What the agent does
Removing six is two jobs that look like one: rewriting imports, and admitting in the packaging that Python 2 is gone. The agent reads the packaging first. In a project like coursera-dl/edx-dl, setup.py still carries the classifiers Programming Language :: Python :: 2 and 2.7, .github/workflows/pythonapp.yml tests [2.7, 3.5, 3.6, 3.7, 3.8], tox.ini lists py27,py34,pypy, and the Dockerfile starts from python:3.4-slim. None of that can stay if the compatibility layer goes, so the pull request changes the classifiers, adds a python_requires, and rewrites the matrix. The agent states the new floor it chose and why; your prompt can fix it instead.
The imports are mechanical. edx_dl/edx_dl.py takes CookieJar from six.moves.http_cookiejar and its URL helpers from six.moves.urllib.error, .parse and .request; these become http.cookiejar and urllib.error, urllib.parse and urllib.request. test_utils.py loops with six.iteritems(strings) three times, which becomes strings.items(). edx_dl/utils.py and edx_dl/parsing.py both have from six.moves import html_parser, and that one is not mechanical.
Those two files call html_parser.HTMLParser().unescape(...), once in clean_filename and twice on the metadata string in parsing.py. On Python 3, six.moves.html_parser is the standard html.parser module, and its HTMLParser.unescape method was removed in Python 3.9. So before the agent changes anything, it runs the suite on a current interpreter and records the failure: AttributeError: 'HTMLParser' object has no attribute 'unescape', in every test that reaches a filename or a metadata block. The package has been broken on modern Python with six installed and working. The fix is html.unescape(s), a function that has existed since Python 3.4, and the agent shows the same tests passing after it.
The last step is the one a search-and-replace cannot do: find out whether six can leave the environment. The agent deletes it from requirements.txt and from the deps in tox.ini, resolves the remaining requirements, and reads what still asks for it. Here html5lib>=1.0.1 does, since html5lib 1.1 declares six>=1.9. That matters more than it looks. parsing.py calls BeautifulSoup(page) four times without naming a parser, so Beautiful Soup uses the best one it can import. With six forced out, import html5lib raises ModuleNotFoundError, Beautiful Soup falls back to html.parser without an error, and the tests still pass on a different parser. The agent leaves six installed as html5lib's dependency and says so.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Remove the six compatibility library from this Python project's own code. Read setup.py or pyproject.toml, the requirements files, tox.ini, the CI workflows and the Dockerfile, and list every place that still declares or tests Python 2. Unless a file says otherwise, set the minimum to the oldest Python 3 release that still receives security fixes, state it, and update the classifiers, python_requires, tox envlist and CI matrix to match. Before editing any code, create an environment on a current Python release, install the requirements and run the existing tests; record every failure with its exception. List each six use: six.moves imports, iteritems, itervalues, string_types, text_type, PY2 or PY3 branches, add_metaclass, and any from __future__ import. Replace each with the Python 3 form. Where a six.moves name maps to a standard library attribute that later Python versions removed, such as HTMLParser.unescape, use the current replacement and say so. Leave __future__ imports alone and report them. Then remove six from the project's requirement files, resolve the remaining requirements and print which installed packages still require six; do not uninstall it if another dependency needs it, and explain what would change if it were removed. Run the tests again on the oldest and newest supported versions and show before and after. Change no behaviour beyond what the port requires. Open a pull request.",
"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
A pull request with the import changes in edx_dl/edx_dl.py, edx_dl/utils.py, edx_dl/parsing.py and test_utils.py, the html.unescape fix, and the packaging edits in setup.py, requirements.txt, tox.ini and the workflow. Its description has a before and after for the test suite on a current Python: the AttributeError failures first, the passing output second. A table maps each six name to what replaced it.
A separate section answers whether six is gone. For this repository the answer is no: it is out of the project's own imports and requirement lines, and still installed because html5lib requires it. The description quotes the resolver output that shows this and notes the four BeautifulSoup(page) calls that pick their parser by what is importable, as a risk to fix by naming the parser.
The reviewer checks the chosen Python floor against who still installs the tool, and checks that .travis.yml and the Dockerfile were either updated or listed as stale. The agent also reports what it saw and left alone, such as the regular expression in parsing.py that Python now warns about for its \d in a non-raw string.
Run it on every Monday morning
Compatibility code comes out best in small, regular pieces. Add an automation to the Project with a schedule trigger, a cron of 0 7 * * 1 and your timezone, and save the prompt above. Each Monday a run starts with that prompt and opens a pull request if anything is left to remove. After the pull request merges, the next run finds no six imports, changes nothing and says so in its answer.
To keep the weekly run useful after that, widen the prompt's first sentence to name the next layer, for example future or mock, and keep the rest. A schedule fires at most hourly, and turning the automation back on after a pause moves it to its next slot instead of firing the Mondays it missed.
The use case this job follows, with its automation and trigger: Upgrade dependencies weekly.
Why run it on Rush?
The useful discoveries in this job only appear when the code is executed on an interpreter newer than the one it was written for. The run creates that interpreter in its own sandbox, runs the suite before touching a line, and keeps the failing output. That is how a removed standard library method shows up as a fact in the pull request and not as a bug report from a user three weeks later.
The dependency question is settled the same way. Resolving the requirements and then importing each package with six absent takes a throwaway environment, which is what a sandbox is. The reviewer gets the resolver's own words on why six stays installed, next to a diff small enough to read in one sitting.
Frequently asked questions
What replaces six.moves.urllib?
The standard library modules it pointed to on Python 3: urllib.request for urlopen, Request and the opener classes, urllib.parse for urlencode and urlparse, and urllib.error for HTTPError and URLError. six.moves.http_cookiejar is http.cookiejar. The names inside those modules are unchanged.
Why does HTMLParser().unescape fail after the port?
It fails before the port too, on Python 3.9 or newer. The method was deprecated in Python 3.4 and removed in 3.9, and six.moves.html_parser only forwards to the standard module. Use the function html.unescape, which takes the string and returns it with entities decoded.
Can I uninstall six once my code no longer imports it?
Only if nothing else you install needs it. Libraries such as html5lib 1.1 still declare it, so a resolver puts it back. The agent removes it from your own requirement lines and prints which packages keep it in the environment.
Does the agent drop Python 2 without asking?
It treats removing six as the decision to drop Python 2, because the two cannot be separated, and it makes every packaging change that follows from it in the same pull request. If the prompt names a floor, it uses that. Nothing merges without your review, and the description lists each file that declared Python 2.
More jobs like this: Upgrade dependencies weekly
- Replace passlib in a FastAPI App Without Breaking Stored Hashes (passlib)
- Update Gin Dependencies Every Week (Gin)
- Update GitHub Actions in a CMake Project's CI (CMake)
- Update the GitHub Actions in a Cobra CLI's Workflow (Cobra)
- Upgrade Pinned GitHub Actions to the Latest Releases (GitHub Actions)
- Update the GitHub Actions in a Gradle Build's Workflow (Gradle)