Skillselion Research

Research on the agent ecosystem

We index the whole AI agent skill ecosystem, which means we can also measure it. These reports count what is there, listing by listing and install by install, and publish the raw data so you can check the arithmetic yourself.

Frozen editions. We compute each census from a complete catalog dump on a single date, and we never revise the published numbers. A citation from 2026 still checks out in 2030.
Method in the open. Every report states its source, its date and its caveats, including the places where a number is softer than it looks.
Raw data attached. Editions ship the full dataset as a download, free to reuse with attribution.

Reports

EditionAugust 202679,827 rows · CSV

The Agent Economy Census

A complete count of the AI agent skill ecosystem: 79,827 listings, 157,833,061 installs, frozen to a single date, with the raw data attached.

Install inequality reaches a Gini of 0.96, against 0.63 for the most unequal country the World Bank tracks
A solo educator rivals Microsoft on installs; OpenAI trails 70x behind
Only 55 of 79,827 listings (0.07%) carry a price
ByteDance's Lark is the biggest single publisher, at 15.4% of all installs
2.4M installs go to skills built to interrogate the person using them

Read the report

EvergreenContinuously updated

The State of AI Agent Skills 2026

The live snapshot: how many agent skills, MCP servers and marketplaces exist right now, what categories dominate, and which tools lead on installs. It refreshes as the catalog moves.

Ecosystem totals by tool type
Adoption by build stage and category
Current most-installed leaderboards

Read the report

For press and researchers

Every report is free to cite and every edition ships its dataset. If you need a cut we have not published, say a single category, one publisher or a full distribution, email info@skillselion.com with the question. We will run it and send the method alongside the answer. Attribution format: “Skillselion Agent Economy Census, August 2026” with a link to the report.

FAQ

Can I cite Skillselion research or reuse the data?

Yes. All Skillselion research is free to reuse with attribution (CC BY 4.0): name the report and edition, for example "Skillselion Agent Economy Census, August 2026", and link the report page. Raw datasets are downloadable from each edition.

What is the difference between a census edition and the evergreen report?

A census edition is frozen: it is computed from a complete catalog dump on one date and its published numbers never change, so a citation stays verifiable forever. The evergreen State of AI Agent Skills report tracks the live catalog and changes as the ecosystem moves.

Where does the underlying data come from?

From the Skillselion production catalog, which aggregates public registry data, chiefly the skills.sh registry, alongside GitHub and other public sources. Each report states its own method, date and caveats.

Are journalists and researchers able to request specific cuts of the data?

Yes. Email info@skillselion.com with the question and we will run the cut against the census dataset and share it, along with the method used.