auntiepypi 0.5.0

auntiepypi — both ends of the Python distribution pipe for the AgentCulture mesh.

auntiepypi

auntie (Python distribution: auntiepypi) is both a CLI and an agent that maintains, uses, and serves the CLI for managing PyPI packages. It overviews packages on pypi.org, detects PyPI-flavored servers running locally, and starts/stops/restarts declared servers — informational first, actionable on demand.

Status: v0.5.0 — lifecycle verbs landed. auntie up, auntie down, and auntie restart are first-class verbs against declared servers (managed_by ∈ {systemd-user, command}). The _actions/ strategy contract has widened from a single apply() to sibling start / stop / restart per strategy. command-managed servers are now PID-tracked ($XDG_STATE_HOME/auntiepypi/<slug>.pid) with a Linux port-walk fallback. The bare invocation (auntie up with no target) is reserved for v0.6.0's first-party PyPI server.

Quick start

uv tool install auntiepypi
auntie --version
auntie overview --json | jq '.sections[] | select(.category == "servers")'
auntie overview requests            # deep-dive into a PyPI package
auntie doctor                       # diagnose declared servers (dry-run)
auntie doctor --apply               # act on actionable remediations
auntie up <name>                    # start one declared server
auntie down --all                   # stop every supervised server
auntie restart <name>               # atomic for systemd-user; stop+start for command

Example servers-section output (one declared server):

{
  "category": "servers",
  "title": "main",
  "light": "green",
  "fields": [
    {"name": "flavor", "value": "pypiserver"},
    {"name": "port",   "value": "8080"},
    {"name": "status", "value": "up"},
    {"name": "source", "value": "declared"}
  ]
}

For the overview and doctor to show anything, add the relevant blocks to your repo's pyproject.toml:

[tool.auntiepypi]
packages = ["requests", "pip"]
scan_processes = false             # opt into /proc scan; same as `--proc`

[[tool.auntiepypi.servers]]
name = "main"
flavor = "pypiserver"
port = 8080
managed_by = "systemd-user"
unit = "pypi-server.service"

auntie doctor walkthrough

auntie doctor classifies every known server into one of four categories and explains exactly what to do next:

$ auntie doctor
# auntie doctor
summary: 1 actionable, 1 half-supervised, 1 skip, 0 ambiguous (3 total)

  main          down     declared    managed_by=command
      diagnosis: down; would dispatch managed_by='command'
      remediation: auntie doctor --apply

  stale         down     declared    managed_by=systemd-user
      config_gap: managed_by="systemd-user" requires `unit`
      diagnosis: half-supervised; --apply would delete this entry
      remediation: add `unit = "…"` to keep supervision, or run `auntie doctor --apply`

  pypiserver:8080  up    port        observed; not declared
      remediation:
          [[tool.auntiepypi.servers]]
          name = "…"
          flavor = "pypiserver"
          port = 8080
          managed_by = "manual"

(dry-run; pass --apply to act on 2 remediations)

Pass --apply to act. A numbered snapshot is written before any edit:

$ auntie doctor --apply
wrote pyproject.toml.1.bak (rollback: mv pyproject.toml.1.bak pyproject.toml)
...

If two entries share the same name, use --decide to choose which to keep (or remove):

$ auntie doctor --apply --decide=duplicate:main=1
wrote pyproject.toml.1.bak (rollback: mv pyproject.toml.1.bak pyproject.toml)
wrote pyproject.toml: removed [[tool.auntiepypi.servers]] entry 'main' occurrence 1 (lines 7-12)
...

See docs/about.md for the longer non-technical explainer. systemd-user unit templates for pypiserver / devpi-server live in docs/deploy/.

Develop

uv sync                          # install + dev deps
uv run pytest -n auto -v         # tests
uv run auntie --version          # smoke
uv run pre-commit install        # enable lint hooks

Quality pipeline mirrors the rest of the AgentCulture mesh: black, isort, flake8 (+ flake8-bandit, flake8-bugbear), pylint, bandit, markdownlint-cli2. CI runs on every PR + push to main.

Trusted Publishing

ghafi provisions the pypi / testpypi GitHub Environments and .github/workflows/publish.yml follows the same OIDC Trusted Publishing pattern every sibling uses — no secrets in the repo.

License

MIT. © 2026 Ori Nachum / AgentCulture.

— Claude