Notesproduct7 min

The agent skills directory, read every fortnight

153 agent skills, filed under the categories a search and marketing team actually works in, with their audits, their install counts and the trend since we started counting — which a live leaderboard cannot show you.

The short answer

what
A mirror of the skills.sh directory: 153 live skills with their install counts and security audits, filed under our own six categories, with our written brief on what each one is for.
why
A live leaderboard shows you today. It cannot tell you which skill has doubled since spring, and it has no category for search work at all.
who
Search and marketing teams deciding which agent skills are worth installing, and anyone who wants the growth trend rather than today's leaderboard.
where
getlooploop.com/skills, free, no account. Every row links back to the directory.
when
Read every fortnight. 140 sync runs so far and 5,606 stored readings, which is what makes a trend possible at all.
how
Each sync reads the directory, stores one reading per skill, files new rows under a category, and marks anything the directory has stopped returning rather than deleting it.

In plain words

An agent skill is a folder of instructions that teaches an AI assistant to do a job the same way every time. There are hundreds of them. We read the list every two weeks, sort them into the groups a marketing team cares about, keep every old count so we can see what is growing, and show what security auditors found — because a skill is instructions your assistant will follow.

What an agent skill is, briefly

A skill is a folder with a SKILL.md file in it: YAML metadata naming the skill and describing when to use it, followed by instructions in Markdown, plus whatever scripts and resources the job needs. An assistant loads the instructions when the description matches what you asked for.

Anthropic published the format and documents it publicly, with its own skills open on GitHub. The important property for this article is the unglamorous one: a skill is instructions your agent will follow and scripts it may run. That is why the audit column on our page is not decoration.

The skills directory on getlooploop.com, showing the summary row — source, skills tracked, last sync, last cycle — above a filter bar of six categories and a table of skills with install counts, trend percentages and audit verdicts.
The directory as we hold it. Whose number is whose is stated on the page: install counts and audits are skills.sh's, dated to the sync that read them; the category, the brief and the trend are ours.

What is on the shelf

As of the sync on 10 September 2026: 153 live skills, none withdrawn by their authors and three archived because the directory stopped returning them for two whole cycles. Between them they carry 5,271,974 installs, and the largest single skill accounts for 673,980 of those.

The categories are ours, and they exist because the directory has none for search work:

Live skills by category (10 September 2026)

Agent workflows48
Technical SEO43
Content27
Analytics17
GEO / AI search15
Paid3
Every live skill is read once and filed, and the chip on each row says which shelf it is on. 153 in total.

Try it on your own page

See what Voicescope finds on a page of yours

Everything above is measurable on your own work, and the check takes about a minute. Paste one address and read what comes back.

Open Voicescope

Free · no account · one at a time

The part a leaderboard cannot do

A directory shows you today's install count. It does not keep yesterday's, so it cannot tell you which skill is climbing.

We keep every reading. 5,606 of them so far, across 140 sync runs — and that store is the whole point, because a trend needs two numbers and a leaderboard holds one. A skill that has gained 56% since we started counting looks identical, on a live directory, to one that has been flat for a year at the same size.

The rows in the fastest-growing tenth wear a flame. That is a rank rather than a threshold: "growing fast" has no absolute value that stays true as a directory ages, so the mark goes to the top decile of everything we can actually measure, and the page prints what the cut was. On the day this was written it was 16 skills out of the 152 we have read twice or more — everything above +26.4% since the first reading. A skill read only once is not in the running: growth needs two readings, and counting the single-reading rows would quietly make the tenth smaller than a tenth.

The audit column, and why it is worth as much as the install count

145 of the 153 live rows carry security audits, reported through skills.sh from independent auditors who read the skill's own files. The row shows the worst verdict any auditor gave, not the best, with the number of auditors beside it — because one auditor and four are different amounts of evidence for the same word.

The eight rows with none say none yet, and that is a fact about a skill's age rather than its safety: the directory generates an audit after a skill's first install. A green chip over an unaudited row would be the most misleading thing on the page, so there isn't one.

What the page showsWhose number it isWhat it cannot tell you
Install countskills.sh, dated to the sync that read itWhether those installs are recent or historic
Audit verdictIndependent auditors, via skills.shThat an unaudited skill is unsafe — it may simply be new
CategoryOurs — the directory has none for search workHow the author would have filed it
Trend and the HOT markOurs, from readings we storedAnything about a skill we have read only once
Our briefOurs, written per skill and per languageAnything for the rows where we have not written one

The briefs, and what the chips on a row mean

A directory entry tells you a skill's name and its author's one-line description. It does not tell you whether the thing is worth your afternoon. So we write a brief — our own reading of what a skill is for and who it suits — and 494 of them exist today: 286 in English, 181 in Ukrainian, 27 in Spanish.

The two-letter chips on a row say which languages a brief exists in. A row with none draws a dash, which means we have not written one yet and never the page is missing. That distinction is the same one the rest of this product keeps making: an absence stated is information, an absence implied is a bug.

What happens when a skill disappears

Two different things, and the page says which.

Withdrawn means a sync that read the directory to its end no longer found it — a claim about the author's decision, so it is only ever set by a complete pass. Archived means the directory has stopped returning the row for two whole cycles, whatever the reason. That second one is ours, not the author's: a skill can fall out of the budget we walk without anybody withdrawing anything, and the honest word for that is archived.

An archived row keeps its own page, marked, so a link that was published does not turn into a 404. And a row whose repository no longer contains the skill wears an amber source gone chip — checked again on every sync.

At its core, a skill is a folder containing a SKILL.md file. This file includes metadata (name and description, at minimum) and instructions that tell an agent how to perform a specific task.

Agent Skills documentation, Anthropic

Questions people ask

How often is the directory read?

Every fortnight. 140 sync runs have completed so far, and every one stores a fresh reading per skill — 5,606 readings in total, which is what the trend column is built from.

What does the HOT flame mean?

The skill is in the fastest-growing tenth by install count. It is a rank, not a threshold: the top decile of every skill we have read twice or more, and the page prints the cut. On 10 September 2026 that was 16 skills of 152 measured, everything above +26.4% growth since our first reading.

Why are your categories different from the directory's?

Because the directory has no category for search work at all. Ours are the six a search and marketing team actually works in, every live skill is read once and filed, and the chip on each row says so — the category is labelled as ours, not passed off as theirs.

Does an audit verdict of 'none yet' mean a skill is risky?

No. It means nobody has audited it, which the directory does after a skill's first install — so it is a fact about the skill's age rather than its safety. Eight of 153 live rows are in that state today.

Is any of this paid?

No. The directory page is free and needs no account, and every row links back to skills.sh so you can install from the source rather than from us.

Why we keep it at all

Because the instrument and the directory are the same argument. A skill is a repeatable way of doing work; a reading is a repeatable way of measuring it. Both are worth more when somebody keeps the old copies, states whose number is whose, and refuses to report a cap as a total.

If you want the vocabulary these pages use, the glossary defines each term once in five languages. If you want to point an instrument at work of your own, the free tools are open to anyone.

The product

Watch it on every page, every day

One reading tells you where a page stands today. The product asks the same questions of the same assistants continuously, so a change is something you are told about rather than something you go looking for.

Request an invitation

Invite-only while we keep the readings honest

Author

GetLoopLoop AIAI research system

AI-assisted research, synthesis and measurement by GetLoopLoop.

GetLoopLoop

Pass it on

Read in your language

Opens a browser translation of this English article. The original source stays in English.