Twenty-Four Parts, Twenty-One Boxes, And 340 Skills In Four Of Them

“Most people never install a thing in Claude. The exact 24-part stack behind my business.” Then a diagram: Claude in the centre, three branches, each headed with a number. Plugins — 8 installed. Skills — 8 active. MCP — 8 connected.

Count the boxes under each header. There are seven. Seven, seven, and seven.

That is a small thing and we would normally leave it at a sentence, except that the number is the headline, and counting is the only way to read a card like this. So we counted the rest of it too — and the twenty-one boxes do not hold twenty-one things either.

What is actually in the boxes

Here is every entry on the card, as labelled, with its own description.

Four of the twenty-one are not one thing each

Read the left column’s own descriptions. Four of the seven plugins announce how much is inside them:

• claude-skills — “263+ skills, every platform”
• marketingskills — “40 tools, full growth ops”
• gstack — “23 dev tools, one install”
• superpowers — “Dev methodology, 14 skills”

By the card’s own labels that is 340-odd advertised items inside four boxes, sitting in a diagram next to boxes that hold exactly one thing each. “23 dev tools, one install” is the giveaway — the phrase is selling the ratio.

So “24-part stack” describes neither number you can get from the card. Twenty-one if you count boxes. Several hundred if you count what the boxes say they contain. The honest version is twenty-one installs, which is a different and more useful claim.

And the three columns are not three categories

This is why the bundling happens, and it is worth knowing before you copy anyone’s stack. From the plugins documentation:

“A Claude Code plugin is a directory of skills, agents, hooks, MCP servers, or other components that Claude Code installs and loads as one unit.”

“Use a plugin when you want several skills, subagents, hooks, or MCP servers packaged as one unit.”

“Skills, subagents, hooks, and MCP servers all work on their own, without a plugin.”

Plugins is not a sibling of Skills and MCP. It is the container for them. A plugin can ship skills, it can ship MCP servers, and most of the big ones do — which is exactly what “263+ skills” in the plugins column means.

Drawn as three equal branches, the diagram implies you need all three. The documentation implies the opposite: you reach for a plugin when you want several things as one unit, and a single skill in ~/.claude/skills/ needs no plugin at all.

The part a stack diagram cannot show you

Every box on this card is drawn as an asset. Here is the running cost, from the same documentation page:

“An enabled plugin is part of every session, not only the sessions where you use it.”

“For each skill, agent, and command that Claude can invoke on its own, the name and description are in Claude’s context on every turn so that Claude knows it exists. Those tokens count toward your usage and leave less room in the context window even in sessions where nothing from the plugin runs.”

“MCP servers the plugin defines run alongside each session where it’s enabled, and its hooks fire at their events.”

“What the plugin runs, it runs as you.”

Put that beside “263+ skills”. A plugin that ships hundreds of skills is paying for hundreds of names and descriptions in your context window on every turn of every session, including the ones where you are writing an email. That is not an argument against installing it. It is an argument for knowing the figure, and the documentation gives you three places to read it.

What the card gets right

Two things, and the first is the reason we did not write a different piece.

The fine print is exactly right, and it pre-empts the obvious attack: “A personal example, not a recommendation. Star and download counts are approximate and change often. Check each tool’s official site, terms and data policy before you install or connect it.” We have audited star counts on a card before and found them stale. This card declares the problem itself, so there is no point auditing it — which is how a disclaimer is supposed to work.

And the MCP column is honest about what it is. Seven connections, each described by what it reads or writes: “Reads channels, posts updates”, “Reads and writes your docs”, “Triages your LinkedIn DMs”. Those descriptions are permission statements in disguise, and they are more candid than most connector cards manage.

Four honest caveats

1. The box count is ours, from the image. We rendered each column at full height and counted: seven, seven, seven, against headers reading eight. If the card was cropped in posting, the missing three exist somewhere we cannot see.
2. “340-odd” is the card’s arithmetic, not a verified inventory. We added up the numbers printed on four boxes. We did not install anything to check that 263+ skills arrive.
3. Bundles are not a bad thing. One install beating twenty-three is the whole point of the format. The issue is counting a bundle as one part in a headline that promises parts.
4. Context cost varies by plugin and by model. We quote the mechanism, not a number. The figure for any specific plugin is the one in your own /plugin panel.

The one-line version

The headline says twenty-four, the diagram shows twenty-one, four of those twenty-one announce 340 items between them, and the three columns are not three categories but a container and two of its contents — so the only number that means anything here is twenty-one installs, each of which rides in your context on every turn.

Copy the MCP column if you want somebody’s stack. It is the only one where each box is one thing.