Why

With recent agentic evolution, integrating LLM CLIs (e.g., Claude Code, Copilot CLI, Gemini CLI) through manually rendered prompts (such as in #55917) creates maintenance overhead and tight coupling to specific tools.

Introducing a structured Agent Skills framework under .github/skills/:

• Provides a generic, tool-agnostic solution
• Reduces maintenance burden
• Improves consistency across locales
• Makes terminology rules explicit and centralized
• Enables scalable automation for future locale additions

This approach clearly separates:

• Global Airflow terminology rules
• Locale-specific translation guidelines

which improves clarity and long-term maintainability.

How

Add Airflow terminology–specific translation skills under .github/skills/translation.

For SKILL.md, we should only mention global-level Airflow terminology guidelines.

e.g., use Dag instead of DAG; some Airflow terms should remain in English, such as Dag and XCom.
Each locales/{locale-name}.md file should document locale-specific guidelines, based on the existing locale files under airflow-core/src/airflow/ui/public/i18n/locales/{locale-name}/.

.github/skills/airflow-translations
├── locales
│   ├── it.md
│   # ... all the locales we currently have, each with locale-specific skills
│   └── zh-TW.md
└── SKILL.md # global-level Airflow terminology guidelines

If your locale does not yet have ‎airflow-core/src/airflow/ui/public/i18n/locales/{locale-name}/README.md:

Please describe why we selected specific terms for the local translation in that file.

Sub-issues

• Global-level agent skills
• Locale-specific agent skills

Committer

[x] I acknowledge that I am a maintainer/committer of the Apache Airflow project.