In Q1 2026, 13.0% of newly written agent skills were in a language other than English, rising to 16.3% one quarter later — a three-point increase in three months across 255,068 skills. For context, GitHub-wide non-English documentation took ten years to grow from 3.7% to 13.0%, indicating AI development is accelerating outside traditional hubs like San Francisco. Notably, English is not a proxy for American developers; GitHub's fastest-growing population is India, which writes in English alongside Nigeria and Singapore.

A skill is a SKILL.md file containing instructions for an AI agent in plain prose, per Anthropic's October 2025 specification. Nine months later, 3.8 million skills existed across 282,200 public repositories in the Git Skills dataset. Skills differ from traditional software: written in human language and executed by multilingual models, they have no technical requirement for English.

Analysis of 1,870,299 distinct skills shows 85.3% English, 6.2% Chinese (led by 104,985 simplified vs 9,112 traditional), 1.7% Japanese, 1.6% German, 1.2% Korean, 1.1% Portuguese, 0.9% Spanish, and 0.4% French. Non-English skills at 14.3% roughly match GitHub documentation's 13.0%, but Chinese representation is markedly higher at 6.2% vs 3.3%. Published figures vary widely (65.0%–99.7% English) due to different populations: curated marketplaces, domain slices, and English-seeded crawls.