What languages are agent skills written in?

In the first quarter of 2026, 14.3% of newly written agent skills were not in English, with Chinese skills accounting for 6.2% of the total. This is a significant shift from just a quarter earlier, when English accounted for 13.0% of the skills. The fastest-growing developer population for agent skills is India, followed by Nigeria and Singapore, all of which use English.

However, a language count cannot identify these developers, as English is not a proxy for American. The non-English share of agent skills is therefore a lower bound, and should be read as such. Skills are human-written instructions for AI agents, and there is no technical reason to write them in English. Developers in Shenzhen or São Paulo can write procedures in their native language, and the AI will follow them.

The claim that the majority of agent skills are not in English is stronger than it may initially appear, as English is not a proxy for American developers. English is not a proxy for American developers. The data shows that skills are markedly more Chinese, with 6.2% of skills written in Chinese compared to 3.3% in English-speaking countries.

The rise in non-English skills is carried by two groups - European languages and Chinese - rather than all non-English languages. Japanese and Korean have seen no growth in non-English skills. Commit history exists for only 24% of skills, and the subsample leans toward heavily copied ones. Therefore, the rise in non-English skills may be an artifact of a selection effect, rather than a genuine change in what people write.