My agent keeps saying it is the start of the conversation. I keep prompting but it just thinks a bit, then it start summarizing the chat even though we are only on 25% context usage, and then it says the conversation has just started even though we are 3 prompts in. (see image) I got it to work once in this conversation but even when starting a new agent it happens and just ignores me.
Steps to Reproduce
Request id: c0c6b204-0a29-42eb-9099-6f84a206cc35
Expected Behavior
the agent should interpret my question and do its thing instead of ignoring everything that happened
Hey, this is a known bug with chat summarization. After summarization triggers, the agent can lose track of the conversation and start treating old messages like new ones.
Your build (v2.6.14, March 8) is from right before the fix shipped. Update to the latest version via Help > Check for Updates, and that should fix it.
For now, if it happens again, the quickest workaround is to start a new chat. You can use @ mentions to pull in the relevant files.
Thank you for responding. The update did not fix anyting, but I messed around and eventually it stopped by itself when I tried different work. Now I noticed that it only seems to happen when I include the cloudinary url in the prompt. That url points to an image.
It only happens when I copy paste the url from the browser directly into the chat. When typing the url manually it does not happen.
I tested and it seems that any url pointing to any internet image will make the agent very confused, causing immediate chat summarizing, and then trying random stuff. Mixing the link with other things in the prompt does not solve it.
Example of it: 0cea9664-9db3-43b5-834b-40a8a0c4dab4
