(Note to OpenAI Staff / Community Managers: This is a verified server-side infrastructure bug under active support tickets #13152806 and #13205756. Front-line support is trapped in a triage loop, but the engineering data below indicates an active platform-level Git server regression.)
Bug Description
The `@Sites` editing workflow fails during the repository retrieval step when trying to open a site for editing using a fresh checkout. The remote Git server throws an HTTP 500 error, blocking all site updates.
Request to open a site slug (e.g., `linda-henry-practice`) for editing using a fresh checkout.
The `@Sites` workflow successfully authenticates but fails during the standard clone operation.
Expected Behavior
The repository should clone cleanly into the workspace environment.
Actual Error Result
```text
git clone failed with exit code 128:
Cloning into ‘/workspace/sites/.linda-henry-practice.edit.qtrmkmz5’…
error: RPC failed; HTTP 500 curl 22
The requested URL returned error: 500
fatal: expected ‘packfile’
```
Critical Diagnostic Findings
* **Shallow Clone Works:** Manually forcing a shallow, depth-1 clone of the main branch (`d25dfe5f0bebf9956009803c6c3d96aa8fdd0e7b`) **succeeds perfectly**.
* **Conclusion:** The source files, credentials, and main branch are entirely intact. The failure is strictly a server-side Git infrastructure issue occurring when the backend tries to compile or stream the full repository history/packfile.
Environment & Impact
* **Impact:** Complete blocker. Cannot edit or update the published site.
* **Account Type:** ChatGPT Pro (Paid Tier)
* **OpenAI Support Tickets:** #13152806 and #13205756
Is anyone from the engineering or DevRel team able to flag this backend Git server timeout/packfile generation bug? Front-line support appears stuck in an automated loop. Any help from staff to look at Project ID `appgprj_6a680e2625c88191b7e02349cde8c59c` would be greatly appreciated.
I’m seeing what appears to be the same issue on another fairly large ChatGPT Site.
The published site remains online and is functioning normally, but opening it for editing with a fresh checkout fails during the standard clone operation.
The failure is reproducible and occurs after roughly 80 seconds with:
I contacted OpenAI Support on August 25 and the case has already been escalated.
After seeing OpenAI Support’s earlier recommendation for this issue, I also performed a diagnostic shallow, depth-1 clone of the main branch, without modifying or publishing anything.
That shallow clone succeeded:
Branch: main
Depth: 1
Shallow repository: true
Commits retrieved: 1
It successfully retrieved the current main commit.
I did not modify any code, commit, push, publish, roll back, or make any database/D1 changes. The currently published site was not affected.
So in my case as well, the standard/full clone fails with HTTP 500 / expected 'packfile', while a depth-1 clone of main succeeds.
Adding this as another data point because the behavior appears to closely match the server-side Git/packfile issue described in this thread.
Thank you so much for adding this data point! This is incredibly valuable validation.
The fact that your standard clone is timing out after exactly ~80 seconds strongly points to a server-side gateway or database timeout limit on OpenAI’s Git history packfile generation. It completely matches my findings that the source files themselves are fine (since --depth 1 succeeds), but the backend infrastructure is choking when trying to stream the full repository history.
Hopefully, an OpenAI engineer or Community Manager scans this thread soon to coordinate a fix for both sites.
Unfortunately I still have this problem. I did notice that it was the only project with 250+ MB size. All the other ones are under 250MB so that might be the magic number? Hopefully this gets fixed soon…
I’m still getting this error as well. My asset heavy ChatGPT Site remains live, but downloading its Git repository repeatedly fails. This prevents maintenance and adding a complete data-export endpoint.
Exact error
error: RPC failed; HTTP 500 curl 22 The requested URL returned error: 500
fatal: expected 'packfile'
Just wanted to add my voice here as well. I have a fairly large ChatGPT site that I published, and when I came back the next day to make changes, I’m being completely blocked with the 500 error. I tried doing the “Retrieve my site using a shallow, depth-1 clone of the main branch. Do not modify, overwrite, roll back, or publish anything.” prompt but got the same error:
I created a brand new simple site, and it worked fine, so the bug is limited to the large site I built. I’ve asked Chat GPT to try and download everything it can from the site and create a new one. We’ll se if that works. Pray for mojo lol!
Apologies that you are still facing this issue. Do you have a case open with us already? If so, can you please share the case id so we can review your account specific details. Thanks!
Hi everyone! We rolled out an update to how Sites reads its source repos around September 21. Can you try editing your existing site again? If you still see the 500 or “expected packfile” error, let us know when you tried and your time zone.