automate

Automate

Create, manage, and debug automated workflows and schedules.

Routing

User Intent	CLI Command
List workflows	python -m cli workflow list
Create a workflow	python -m cli workflow create or python -m cli workflow create --from-template NAME
Run a workflow	python -m cli workflow execute ID [--dry-run]
Schedule a workflow	python -m cli schedule create
Enable/disable schedule	python -m cli schedule enable ID / python -m cli schedule disable ID
Debug a failing workflow	See Diagnostics section below

Not This Skill

• User wants to run a one-time query → use ask
• User wants to create an agent or tool needed by a workflow → use configure
• User wants to create a data product that a workflow will query → use curate

Key distinction: "Set up a daily email report" is automate. "Set up an agent" is configure. The word "set up" alone is ambiguous — look for scheduling/recurring/automation language to choose automate.

Creating Workflows

Step 0: Verify the query works first

Before building a flow, make sure the underlying query or task works interactively. Use the ask skill to run it once — if the query fails or the agent gives bad results, fix that before automating it. Debugging a broken query inside a scheduled flow is much harder than testing it interactively.

Step 1: Look up the agent

Before building a flow, identify which agent it should use. Both default agents (by ref name) and custom agents (by UUID) work in flows.

Use the configure skill's agent commands to look up agents:

# List all agents — look for the one that matches the user's intent
python -m cli agent list

# Get a specific agent's config, including its input_json_schema
python -m cli agent get <agent_uuid>

The agent's input_json_schema tells you exactly what inputs the flow node needs. This matters because different agents require different inputs — a custom agent might need project_name and report_type, not just message and data_product_id.

If the user has been working with a specific custom agent (e.g., they just created or configured one), use that agent's UUID — don't substitute a default agent.

Step 2: Build the flow from a template or from scratch

Three templates are available — list with python -m cli workflow templates:

1. agent-only — Run a single agent. The simplest flow — good for running any agent on a schedule. Required: name, agent_id, query

2. query-and-email — Query a data product, email the results. Required: name, agent_id, data_product_id, query, tool_id (SMTP), email, subject

3. query-only — Query a data product (no email). Required: name, agent_id, data_product_id, query

Example: python -m cli workflow create --from-template agent-only --agent-id <uuid> --query "Generate the weekly summary"

If the agent needs additional inputs beyond message (check input_json_schema), build the flow JSON from scratch instead. See references/agent-schemas.md for how to wire agent inputs into flow nodes — it covers both default and custom agents.

Scheduling

python -m cli schedule create \
  --workflow-id <uuid> \
  --name "Weekly sales report" \
  --cron "0 9 * * 1" \
  --timezone "America/Los_Angeles" \
  --timeout 60

• Cron format: minute hour day month weekday (e.g., 0 9 * * 1 = Monday 9am)
• Minimum interval: 60 minutes
• --timezone defaults to America/Los_Angeles
• --timeout defaults to 60 minutes
• --disabled creates the schedule in disabled state (useful for testing)
• Auth is handled automatically — the CLI finds the user's managed auth config
• To manage existing schedules: python -m cli schedule list/get/enable/disable/delete

Diagnostics

When a workflow or schedule isn't working, diagnose using the CLI:

1. Schedule not firing?

   • python -m cli schedule get <id> — check enabled and next_run_at
   • If disabled, enable with python -m cli schedule enable <id>
   • If next_run_at is in the past, the scheduler may be stalled — escalate to the user's admin

2. Execution failing?

   • python -m cli workflow list — check recent execution status
   • Auth errors → use the setup skill to re-authenticate
   • Tool/agent not found → verify IDs with python -m cli tool list / python -m cli agent list

3. Issues beyond CLI access (stuck executions, duplicate runs, scheduler pod health) require admin/SRE intervention. Describe the symptoms to the user and suggest they contact their Alation administrator.

Common Mistakes

Mistake: Creating a new SMTP tool when the user asks to email results. Why it seems reasonable: the flow needs an email tool. Instead: List existing tools first (python -m cli tool list). There is likely already an SMTP tool configured. Only create a new one if none exists.

Mistake: Using a default agent when the user has a custom agent for the task. Why it seems reasonable: default agents are easier to reference. Instead: If the user created or configured a custom agent, use its UUID. Check python -m cli agent list to find the right one. Look up its input_json_schema to know what inputs the flow node needs.

Mistake: Hardcoding agent/tool IDs in the flow definition without verifying they exist. Why it seems reasonable: the user provided the names. Instead: Look up IDs using python -m cli agent list or python -m cli tool list.

Mistake: Building a flow without checking what inputs the agent expects. Why it seems reasonable: most agents just need message. Instead: Run python -m cli agent get <uuid> and check input_json_schema. Custom agents often need additional inputs like data_product_id or domain-specific parameters.

Sharing URLs

After creating a workflow, share the url from the CLI response so the user can view it in the Alation UI. Schedule responses include a workflow_url pointing to the parent workflow.

Red Flags

• "I'll create an SMTP tool for this workflow" — check existing tools first.
• "Let me build the flow JSON from scratch" — try templates first.