ChatGPT fails on Python and ZIP and Throws ClientError

This is still an issue. I’ve hit a persistent ChatGPT code-execution/file-generation problem while developing a Python/OpenCV/YOLO application in a ChatGPT Project.

The project has been developed iteratively through several working Python versions. ChatGPT had previously been able to read the existing source, modify it, generate a new .py file, syntax/compile-check it, and provide the resulting file for download. We successfully used that workflow through development of the current V6.2 detector.

When we attempted to generate V6.3, the execution environment started failing. ChatGPT could still read and analyse the source code, logs, CSV output, screenshots and other project files, and could specify the required code changes, but attempts to actually execute Python or generate/validate the new Python file failed with execution/runtime errors. Some attempts reported the code execution state resetting or a client error; others simply remained stuck on “Working” for a very long time.

We tried to isolate whether the problem was caused by the existing conversation. A new chat inside the same Project also failed, suggesting it wasn’t just a corrupted conversation. We then tried a normal chat outside the Project, where execution initially worked: it successfully generated V6.2 and performed Python compile/AST/static checks. That demonstrated the source code itself and the requested workflow were valid.

Unfortunately, subsequent attempts to generate V6.3 outside the Project also became unreliable. We tried fresh chats, reduced the amount of evidence supplied, supplied the authoritative V6.2 source plus logs/CSV separately, and even removed video analysis as a prerequisite so the task was simply “modify this known-good Python file according to this specification, create V6.3 and compile-check it.” Those attempts either failed execution, had file/video-transfer problems, or sat indefinitely on “Working.”

Importantly, the Python application itself is not what is failing here. V6.2 compile-checks and runs successfully on the local Mac and has been running against the live camera feed for hours. It is generating logs, CSV data, evidence images and Pushover notifications normally. We also have a well-defined V6.3 specification based on live testing. The blocker is specifically getting ChatGPT’s execution/file-generation environment to produce and validate the next .py file.

We have now set up an hourly ChatGPT task that attempts a minimal execution test: create a small text file, read it back, create a trivial Python file and compile-check it. It stays silent while execution remains unavailable and is supposed to notify us when the complete test succeeds.

Current status: the Project and all the engineering evidence remain intact, and ChatGPT can continue analysing the project normally, but reliable code execution/file generation is still unavailable. The next version is effectively waiting on that environment rather than on further debugging or design work.

This has persisted across multiple chats and enough different test cases that it doesn’t appear to be a single bad prompt, oversized video, broken Python source file, or one stalled conversation. The particularly frustrating part is that the coding project itself is currently running successfully—the hosted execution environment needed to produce the next revision is the thing preventing development from continuing.