Dairy Data Tools

Free tools for dairy nutritionists and their AI agents. Give your agent a farm's own DairyComp and EZfeed exports and ask about a pen: how it's doing, what a ration change did, and whether it's just the season. The numbers come from small, tested scripts, not from the AI's guesswork.

By Pádraig Lucey, MVB, MPVM, Ph.D. (Dairy Data Science).

Get it

In the Claude app (desktop or claude.ai, a paid plan with code execution on): Customize > Plugins > Add marketplace > padge01/dairy-data-tools, then Add Dairy Data Tools. Attach the farm's export files to a chat and ask.

In Claude Code:

claude plugin marketplace add padge01/dairy-data-tools

claude plugin install dairy-data-tools@dairy-data-tools

In other agents (Codex, Cursor, Gemini CLI and others that read Agent Skills):

npx skills add padge01/dairy-data-tools

In Python, without an agent (Python 3.10 or later; nothing else to install):

pip install dairy-data-tools

dairy-data-tools pen --folder "Farm Exports" --pen 21

Every question has a command: dairy-data-tools --help lists them. The ration change is dairy-ration-change.

What's in it

Ask in plain words; the right skill picks itself. No commands to learn.

Plain Python, checked number for number against the author's earlier R version: every count says what it's out of, milk is shown as a spread, and the farm is compared with itself, never with published targets.

A sample herd (sample-herd/): a made-up California dairy's exports, to try every skill before using a real farm's.

Its neighbours, separate tools from the same place: BESTPRED (USDA's best prediction of 305-day and mature-equivalent milk; in this marketplace, and at https://bestpred.dairydatascience.com/mcp) and the Dairy Ration Lab (a free ration balancer).

What it does with your files

• The scripts read the export files where they are. They change nothing, keep nothing, and send nothing over the internet. They use Python's standard library only.
• In Claude's chat the files you attach and the scripts run in Claude's own code sandbox. What the agent reads goes to the AI company as part of the conversation, as anything you attach does, unless you run a model on your own computer.
• Event remarks, protocols and technician names are left out of what the scripts report, because they are free text typed on the farm and could contain anything, including instructions aimed at an agent.
• Nothing reads DairyComp's own files or backups, and no figure DairyComp calculates for itself (such as 305-day or mature-equivalent milk) is used: the kit works from raw records.

Good to know

• Weights are in pounds. Pens are compared with themselves and with the farm's other pens, never with published targets.
• It's a teaching and planning tool, not veterinary or nutrition advice: check anything before it changes what cows are fed.
• DairyComp 305 and EZfeed are Valley Ag Software's products; this project isn't affiliated with them.

Cite it

If it helps your work, please cite it: GitHub's "Cite this repository" button gives the reference (CITATION.cff).

Licence

Code under Apache-2.0 (LICENSE); the skills' text under CC BY 4.0 (LICENSE-skills.md). Questions and fixes: issues.

Working on it

• plugins/dairy-data-tools/: the plugin people install (skills, sample herd, README, licences).
• tools/dairy.py, tools/dairytools/ and the ration-change scripts in tools/: the one copy of the code; python tools/sync_scripts.py copies it into each skill and writes this README. pyproject.toml builds the same code as the dairy-data-tools package on PyPI.
• tools/make_sample_herd.py: writes the sample herd (fixed seeds, same files every run).
• tests/: python -m pytest -q tests (made-up files with known answers, the sample herd, the directory's file limits).