Subham-KRLX opened a new pull request, #64012: URL: https://github.com/apache/airflow/pull/64012
Problem: AI coding tools (Claude Code, Gemini CLI, Cursor, etc.) currently treat Apache Airflow as a generic Python project. They are unable to detect whether they are running on the Host or inside the Breeze container, causing them to guess commands that fail 50% of the time and break contributor workflows (e.g., attempting pytest on host when system dependencies are missing). Solution: This PR introduces a Breeze-aware Agent Skills system that exposes Airflow's contributor workflows as machine-readable instructions. Stateless Detection: Implements a detection chain (BREEZE_HOME → /.dockerenv → /opt/airflow) in scripts/ci/prek/breeze_context.py to reliably identify the runtime context. Standardized Skills: Defines core workflows (stage-changes, run-static-checks, run-unit-tests) using the [agentskills.io](https://agentskills.io/) standard, allowing AI agents to choose the correct command (e.g., uv run pytest vs breeze exec pytest) automatically. CI Stability: Resolves a critical SystemExit: 1 test collection error in sub-packages and cleans up workspace-level dependency circularity. Key Components: .agents/skills/: Three skills in agentskills.io standard (YAML + Markdown), ensuring documentation-as-code consistency. validate_skills.py: Pre-commit hook enforcing standard compliance and SPDX license header compatibility. scripts Package: Lightweight distribution for agent utilities, isolated from main Airflow core dependencies. Verification Proof: ✅ 298/298 Tests Passed: Comprehensive unit and end-to-end scenario coverage for context detection. ✅ CI Collection Fixed: airflow-ctl and task-sdk tests now collect with 0 errors (isolated via norecursedirs). ✅ Static Checks Green: All prek hooks (ruff, mypy, newsfragments) passing locally. ✅ Rebased: Synced with the latest apache/main branch. closes: #62500 Was generative AI tooling used to co-author this PR? Yes — Gemini (Code Research) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
