What the agent does
Python 3.12 removed the imp module after a long deprecation, so any import imp now raises ModuleNotFoundError: No module named 'imp'. The common place is a setup.py that loads a version file with imp.load_source, which turns every pip install . into a build failure on 3.12. The run starts with that traceback, and the agent clones the repository into its own cloud sandbox.
In a Python tool like MJL85/natlas, setup.py lines 20 and 21 read import imp and _version = imp.load_source('', 'natlas/_version.py'), and natlas/_version.py holds __version__ = "0.12.1". The rest of setup.py lists graphviz, pydot, pysnmp>=4.2.5, pyparsing and netaddr>=0.7.14 as install_requires, the natlas package and a natlas/natlas.py script. The agent reproduces the failure with a 3.12 interpreter: pip install . stops inside the build backend with ModuleNotFoundError: No module named 'imp', and uv pip install . reports the same line under The build backend returned an error.
The replacement is the importlib.util sequence the Python documentation gives for imp.load_source: spec_from_file_location('natlas_version', 'natlas/_version.py'), module_from_spec, then spec.loader.exec_module, after which _version.__version__ reads as before. The agent keeps the version in one file rather than duplicating the string into setup.py. With that change pip install . on 3.12 completes and importlib.metadata.version('natlas') returns 0.12.1.
It then runs the tool, because 3.12 removed more than imp. natlas-cli.py line 33 imports LooseVersion from distutils.version, which 3.12 also dropped, and uses it at line 223 to compare a module's require_api against natlas.__version__. The agent replaces that comparison in the same pull request, with packaging.version.Version as a new requirement or a tuple comparison of the dotted parts, and says which it chose. The module loader at line 122 uses __import__ and needs no change.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "pip install . fails on Python 3.12 with ModuleNotFoundError: No module named 'imp'. Find every use of the imp module, replace imp.load_source in setup.py with importlib.util so the version is still read from its file, check the entry point for other modules Python 3.12 removed such as distutils, fix those too, install and run the package on Python 3.12 to confirm, and open a pull request that quotes the errors and explains each replacement.",
"resources": [
{
"url": "https://github.com/your-org/your-app"
}
]
}- Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
- Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
- Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.
What comes back
What comes back is a pull request touching setup.py and natlas-cli.py. Its description quotes the imp error from the build backend and the distutils error from the CLI, shows pip install . completing on 3.12 with the version read back, and lists the Python release that removed each module.
A reviewer checks that the version is still read from natlas/_version.py rather than copied, that an older interpreter still installs the package (the importlib.util calls exist there too), and that the version comparison keeps its meaning for dotted strings such as 0.12.1. If packaging is added, it appears in install_requires and in whatever lock the project keeps.
The session keeps the failing install, the passing install and the CLI's output, so the reviewer can see the proof without a 3.12 interpreter at hand.
Run it on every new Python version in CI
Create an automation in the Project with this prompt and a github trigger on workflow_run.failure. When a workflow fails after a Python version moves, such as actions/setup-python picking up 3.12, the run starts with the event after the prompt, reads the failing job's log, and opens the pull request.
A repository without CI can run the same prompt on a schedule trigger once a month, installing the package on the current Python release; a run that finds nothing to fix ends without a pull request and says so.
The use case this job follows, with its automation and trigger: Fix a bug from an alert.
Why run it on Rush?
The failure sits in setup.py, so it never shows in a test suite that already has the package installed. A Rush run installs from source on the Python version that broke it, in its own sandbox, and keeps the build backend's traceback next to the fix, so the pull request proves the install rather than asserting it.
Removed standard-library modules come in groups: imp and distutils left in the same release. One run fixes the install, runs the entry point, and finds the second removal before a user does. Several packages with the same imp.load_source pattern can be fixed at once, one run and one pull request each.
Frequently asked questions
Why not read the version with a regular expression instead of importing the file?
That works too, and the agent uses it when the version file has imports of its own that would run on exec_module. Here _version.py is one assignment, so loading it with importlib.util keeps __version__ as the single source and changes the least.
Does this break installs on older Python versions?
No. importlib.util.spec_from_file_location and module_from_spec exist on every Python 3 release the package can run on, so the same setup.py installs on an older interpreter and on 3.12 alike.
What replaces distutils.version.LooseVersion?
packaging.version.Version is the standard replacement and is what pip itself uses. It adds packaging to the requirements; for a project that wants no new dependency, the agent compares tuples of integers from the dotted string instead and says so.
Will it also move the project to pyproject.toml?
Not in this pull request. The ask is the 3.12 failure, and a packaging migration is its own change. The agent notes in the description that setup.py could become pyproject.toml with a dynamic version if the maintainer wants that next.
The same job on other stacks
More jobs like this: Fix a bug from an alert
- Fix "ModuleNotFoundError: No module named 'kafka.vendor.six.moves'" on Python 3.12 (kafka-python)
- Fix "ModuleNotFoundError: No module named 'moviepy.editor'" (MoviePy)
- Fix "ModuleNotFoundError: No module named 'pip.req'" in setup.py (pip)
- Fix "ModuleNotFoundError: No module named 'pipes'" on Python 3.13 (pipes)
- Fix "ModuleNotFoundError: No module named 'pkg_resources'" after a setuptools upgrade (setuptools)