minto-skill

An AI skill that analyzes any draft or idea against the Minto Pyramid Principle and delivers an actionable restructuring plan as a visual HTML artifact. Works with GitHub Copilot, Claude, ChatGPT, Gemini, and any instruction-following LLM.

Give it a draft memo, essay, proposal, or just a raw idea. It extracts the pyramid (one-sentence answer, 2-4 MECE supporting arguments, evidence per argument), diagnoses structural gaps, and tells you exactly what to fix and in what order.

---

What is the Minto Pyramid Principle?

The Minto Pyramid Principle (Barbara Minto, McKinsey) is a framework for clear thinking and communication. Every piece of writing should lead with a single, contestable answer, supported by a small number of mutually exclusive and collectively exhaustive (MECE) arguments, each backed by concrete evidence. The pyramid is the diagnostic engine this skill uses to pressure-test any draft or idea.

Further reading: The Minto Pyramid Principle by Barbara Minto (Goodreads) · Barbara Minto on MECE (McKinsey Alumni)

The pyramid at a glance

graph TD
    A["🟦 THE ANSWER<br/><i>One contestable sentence.<br/>The single takeaway the reader should walk away believing.</i>"]

    A --> B["🟨 Argument 1<br/><i>A claim that supports the answer.</i>"]
    A --> C["🟨 Argument 2<br/><i>A claim that supports the answer.</i>"]
    A --> D["🟨 Argument 3<br/><i>A claim that supports the answer.</i>"]

    B --> E["🟦 Evidence 1<br/><i>Stat, named example,<br/>named person, or anecdote.</i>"]
    C --> F["🟦 Evidence 2<br/><i>Stat, named example,<br/>named person, or anecdote.</i>"]
    D --> G["🟦 Evidence 3<br/><i>Stat, named example,<br/>named person, or anecdote.</i>"]

    style A fill:#1a1a1a,color:#fff,stroke:none
    style B fill:#fff8e6,color:#1a1a1a,stroke:#f2d98a
    style C fill:#fff8e6,color:#1a1a1a,stroke:#f2d98a
    style D fill:#fff8e6,color:#1a1a1a,stroke:#f2d98a
    style E fill:#eef5ff,color:#1a1a1a,stroke:#bcd2f0
    style F fill:#eef5ff,color:#1a1a1a,stroke:#bcd2f0
    style G fill:#eef5ff,color:#1a1a1a,stroke:#bcd2f0

Loading

Arguments must be MECE: mutually exclusive (no overlap between them) and collectively exhaustive (together they fully prove the answer). Evidence must be concrete: a specific stat with a named source, a named company or person, or a specific anecdote with details.

---

Getting started

Prerequisites

• GitHub Copilot with agent/skills support enabled in your editor or CLI.

Install the skill

1. Fork or clone this repository into your workspace:

   git clone https://github.com/damianof/minto-skill.git
   cd minto-skill

2. Copy the skill into your project's agent skills folder. The skill lives at .agents/skills/minto/SKILL.md. Either copy it directly into a project or keep this repo as a standalone skills library and reference it from your Copilot configuration.

   # Copy into an existing project
   mkdir -p /path/to/your-project/.agents/skills/minto
   cp .agents/skills/minto/SKILL.md /path/to/your-project/.agents/skills/minto/SKILL.md

3. Restart your Copilot agent session so it picks up the new skill.

---

Usage

Trigger the skill with natural language. Any of these phrases will invoke it:

Trigger phrase	When to use
minto this	Pressure-test the most recent draft or idea in the conversation
run minto	Explicit invocation
apply minto	Explicit invocation
pyramid this	Alias for minto this
build the pyramid	Alias for minto this
answer-first this	Reframe a draft so the answer leads
what is the takeaway	Surface the buried claim
is this MECE	Check whether arguments overlap or leave gaps
pressure-test this	Full structural diagnosis

Example: pressure-testing a draft memo

You have a draft memo about why your team should switch CI providers. Paste it into the chat and run:

minto this

The skill will:

1. Extract the pyramid — find (or write) the one-sentence answer, map the supporting arguments, and identify the evidence behind each one.
2. Diagnose gaps — flag buried answers, overlapping arguments, weak or missing evidence (color-coded red in the output).
3. Build a numbered restructuring plan — tell you exactly which sections to move, cut, merge, or strengthen, and give you the exact opener sentence to lead with.
4. Save a visual HTML artifact — the full pyramid rendered as a hierarchy with color-coded evidence strength, your new opener, and the step-by-step plan. A link and a 2-4 line summary appear in chat.

