Labs.