Custom Agent - Instructions | The place for Zendesk users to come together and share
Skip to main content
Question

Custom Agent - Instructions

  • June 11, 2026
  • 1 reply
  • 9 views
D

I am jumping into custom agents for the first time and struggling to get the output I’m looking for. Ideally the agent will read a new ticket, identify any image or pdf attachments (inline or attached) and put the content of those images in as an internal note. 

Right now, I can get it to paste a lot of the metadata or what it “assumes” the content is but not the actual values. For example if the screenshot has a list of 5 emails + IDs I just want the plaintext version in an internal note. 

My current instructions read like 

  1. When a ticket is created, look for any images (e.g., .png, .jpeg, .jpg) or PDFs. These files could be attachments but might also be in-line and there could be multiple images.
  2. If an image or PDF is found, paste all of the exact text output captured. Do not alter, summarise, or paraphrase the extracted text. Repeat this for all images or PDFs found
  3.  Add the tag image_read to the ticket

There was an interesting demo of this concept at the showcase yesterday, not sure if I’m missing something or I just need some tweaking on the instructions.

    1 reply

    O

    Whenever a new ticket is created, inspect all attached and inline images and PDFs. Extract all visible text using OCR. Return the extracted text exactly as it appears, preserving formatting, line breaks, email addresses, IDs, and other values. Do not summarize, infer, describe, or interpret the content. If text cannot be read with confidence, indicate the unreadable portion rather than guessing. Create an internal note containing the full extracted text from each file and add the tag 'image_read’ 
    human to dog translator

    Terms & ConditionsAccessibility statement