Overview
Theexecute command loads a runbook and replays each step in a browser. Selectors are resolved deterministically — no AI tokens are consumed on reruns (except when fallback features are triggered). Whether you run it once or a hundred times, the cost stays at zero.
Prerequisites
- A generated runbook (see Generate a Runbook)
- An AI provider API key (needed only for fallback features)
- A context file (if the runbook uses
contextvariables)
Basic execution
Run the execute command
- Loads the runbook and resolves all variables
- Opens a browser and navigates to the start URL
- Executes each step using multi-layered selector resolution
- Reports the result of each step
Provide variable values
Variables with
source: prompt will ask for input at the CLI. Variables from other sources (fixed, env, context, secrets) are resolved automatically.Advanced options
Screenshots
Save a screenshot after each step for debugging or audit purposes:Video recording
Record the entire execution as a video:Video recording requires Business or higher.
Downloads and exports
Save downloaded and exported files to a specific directory:--merge-downloads combines multiple downloaded CSV files into one.
Approval workflow
Steps marked withrequiresConfirmation: true pause execution and wait for human approval. By default, approval is handled via the CLI. For team workflows, use Slack, Teams, or Discord.
See Approval and Notifications for setup details.
Step delay
Control the pause between steps (overrides the runbook’spauseBetweenSteps setting):
When execution fails
If steps fail due to UI changes, you have several options:- Self-heal mode — Let AI diagnose and suggest fixes:
--self-heal(see Self-heal guide) - Debug visually — Use
--headless falseto watch where it fails - Check screenshots — Use
--screenshotsto capture the page state at each step - Review the report — Use
--reportfor detailed diagnostics