Where does the bug appear (feature/product)?
Cursor IDE
Describe the Bug
Request ID: 1b5a76a7-4776-4b40-a2b2-746c0d298ae3
[internal] Failed to run step, exceeded max retries
Rpe: [internal] Failed to run step, exceeded max retries
at iw1 (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:35723:27646)
at fog (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:35723:26375)
at hw1 (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:35724:7469)
at kog.run (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:35724:12370)
at async qle.runAgentLoop (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:48124:2927)
at async k7g.streamFromAgentBackend (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:48195:12809)
at async k7g.getAgentStreamResponse (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:48195:21069)
at async sut.submitChatMaybeAbortCurrent (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:35854:1995)
at async no (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:47078:33305)
Screenshots / Screen Recordings
Operating System
MacOS
Note steps tried:
restart curosr
uninstall and reinstall
change Http settings
turn on/off VPN
try different models
Steps to Reproduce
Steps to Reproduce
Open Cursor IDE on MacOS system, open existing local project folder. ask it any sort of request, small or large.
Select part of project source code, send AI chat request to ask code analysis & modification via Cursor built-in AI dialogue.
After waiting for several seconds, popup error alert: An unexpected error occurred, Request ID:1b5a76a7-4776-4b40-a2b2-746c0d298ae3,
Expected Behavior
After sending code-related query in Cursor chat panel, the backend agent connects normally without ping timeout, AI can receive request and return corresponding code analysis/answer content stably, no unexpected error popup.
Thank you very much for the Request ID that is extremely helpful in helping us pinpoint the source of the bug. This conversation has grown larger than the model’s context window, and Cursor’s automatic summarization can’t shrink it enough to recover, so the request retries and fails with that error.
The best fix is to start a new chat. That resets the context, and you should be back to normal right away. For long tasks, a fresh chat per task helps avoid this.
We are tracking this behavior and are working to improve it, so I’ll let you know any updates there. But for now, the easiest workaround and fastest way to get unblocked is to start a fresh chat regularly. Let me know if you have any follow-ups or you are seeing this error a lot more than you expect.
One more thing just to sanity check. What does your prompt context usage look like? Click on the circle indicator and let me know what it looks like when you kick off a fresh chat with a simple message. E.g. Send the message “Hi” to the agent and let me know what the context usage looks like. Just want to make sure it’s reasonable and not tremendously huge. You could have a bunch of skill files or things that are cluttering up the context on each turn, unexpectedly.
Yes, that is extremely large, and that is definitely what is causing the issue.
The request reaches us fine; the problem is your context window is nearly full before you type anything. You have ~150k tokens of Skills installed, and every Skill’s name + description loads on every request. That eats over half the model’s window up front, so even a brand-new chat with a one-word message can instantly hit the limit and error. A fresh chat inherits the same oversized baseline, which is why it kept happening.
The fix is to trim your Skills:
Open Cursor Settings (Cmd+Shift+J) → Rules, Skills, Subagents. Look for your skills listed here and remove those that you don’t need.
For ones that you do need, add this line to the top header section: disable-model-invocation: true to rarely-used Skills so they only load when you type /skill-name. This means that you must invoke the skill manually, and therefore, it doesn’t clutter up your chat context on each chat. Personally, I have dozens of skills, and the vast majority are non-model-invocation skills, so I don’t clutter my context.
If you don’t see a lot of skills listed in Cursor settings, they are likely getting picked up from another AI coding product on your machine. You can toggle this switch to disable them.
