▶
ExecutionTrigger
Automation/LLM/Healing
Uses LLM to diagnose automation failures and suggest/apply healing actions
Scores range from 0 to 10. Higher values mean more impact, exposure, or operational weight.
Trigger
LLM model (vision-capable preferred)
Optional screenshot at the time of failure as base64 PNG, JPEG, WebP or GIF (a data URL is fine). Ignored when Image is connected
Optional screenshot at the time of failure as an image, e.g. from the Screenshot node. Takes precedence over the base64 Screenshot
Screen frame of the screenshot, from the capture node. Coordinates are desktop input coordinates (ready for the mouse nodes) when Frame is connected, otherwise pixels of the original screenshot
The error message from the failed action
Type of action that failed (click, type, wait, find, etc.)
The target of the failed action (selector, text, or an x,y point in desktop input coordinates (ready for the mouse nodes) when Frame is connected, otherwise pixels of the original screenshot)
Additional context about what the automation was trying to do
Current page HTML (for selector-based failures); only the first 30,000 bytes are sent
Continue (Healed)
Could not heal
Full healing result
Failure diagnosis
Healed value (new selector, text, or an x,y point in desktop input coordinates (ready for the mouse nodes) when Frame is connected, otherwise pixels of the original screenshot)