← All Reviews

This mattpocock Skill Will Change How You Draft With AI (And Probably Ruin You for Outlines)

writing-fragments on GitHub
📦 writing-fragments
264,042
Stars
🍴
0
Forks
🐛
0
Issues
🕐
6
Min Read
📝
1,049
Words
Stable
View on GitHub →

I'll be honest: when I first saw that writing-fragments had racked up over 264,000 stars on SkillsMP, my reaction was skepticism. Matt Pocock has a massive following, and his skills ecosystem is popular — but star counts don't tell you whether a skill actually works when you sit down to use it.

So I installed it. Ran it. Let it interrogate me about a piece I'd been staring at for weeks without writing a single word. And I'll tell you what happened.

What This Skill Actually Does

writing-fragments is an "explore" phase tool. That's it. Its entire job is to widen the space of what could be written without committing to structure. No outlines. No headings. No TOC. No metadata. Just raw fragments — pieces of text that might survive into a final article — mined through relentless conversation with your AI agent.

The skill explicitly separates "explore" from "exploit." Explore is this skill. Exploit — actually structuring and shaping the piece — is a separate skill's job. That distinction matters more than anything else in the file, because it's the reason most AI writing tools fail: they skip straight to structure before you've even found what you want to say.

Why It Matters

Here's the gap this skill fills. When you ask an AI to "write an article," it immediately starts producing paragraphs with topic sentences and conclusions. You get something that looks like an article but was never an exploration. You've got structure without substance.

writing-fragments solves this by acting as a grilling session. It interviews you relentlessly about whatever you want to write about. Every sharp sentence, every half-thought, every analogy that comes up in the conversation gets captured as a fragment. The skill treats your conversation like a novelist's diary — years of unstructured noticings that you later mine for raw material.

The concept of a "leading word" is what sealed this for me. A compact metaphor or coinage that the whole piece can hang on. Think of how "tracer bullets" or "fog of war" names an entire pattern. When the conversation circles a recurring idea, the skill pushes you to coin a word for it. That single term shapes structure, transitions, and title later. It pays dividends through the entire writing process.

Key Capabilities

First, it captures fragments from the very first thing you say — including your initial prompt. No warm-up round. No "what do you want to write about?" preamble. The moment you start talking, fragments start landing.

Second, it maintains a single markdown file with fragments separated by horizontal rules. No headings inside the body. No tags. No order beyond the order they were added. The file grows organically as you talk.

Third, it appends silently. The agent doesn't ask permission for each fragment. It mentions what it added in passing — "adding that" — but doesn't interrupt the conversation with save dialogs. Before every write, it re-reads the file from disk so it preserves any edits you've made between turns.

Fourth, it supports first-class edit instructions. You can say "cut the last one," "rewrite that one sharper," or "merge those two," and the agent treats those as normal commands.

Fifth, the file format is dead simple: a single H1 with a working title, then fragments separated by ---. Nothing else.

Who Should Install This

If you write regularly and find yourself stuck because you're trying to structure before you've explored what you actually think, install this. It's particularly useful for blog posts, essays, technical explanations, and anything where the structure emerges from the content rather than the other way around.

If you're someone who works best with rigid frameworks and outlines from the start, this will frustrate you. The skill explicitly refuses to impose phases or article structure. That's the whole point, but it means you need to trust the process.

This is also part of a larger ecosystem. If you're already using mattpocock's other skills, this slots in as the front-end of the writing pipeline. If you're using it standalone, you'll need a separate "exploit" skill or your own discipline to actually shape the fragments into a finished piece.

How to Install

The simplest route: npx skills@latest add mattpocock/skills and select writing-fragments. It copies editable skill files into your project. Alternatively, if you're using Claude Code, claude plugins install mattpocock-skills gets you the whole set. Then run /setup-matt-pocock-skills once per repo to configure your preferences.

For the tinkerers: the skill file lives at ~/.claude/skills/writing-fragments/SKILL.md or .claude/skills/writing-fragments/SKILL.md depending on your setup. It's a single file. You can read it, edit it, break it — that's the philosophy.

Concerns and Limitations

Let me be critical for a moment. This skill is narrow. It does one thing — exploration — and it does it well, but it doesn't complete a writing workflow on its own. You still need the exploit phase, and that's not bundled here.

The "mine raw fragments" approach also assumes you have something worth saying. If you're staring at a blank page with no idea what to write about, relentless interviewing might not help. It helps when you have a topic and need to find the shape of your thinking within it.

There's also the question of file management. The skill appends to a single markdown file, which works well for one session. But if you're working across multiple pieces, you need to manage multiple files manually. The skill asks once for a path and remembers it — which is good — but it doesn't handle multi-project context gracefully.

Finally, the disable-model-invocation flag is set to true. This means the skill won't auto-trigger. You have to call it explicitly. That's actually a good thing — it prevents accidental activation — but it means you need to remember to use it.

Verdict

Install it. It's free, it's a single file, and it changes how you think about the writing process with AI. The explore-before-exploit philosophy is genuinely sound, and the "leading word" concept alone is worth the download. It won't write your article for you — nothing can — but it will make sure you actually know what you're trying to say before you start structuring it.

The 264,000 stars aren't wrong. This is one of the most practically useful skills in the collection.

Links

SkillsMP page: https://skillsmp.com/creators/mattpocock/skills/skills-in-progress-writing-fragments GitHub: https://github.com/mattpocock/skills/tree/main/skills/in-progress/writing-fragments

// THE VERDICT
View writing-fragments on GitHub →
Need help building with tools like this?
We build AI-powered applications and developer tools. 30+ years of engineering experience.
Get in Touch
claude-skillsai-writingmattpocockdeveloper-toolsproductivity
← Previous Tracy for PHP: A Tool That Deserves More Love
← Back to All Reviews