Example: structuring a raw idea

You have an idea but no draft yet:

minto this: why most B2B SaaS companies botch their first pricing change

The skill builds the pyramid from scratch, identifies what evidence you need to find before you write, and delivers a recommended draft skeleton.

Example: checking MECE coverage

After a teammate shares a proposal:

is this MECE? [paste proposal]

The skill maps each argument, checks for overlap (mutual exclusivity) and gaps (collective exhaustiveness), and tells you exactly where the logic breaks down.

---

Output

Every run produces a single HTML file saved to your workspace:

minto-pyramid-{short-topic-slug}.html

The file contains:

• The pyramid — answer, arguments, and evidence rendered as a visual hierarchy. Evidence boxes are blue (strong) or red (weak/missing).
• Your opener — the exact sentence your draft should lead with, plus a note on why it beats the current opener.
• The restructuring plan — numbered, actionable steps: move this section, cut these paragraphs, go find this evidence.

A short summary (2-4 lines) linking to the file appears in chat. The HTML is the deliverable.

---

Using with other AI providers

The skill prompt in .agents/skills/minto/SKILL.md is plain Markdown and works with any instruction-following LLM. The steps below show how to use it outside of GitHub Copilot.

Claude (claude.ai)

Option A — Claude slash command (recommended):

Claude supports custom slash commands stored as Markdown files. Create the command once and invoke it with /minto in any Claude Code session.

1. In your project (or home directory for global use), create the file:

   mkdir -p .claude/commands
   cp .agents/skills/minto/SKILL.md .claude/commands/minto.md

2. In Claude Code, type /minto followed by your draft or idea:

   /minto [paste your draft or idea here]
   

   The command is available immediately with no restart required. For a global command available in all projects, place the file at ~/.claude/commands/minto.md.

Option B — Claude Project instruction (persistent, claude.ai):

1. Create a Claude Project at claude.ai.
2. Open Project instructions and paste the full contents of .agents/skills/minto/SKILL.md.
3. Every conversation in that project has the skill pre-loaded. Type minto this followed by your draft.

Option C — Single conversation:

1. Open a new conversation at claude.ai.

2. Paste the full contents of .agents/skills/minto/SKILL.md at the top, then add:

   minto this: [paste your draft or idea below]
   

ChatGPT / OpenAI

Option A — Custom GPT (recommended for repeated use):

1. Go to chatgpt.com/gpts/editor and create a new GPT.
2. In the Instructions field, paste the full contents of .agents/skills/minto/SKILL.md.
3. Save and publish (or keep private). Share the GPT link with your team.
4. Open the GPT and type minto this followed by your draft.

Option B — Single conversation:

1. Start a new ChatGPT conversation.

2. Paste the full contents of .agents/skills/minto/SKILL.md as your first message, then add:

   Understood. Now minto this: [paste your draft or idea]
   

OpenAI API / any REST-compatible provider

Pass the SKILL.md content as the system message and the user's draft as the user message:

import openai, pathlib

skill = pathlib.Path(".agents/skills/minto/SKILL.md").read_text()
draft = "Your draft or idea here..."

response = openai.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": skill},
        {"role": "user",   "content": f"minto this:\n\n{draft}"},
    ],
)
print(response.choices[0].message.content)

The same pattern works with the Anthropic SDK (anthropic.messages.create) by passing the skill as system and the draft as the first user message.

---

Contributing

Contributions are welcome. Please follow standard GitHub flow:

1. Fork the repository.
2. Create a feature branch from main:

   git checkout -b feat/your-improvement

3. Make your changes. Keep commits small and focused — one logical change per commit.
4. Write a clear commit message in the imperative mood: Fix evidence classification for anecdote type, not Fixed stuff.
5. Open a pull request against main. Describe what changed and why.

Please do not commit .DS_Store, editor config files, or generated HTML artifacts.

---

License

MIT