agent skills.
ecosystem report.
The 2026 Agent Skills ecosystem is large, concentrated, and still easy to misread. This live report analyzes the first 500 all-time skills.sh results, removes entries flagged as duplicates, and separates measurable adoption from quality claims. Use it to understand the market, choose safer skills, or publish work that earns trust instead of chasing installs alone.
Analyzed by skillZs. Data refreshes hourly.
What does the 2026 Agent Skills data show?
The API reports 9,590 catalog entries. In the first 500 non-duplicate all-time results, skills have accumulated 74,573,547 installs. The median entry in this high-adoption sample has 83,095 installs, while the top ten account for 9.6% of sampled installs. Those figures describe the ranked sample, not the entire long tail.
Adoption is not evenly distributed across maintainers. The sample contains 75 unique repositories, and multi-skill repositories can accumulate installs across several entries. That makes repository reputation, distribution, and documentation important, but it does not turn popularity into a security review.
Which repositories lead the sampled Agent Skills ecosystem?
- microsoft/azure-skills: 10,785,949 sampled installs across 30 skills.
- open.feishu.cn: 10,706,463 sampled installs across 27 skills.
- mattpocock/skills: 9,010,912 sampled installs across 46 skills.
- larksuite/cli: 8,896,462 sampled installs across 27 skills.
- coreyhaines31/marketingskills: 3,164,157 sampled installs across 51 skills.
- heygen-com/hyperframes: 3,125,246 sampled installs across 31 skills.
- vercel-labs/skills: 2,515,862 sampled installs across 1 skill.
- anthropics/skills: 2,466,249 sampled installs across 18 skills.
- obra/superpowers: 2,303,480 sampled installs across 14 skills.
- leonxlnx/taste-skill: 2,253,597 sampled installs across 13 skills.
Repository totals are calculated only from the current top-500 all-time response after duplicate flags are removed. They are a reproducible snapshot, not a claim about every skill a maintainer has ever published.
How is the Agent Skills report calculated?
- Request the first 500 entries from the official all-time skills.sh API view.
- Remove entries whose upstream record is explicitly flagged as a duplicate.
- Sum installs, count unique source repositories, calculate the median, and measure the top-ten share.
- Group sampled installs by exact source repository and publish the ten largest totals.
- Refresh the server-rendered analysis once per hour while preserving the published method and caveats.
The upstream skills.sh documentation says leaderboard installs come from anonymous CLI telemetry. The CLI reference explains what is collected and how users can opt out. Opt-outs, non-CLI installs, duplicates, changing rankings, and a capped sample all limit what these numbers can prove.
What should Agent Skill creators learn from the report?
Publish one inspectable source, solve a narrow job, test the public installation path, document permissions and compatibility, and maintain the repository after launch. Distribution matters, but durable recommendations come from usefulness and trust. Follow the creation guide, publishing guide, and security checklist before asking others to install.
Readers choosing a workflow should compare the live top-20 rankings, search the full Agent Skills Directory, and inspect the exact source revision. Treat installs as one signal among task fit, maintenance, permissions, tests, and available audits.
What do people ask about Agent Skills statistics?
How many Agent Skills are in the ecosystem?
The live skills.sh API currently reports 9,590 catalog entries. The total can include duplicates and changes as repositories are discovered.
Does an install count measure Agent Skill quality?
No. It measures adoption reported through anonymous skills CLI telemetry. It does not prove correctness, security, compatibility, maintenance, or task fit.
Can I cite this Agent Skills report?
Yes. Cite skillZs, the page URL, the access date, and the stated top-500 all-time sample. Link to the methodology so readers can understand the source and limits.