---

name: authoring-task-sdk description: "Write, migrate, debug, and review Airflow 3 Dag authoring code using airflow.sdk and Task SDK public interfaces." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Authoring Task SDK

Use this sub-skill when the task is about author-facing Airflow 3 Dag code: creating Dags, migrating Airflow 2 imports, reviewing TaskFlow or standard provider operators, diagnosing Dag discovery/import issues, and validating that a Dag file is parseable and discoverable.

Read First

• references/api-reference.md for verified imports, signatures, public-interface boundaries, and standard provider operator signatures.
• references/workflows.md for task-oriented recipes: new Dag, import migration, dependencies, task groups, dynamic task mapping, assets, timetables, Params, XCom/context, TaskFlow, and standard operators/sensors.
• references/troubleshooting.md for parse/import failures, discovery heuristics, schedule/Param/template mistakes, metadata DB access restrictions, async/deferrable confusion, and optional Graphviz issues.
• scripts/validate_dag_file.py for a bundled safe parse/import helper that validates one Dag file without depending on the original repository checkout.

Scope

This sub-skill covers author-facing Python APIs used in Dag files:

• Dag declaration with airflow.sdk.DAG or @dag, including schedule, start_date, catchup, params, tags, task_group, fail_fast, allowed_run_types, deadline, disable_bundle_versioning, and rerun_with_latest_version.
• TaskFlow decorators (@task, @task.branch, provider decorators exposed under @task.<provider>), @task_group, dependency helpers, dynamic task mapping, and XComArg wiring.
• Assets and asset scheduling with Asset, AssetAlias, inlet/outlet events, asset expressions, and mixed asset/time schedules.
• Timetable selection, event schedules, Params, Jinja templates, context access, XComs, Variables, Connections, and standard provider operators/sensors commonly embedded in Dag files.

Boundaries

• For CLI commands, REST API operations, Dag runs, state management, and operational debugging, route to the operations/CLI/API sub-skill.
• For provider package development, provider tests, hooks/operators implementation, or packaging conventions, route to the provider/extension sub-skill.
• For Helm, Docker images, executor deployment, triggerer/scheduler operations, or production configuration, route to the deployment sub-skill.
• For contributor tooling, Breeze, static checks, PR workflow, or repository development tasks, route to the contribution tooling sub-skill.

Validation

Prefer validating changed Dag files with the bundled helper before recommending broader operational commands:

python skills/apache-airflow/sub-skills/authoring-task-sdk/scripts/validate_dag_file.py path/to/dag_file.py --list-dags
python skills/apache-airflow/sub-skills/authoring-task-sdk/scripts/validate_dag_file.py path/to/dag_file.py --expect-dag-id example_dag

Use --repo-root only when the user explicitly wants the helper to import packages from a current checkout. Do not bake local checkout paths into Dag code or skill documentation.