Skip to main content
Question

AI Skill returning previous/stale response after changing Prompt Input

  • August 28, 2026
  • 11 replies
  • 162 views

HemanthaPindra
Navigator | Tier 3
Forum|alt.badge.img

Hi Team,

I am facing an issue while testing an AI Skill in Automation Anywhere AI Studio using the "AIAE - Skills" model connection.

I have configured a String Prompt Input variable and referenced the same variable correctly in the prompt.

The first test returns the expected response. However, after changing the Prompt Input value to completely different data and clicking Get response again, the AI Skill sometimes continues to return the previous response instead of processing the updated input.

I have already tried:

  • Changing the Prompt Input value to completely different content

  • Verifying that the Prompt Input variable is correctly referenced in the prompt

  • Closing and reopening the AI Skill

  • Testing with simple input values

  • Using a simple echo prompt to verify the input variable

The echo test correctly returns the updated input value, but when using the actual prompt, the skill can still return the previous response.

Has anyone experienced similar behavior with AI Skills using the "AIAE - Skills" model connection?

Could this be related to response caching or the model connection?

Any suggestions would be appreciated.

Thank you.

11 replies

  • Ground Control | Tier 1
  • August 29, 2026

Hi Team,

I am facing an issue while testing an AI Skill in Automation Anywhere AI Studio using the "AIAE - Skills" model connection.

I have configured a String Prompt Input variable and referenced the same variable correctly in the prompt.

The first test returns the expected response. However, after changing the Prompt Input value to completely different data and clicking Get response again, the AI Skill sometimes continues to return the previous response instead of processing the updated input.

I have already tried:

  • Changing the Prompt Input value to completely different content

  • Verifying that the Prompt Input variable is correctly referenced in the prompt

  • Closing and reopening the AI Skill

  • Testing with simple input values

  • Using a simple echo prompt to verify the input variable

The echo test correctly returns the updated input value, but when using the actual prompt, the skill can still return the previous response.

Has anyone experienced similar behavior with AI Skills using the "AIAE - Skills" model connection?

Could this be related to response caching or the model connection?

Any suggestions would be appreciated.

Thank you.

I have seen similar kind of thing earlier. Try ‘Evaluation’ once.


  • Ground Control | Tier 1
  • September 1, 2026

Hi there, I am getting same issue, and i even tried evaluations.. the output is coming as per requirement right there, what to proceed with?? ​@Oli.Morris ​@Aaron.Gleason your suggestions are much appreciated in this scenario.


Aaron.Gleason
Automation Anywhere Team
Forum|alt.badge.img+6
  • Automation Anywhere Team
  • September 1, 2026

Hi there, I am getting same issue, and i even tried evaluations.. the output is coming as per requirement right there, what to proceed with ​@Oli.Morris ​@Aaron.Gleason you suggestions are much appreciated in this scenario.

Try a different LLM and see if that works better. You have at least two: AI Agent and AI Skill. Try the AI Agent one in your AI Skill and see if that improves things.


  • Ground Control | Tier 1
  • September 1, 2026

Hi there, I am getting same issue, and i even tried evaluations.. the output is coming as per requirement right there, what to proceed with ​@Oli.Morris ​@Aaron.Gleason you suggestions are much appreciated in this scenario.

Try a different LLM and see if that works better. You have at least two: AI Agent and AI Skill. Try the AI Agent one in your AI Skill and see if that improves things.

Hi ​@Aaron.Gleason,

I have used both and in both im facing same issue.


Aaron.Gleason
Automation Anywhere Team
Forum|alt.badge.img+6
  • Automation Anywhere Team
  • September 1, 2026

@Vipulkumar Not sure then. Open your browser’s developer tools and turn off caching. Verify that updated data is being sent and received (Network tab). 


  • Ground Control | Tier 1
  • September 1, 2026

@Vipulkumar Not sure then. Open your browser’s developer tools and turn off caching. Verify that updated data is being sent and received (Network tab). 

Tried disabling the cache, and running again still issue persists. :(

 


Aaron.Gleason
Automation Anywhere Team
Forum|alt.badge.img+6
  • Automation Anywhere Team
  • September 1, 2026

@Vipulkumar Sorry, I have no idea. Maybe message caching at your company?

Here’s me trying an AI Skill. First prompt and response:


Modifying the system prompt to have the LLM return XML:

It’s got to be something on your side.


  • Ground Control | Tier 1
  • September 1, 2026

@Vipulkumar Sorry, I have no idea. Maybe message caching at your company?

Here’s me trying an AI Skill. First prompt and response:


Modifying the system prompt to have the LLM return XML:

It’s got to be something on your side.

Hi Aaron,
I have already tried running ai skills before and it was running fine till yesterday, im facing the issue today itself,

I deleted the old ai skill and created a new one: 

first prompt: the output is generated correctly as per the input provided: 
second prompt: the output is same as the first one

Also i would like to highlight that i am referring to input variables, where when im changing the data, it is not getting updated in variables passed in user prompt.
 



 

 


Aaron.Gleason
Automation Anywhere Team
Forum|alt.badge.img+6
  • Automation Anywhere Team
  • September 1, 2026

@Vipulkumar Good troubleshooting. Yep, I’m seeing the same thing. I have no solution. I will ask around.


  • Ground Control | Tier 1
  • September 2, 2026

@Vipulkumar Sorry, I have no idea. Maybe message caching at your company?

Here’s me trying an AI Skill. First prompt and response:


Modifying the system prompt to have the LLM return XML:

It’s got to be something on your side.

Hi Aaron,
I have already tried running ai skills before and it was running fine till yesterday, im facing the issue today itself,

I deleted the old ai skill and created a new one: 

first prompt: the output is generated correctly as per the input provided: 
second prompt: the output is same as the first one

Also i would like to highlight that i am referring to input variables, where when im changing the data, it is not getting updated in variables passed in user prompt.
 



 

 

While testing, the AI Skill appears to be returning stale responses. For example, if you delete the input variable and create a new input variable with the same name, it provides a different output the first time. However, after that, it starts returning the same stale response again.

Instead of relying on Test, you can try using Evaluate to test the AI Skill and verify its responses more reliably.


  • Ground Control | Tier 1
  • September 11, 2026

@Vipulkumar Sorry, I have no idea. Maybe message caching at your company?

Here’s me trying an AI Skill. First prompt and response:


Modifying the system prompt to have the LLM return XML:

It’s got to be something on your side.

Hi Aaron,
I have already tried running ai skills before and it was running fine till yesterday, im facing the issue today itself,

I deleted the old ai skill and created a new one: 

first prompt: the output is generated correctly as per the input provided: 
second prompt: the output is same as the first one

Also i would like to highlight that i am referring to input variables, where when im changing the data, it is not getting updated in variables passed in user prompt.
 



 

 

While testing, the AI Skill appears to be returning stale responses. For example, if you delete the input variable and create a new input variable with the same name, it provides a different output the first time. However, after that, it starts returning the same stale response again.

Instead of relying on Test, you can try using Evaluate to test the AI Skill and verify its responses more reliably.

Hi Rut,

Please go through the chat chain and understand the fact that i have already mentioned that using the evaluations the output is coming fine, but i can’t just rely on evaluations, because i might use ai skill in taskbot/ api skill and in mozart, also i have to perform a live demo of the flow.

However, the issue is now resolved and seems like temporary issue from backend. Thankyou.