AI Anonymous / October 23, 2026 — PyPI metadata practice Full guide: https://jabsystems.io/ai-anonymous/pypi-check-the-package Read check-packages.py and package-fixtures.json before execution. Requires an existing Python 3 installation and HTTPS access to pypi.org. Tested on Windows / Python 3.13.15. No third-party Python dependency required. From the extracted folder: python .\check-packages.py The script reads fixtures next to itself, makes two cached-per-run project GETs, and prints JSON. It does not invoke pip, install/import suggested packages, download wheels/sdists, execute their code, or follow returned project links. Python and the supplied checker itself execute locally; inspect code before running. Authored fixtures: packaging 26.0; packaging 9999.0 (intentional absent release); ai-anonymous-fictional-package-20261023 1.0 (intentional absent project). Observed October 23: release_files_listed (2 files, 0 yanked, >=3.8), release_key_not_listed, not_in_pypi (404). Live state can change. All trust_review and compatibility fields stay not_checked. Exit 0 means report produced, not packages approved. Input is a small authored JSON list, not requirements.txt or PEP 440 parsing. Release lookup uses exact JSON keys: equivalent version spellings are not resolved. 404 is scoped to public PyPI, not private indexes or all distribution channels. HTTP errors other than 404 and request/parse errors become unresolved. No automatic retries, bulk service, persistent cache, security audit, dependency resolution, platform/wheel selection, or compatibility check. Back off on rate limits; do not turn an outage into a missing-package claim. Project links come from latest project metadata; they are not verified ownership or necessarily historical release documentation. Never open them blindly. No AI model, package source audit, malware detection, installation, performance, real organization workflow, or operating systems other than Windows tested. Suggested next step: independently find upstream docs and compare the exact distribution, release and source identity before any separate installation decision. The review prompt is optional and unexecuted. Use only approved non-sensitive input. Social files are prepared for manual posting, not scheduled or published.