Sometimes I keep seeing random output from Cursor, attaching the screenshot for the same. The code change in the files seem to be alright but the steps summary and all is just some random characters.
Steps to Reproduce
Just a normal prompt to add a feature or fix an issue in the code.
Expected Behavior
The output is all weird; it seems like something is corrupted.
Thanks for your reply. It happens in the Chat I guess, right after the prompt in the chat it goes into thinking mode and it spits outs this garbage text.
Here is the Request ID - bee94b02-6e2c-4f71-8d95-0d496486409e.
It is a long conversation, it happened after a few interactions.
I have restarted and started fresh, so far I haven’t come across the problem again.
@dnmahendra Be careful posting this on open forums. It would be quite simple for an attacker to inject a prompt into Sonnet (Cursor is attackable using 3rd-party VSCode plugins) that would encode some secrets you have on your machine and have Cursor dump them in the reasoning output. The attacker can then search open forums for specific patterns, find your post, then decode your secret on their side.
In other words, what looks like garbage text might contain some secrets from your machine (think API keys and crypto seed phrases).
@mojojojo Thanks for the reminder. I ensure Cursor has no access to any secrets at any time. That garbage is mostly what Cursor is trying to summarise the task. When I first saw these messages, even I was worried for a second that I got hacked or something seeing that output.
