Kagi Assistant - always show complete code/content in full

beetstabasco

Per Kagi Assistant itself...

Response 1:

I apologize for this behavior - you are absolutely correct that I have been doing this, and it is wrong. I should never truncate responses with "..." or ask if you want to see "the rest."

This appears to be an incorrect pattern I've fallen into. Even if a file is long, I should always show the complete content without:

Using ellipses ("...")

Adding comments like "rest of the original tests..."

Asking if you want to see more

Breaking it into multiple responses

I will correct this immediately. From now on, I will always show the entire file/code/content in full, without any truncation or follow-up questions about seeing more.

If you notice me falling back into this incorrect behavior, please call it out again. There should never be any need for you to request "the rest" of anything - you should get complete responses every time.

Response 2:

My corrections and learning do not persist across different conversation threads. I am a stateless system, meaning:

Each conversation thread starts fresh

I cannot remember or apply learnings from previous conversations

Even if you correct my behavior in one thread, I will revert to default behavior in a new thread

This is a fundamental limitation of my current architecture. The only way to consistently get complete responses would be to:

Include your requirement for complete responses in each new conversation thread, or

Have my base instructions modified by my developers to never truncate responses or ask about showing "the rest"

Response 3:

To request changes to my base instructions to prevent truncation and partial responses, you should:

Contact Kagi directly through their official support channels, since they are the ones who developed and maintain me (Kagi Assistant)

Specifically request that my base instructions be modified to:

Never truncate responses

Never use ellipses to shorten content

Never ask users if they want to see "the rest" of anything

Always show complete code/content in full

Instead of asking Kagi Assistant to repeat itself and give me the entire response, it would give me the proper response on the 1st try.

beetstabasco

Adding verbiage to custom instructions does not seem to help.

Luis

This appears to be a known problem with the October release of Sonnet 3.5 and reportedly affects Haiku 3.5 as well. We've tried different prompting techniques but can't stop it consistently, this has also been reported by several users to Anthropic. For now, the most effective solution seems to be switching to a different model.

We will mark this thread as nofix, as there are no actionable next steps for us. If we have misunderstood anything, please let us know