Data Agent Airflow skills guide - Data Management - Alibaba Cloud Documentation Center

Skill overview

Data Agent includes the following three Airflow skills, loaded automatically.

Skill	Scenarios	Typical deliverable results
`airflow-dms-authoring`	Author, modify, and validate Airflow 3 DAGs. Supported task types include DMS SQL, lock-free SQL, Notebook, AnalyticDB Spark, Lindorm Spark, DTS data ingestion, and SSH. You can also configure schedules, dependencies, parameters, alerts, and runtime settings.	DAG files, task descriptions, schedule and dependency details, pre-run configuration checklists, and validation results.
`airflow-operation`	Manage DMS Airflow instances and workspaces. Create, query, and update instances. Configure specifications, log directories, DAG and plugin directories, OSS mounts, and upload and publish workspace code.	Instance details, deployment and publish status, Airflow IDs, failure summaries, and next-step suggestions.
`airflow-rest-api`	Manage Airflow runtime configurations and execution status. Supports troubleshooting connections, variables, pools, DAG status, DAG runs, task instances, and DAG failures.	Configuration operation results, DAG run status, failed tasks, error summaries, root cause analysis, and remediation suggestions.

Usage examples

Describe your task in natural language, and Data Agent automatically selects the appropriate skill. For best results, specify the context hierarchically: workspace → Airflow instance → operation.

airflow-dms-authoring

Natural language task description	Data Agent deliverables
"Write a hello world DAG for Airflow 3."	A minimal runnable DAG file, task description, and validation results.
"Write a DMS SQL DAG that runs daily at 2:00 AM to execute this SQL statement."	A scheduled SQL DAG, schedule description, and pre-run configuration checklist.
"Chain these three SQL tasks in a serial DAG: `Task A`, `Task B`, and `Task C`."	A serial dependency DAG and task role descriptions.
"Write a Notebook / Spark / DTS DAG with the following parameters..."	A DAG for the specified task type, parameter descriptions, and runtime configuration suggestions.
"Add failure alerts to this DAG and check whether it is compatible with Airflow 3."	The updated DAG, alert configuration details, and compatibility check results.

airflow-operation

Natural language task description	Data Agent deliverables
"Create an Airflow instance named `<instance name>` in workspace `<workspace name or ID>`, in the `<region>` region with the `<specification>` specification. Here are the network and OSS details..."	Instance creation results, Airflow ID, and deployment status.
"List all Airflow instances in workspace `<workspace name or ID>`."	Instance list, instance status, and actionable instance IDs.
"Upload and publish the local `<DAG directory or file>` to workspace `<workspace name or ID>`."	Uploaded file list, publish status, and whether DAG parsing is pending.

airflow-rest-api

Natural language task description	Data Agent deliverables
"Create a connection named `<connection ID>` in Airflow instance `<Airflow instance ID or name>` of workspace `<workspace name or ID>`."	Connection creation results. Sensitive fields are not displayed.
"Trigger DAG `<DAG ID>` in Airflow instance `<Airflow instance ID or name>` of workspace `<workspace name or ID>`."	DAG run ID, trigger status, and follow-up suggestions.
"Analyze why DAG `<DAG ID>` failed in Airflow instance `<Airflow instance ID or name>` of workspace `<workspace name or ID>`."	Failed runs, failed tasks, key errors, root cause analysis, and remediation suggestions.

Develop Airflow tasks with Data Agent

Airflow task development with Data Agent follows two steps:

Author the Airflow code: Provide the DAG ID, schedule interval, task type, task content, dependencies, runtime configuration, and alert requirements. Data Agent delivers the DAG file, task descriptions, a pre-run configuration checklist, and validation results.
Upload and publish to a workspace: Provide the workspace, Airflow instance, local DAG file or directory, target publish directory, and whether to perform a verification run. Data Agent delivers the upload results, publish status, and verification run results.

Usage examples

Natural language task description	Data Agent deliverables
"Develop an Airflow DAG with DAG ID `<DAG ID>` that runs daily at `<schedule time>`. Run `<Task A>` first, then `<Task B>`."	DAG file, task dependency descriptions, and pre-run configuration checklist.
"Publish the DAG just generated to Airflow instance `<Airflow instance ID or name>` of workspace `<workspace name or ID>`."	Upload results, publish status, and whether DAG parsing is pending.
"Develop `<DAG ID>`, publish it to Airflow instance `<Airflow instance ID or name>` of workspace `<workspace name or ID>`, and run a verification."	DAG file, publish status, and verification run results.

Bind frequently used information with the memory feature

If you regularly work with the same workspace or Airflow instance, ask Data Agent to remember that information. In later conversations, refer to "the default workspace", "the default Airflow instance", or your custom alias, and Data Agent applies the bound details automatically.

Recommended information to bind:

Default workspace name or ID.
Default Airflow instance name or ID.
Preferred region.
Frequently used DAG IDs and DAG publish directories.
Frequently used connection, variable, and pool names.

Usage examples

What you say when binding information	What you say later
"Remember that my default workspace ID is `8685143589400515`."	"List all Airflow instances in my default workspace."
"Bind workspace `<workspace name or ID>` and Airflow instance `<Airflow instance ID or name>` as my default Airflow environment."	"Publish the local `<DAG directory or file>` to my default Airflow environment."
"Remember that DAG `<DAG ID>` is my default debug DAG."	"Trigger my default debug DAG and show me the run results."