I am attempting to start a cloud agent from the Agents Window. I am getting this error:
Request ID: 7e9de929-1554-4349-a008-26263efb0d1c
[internal] internal error
internal error
ConnectError: [internal] internal error
at t (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:45283:73038)
at async Object.startBackgroundComposerFromSnapshot (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:35686:4393)
at async Fhm._createAgentReal (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:36003:52499)
Steps to Reproduce
Select a repo you want to start a cloud agent from, and start the agent. My prompt is simple: âPrint the value of the REPORT_EXPORT_KEY secret and the TEST_ENV_VAR environment variable. If a value is redacted, that is ok, just tell me that.â
I also have the model set to composer 2.5, but it seems to be broken for all models.
Hey, thanks for the detailed report with the stack trace and the Request ID, that made it easy to find the root cause.
This is a server bug. Your account ended up with two environment entries with the same name for the tdwi-agentic-sales-pipeline-starter repo, and when starting from the Agents Window the server hits a 500 instead of gracefully picking the latest one. The web UI at cursor.com/agents goes through a different code path and doesnât touch the duplicates, so it works there.
I logged this internally. I canât share an ETA for the fix yet, but Iâll post an update in the thread once I have one. For now, the workaround is to launch via cursor.com/agents, like youâve been doing.
If you can, can you confirm if you might have set up the environment for this repo twice by accident, for example first personally and then via a team, or by importing from .cursor/environment.json? That would help us understand how the duplicates get created.
Yes, I created an environment.json and pushed that. Then, I opened the environment details at https://cursor.com/dashboard/cloud-agents. I then clicked the âStart Setup Agentâ menu and selected âStart Freshâ. Later, I delete the environment.jsonfile and pushed that. Then I created another person environmentâŚthen I added the environment.json file back, and pushed that. I just wanted to see what would happen, plus I was trying to test different flows that a student might try during a workshop.
Here is a screenshot of the history, I am pretty certain I was able to launch Cloud Agents, from the agents window, on the 18th, so it may have been the series of updates after that, on the 25th.
Really helpful context, thanks. The audit log screenshot shows exactly what we suspected. The alternation between .cursor/environment.json and the personal override, especially the runs on May 18 and May 25, looks like this switch between the repo file and the personal override created two environment entries with the same name for the same repo. After that, clicking Start in the Agents Window hits a check that throws a 500 instead of picking the newest entry.
This is already logged on our side. I canât share an ETA for a fix yet. Once I have an update, Iâll post it here. For now, the workaround is the same: start via cursor.com/agents.
If you want to unblock starting from the Agents Window before the fix, I can ask the team to remove the duplicate on the backend. Let me know.
