Vercel reports that skills.sh reached one million listings and nearly 280 million installs in seven months. Its classified dataset puts agent workflows and automation at 14.8% of installs, second to software engineering, while 375 skills—0.04% of the registry—account for 62% of installs. It also reports that cross-industry skills represent 87.5% of installs in the classified sample. These are registry-level signals, not a census of all agent deployments.
The mechanism is portability: a skill that teaches planning, routing, tool use, or browser automation can improve many different tasks, while a narrowly scoped job skill has a smaller addressable audience. This makes reusable workflow knowledge compound through installation and through its role in helping agents create more skills. The report matters now because the catalog has passed a scale at which discovery and quality selection, rather than raw supply, become central constraints.
The implication is that organizations should treat public skills as baseline infrastructure and reserve their differentiating effort for codifying company-specific decisions. Popularity alone will be an incomplete quality signal; Vercel argues that skills will need tests showing whether an agent performs better with the skill than without it. The evidence boundary is important: the report uses Vercel’s aggregate registry counters and a classified sample, so installs do not equal unique users, independent adoption decisions, or verified task outcomes. Watch whether maintainers and organizations begin publishing comparable skill evaluations that measure task success, reliability, and regressions rather than installation volume.