AI Anonymous · from JAB Systems / October 23, 2026

AI suggested a package. Does it exist?

Before you run the install command, check the distribution name and release. A registry record is only the first check.

PyPI / AI-assisted coding · Tested authored examples · Human dependency review

In this field note · 7 sections
  1. Before you install, write down the claim.
  2. Download a checker you can read.
  3. Ask the registry, without installing the suggestion.
  4. Three authored cases, three useful outcomes.
  5. Found is the start of the review.
  6. Keep an unresolved answer.
  7. Take the package-checking kit.
Warm ivory editorial poster reading “AI suggested a package. Does it exist?” A magnifying glass enlarges a label reading “package name” on a kraft box. An open box behind it holds a question-mark card. Supporting text reads “Check the name before you install” and “A registry record is only the first check.” Footer: AI Anonymous, from JAB Systems, jabsystems.io. AI-generated conceptual artwork.
Original AI-generated conceptual artwork. Illustrative boxes; no package endorsement.
01 / Field guide

Before you install, write down the claim.

An AI-drafted setup may include a plausible package name and a confident install command. Treat that dependency as a proposal. Copy the exact distribution name, requested version and reason it is needed into the review checklist. Check whether Python’s standard library already covers the task.

This is an established packaging workflow applied to AI-assisted coding, not a new PyPI feature. The Python Packaging User Guide explains that the name used in an import can differ from the distribution name installed from a registry. An import statement alone is insufficient evidence for an install command. Ask for clarification instead of guessing a similar name.

The exercise below checks public PyPI metadata. For a workplace package on an approved private index, use that organization’s process; absence from public PyPI does not establish nonexistence.

02 / Field guide

Download a checker you can read.

Use an existing Python 3 installation. We ran this on Windows with Python 3.13.15. The checker uses the Python standard library; there is no dependency-install step. The commands create a new dated folder and stop if it already exists. Choose another empty location for a rerun.

New-Item -ItemType Directory -Path .\pypi-practice-2026-10-23 -ErrorAction Stop
Set-Location .\pypi-practice-2026-10-23
Invoke-WebRequest -Uri "https://jabsystems.io/ai-anonymous/Social/check-the-package-practice.zip" -OutFile .\practice.zip
Expand-Archive -LiteralPath .\practice.zip -DestinationPath .\practice
Get-ChildItem .\practice
Get-Content .\practice\check-packages.py
Get-Content .\practice\package-fixtures.json

Read the script and fixtures before proceeding. It sends project names in HTTPS requests to pypi.org, prints metadata and does not download package archives or follow returned links. Do not put private project names or secrets in this public-index exercise. Running this reviewed checker still executes local Python code; it is not a sandbox.

03 / Field guide

Ask the registry, without installing the suggestion.

From the same PowerShell session, run:

python --version
python .\practice\check-packages.py | Set-Content -Encoding utf8 .\my-report.json
Get-Content .\my-report.json

The PyPI JSON API exposes project metadata and release records. The core request in the complete downloadable script is:

url = "https://pypi.org/pypi/" + quote(name, safe="") + "/json"
request = Request(url, headers={"User-Agent": "AIAnonymousPackageExercise/1.0",
                                "Accept": "application/json"})
with urlopen(request, timeout=15) as response:
    data = json.load(response)

The script validates distribution-name syntax and applies the standard name normalization: lowercase, with runs of dots, underscores and hyphens reduced to one hyphen. It reuses each project response within this run. Its two public requests cover three cases.

Release checking uses an exact key in releases. This deliberately small checker is not a requirements parser or version resolver: it does not handle equivalent version spellings, ranges, extras, direct URLs, platform tags or transitive dependencies. A missing key asks you to inspect the actual releases, not guess another install command.

04 / Field guide

Three authored cases, three useful outcomes.

Observed October 23, 2026, from actual requests on Windows / Python 3.13.15:

  • packaging, release 26.0: project_returned / release_files_listed. Two files listed, zero marked yanked, both with a declared Python requirement of >=3.8.
  • packaging, release 9999.0: project_returned / release_key_not_listed. A project can exist while the requested release key is absent.
  • ai-anonymous-fictional-package-20261023, release 1.0: not_in_pypi, HTTP 404. This is an intentionally fictional name, not an allegation about a real project.

The missing name and version were authored for this exercise. They are not errors observed from an AI model. Download the recorded JSON report; your live result may change. Every case retains trust_review: not_checked and compatibility: not_checked. Exit 0 means a report was produced, not that installation is approved. A declared Python requirement is metadata, not a successful compatibility test.

05 / Field guide

Found is the start of the review.

Independently locate the upstream project’s documentation through a source you trust. Compare its stated distribution name, registry entry, source repository and intended function with the proposal. Do not treat a model-supplied URL, familiar spelling or returned project link as ownership proof. The checker labels its links as coming from the project’s latest metadata; those links may not describe the older requested release.

PyPI’s provenance guidance and source-code review guidance describe further checks. Record maintainer/source identity, relevant release history, provenance, license, dependencies and suitability for your environment. A valid origin does not by itself establish benign code. This exercise does not perform those checks or endorse a package.

If the project or version is unclear, stop and resolve it with the author or an approved maintainer. For mixed public/private environments, follow your organization’s index policy: pip’s installation documentation warns about dependency confusion when adding public/private sources with --extra-index-url. This tutorial changes no index configuration.

An optional proposed AI review prompt extracts dependency claims without running commands or silently repairing names. We did not execute or evaluate it. Verify any extraction yourself, and use only input appropriate for your approved tool.

06 / Field guide

Keep an unresolved answer.

HTTP 404 means this public PyPI endpoint did not return that project at check time. It does not rule out private indexes, other registries, spelling mistakes or an upstream source-only distribution. Other HTTP failures, including rate limits, and request/parse errors are reported as unresolved. The script does not retry automatically. Wait and investigate before retrying; do not label an outage as an invented dependency.

PyPI documents that JSON metadata is supplied at upload time and can differ from archive contents. File counts, yanked flags and requirement fields are observations to investigate. They do not establish safety, compatibility, completeness or whether the package implements the API your script needs.

Tested: two live metadata requests, three authored cases, local boundary checks and extraction/rerun of the downloadable practice kit. Not tested: an AI model, hallucination frequency, package installation or imports, archive contents, source-code security, malware, dependency resolution, real organization workflows, other operating systems or performance. No model output, speedup, security guarantee or approval to install is claimed.

07 / Field guide

Take the package-checking kit.

Practice ZIP · Complete checker · Authored fixtures · Observed report · Human review checklist · Proposed prompt · Practice README

Primary sources reviewed October 23, 2026. The portrait is original AI-generated conceptual artwork, visually reviewed. Its boxes and labels are illustrative.

Explore more AI Anonymous field notes →

Share this post.

Found this useful? Pass it along. Download the artwork, add your own take, and tag @jabsystems.io so I can see your post.