Coding-Agent Rules for Data Science

A reusable, IDE-agnostic ruleset that steers AI coding agents toward
reproducible, analysis-first data-science work in R and Python — for
scientists and researchers doing agentic coding across any field (ecology,
genomics, economics, physics, ML, general data science, and more). Instead of
copying a one-size config, you point your agent at this repo and it builds a
personalized copy tuned to your field, your tools, your machine, and your
preferences.

This is a living document: the ruleset is actively maintained and will keep
evolving over time. What you set up is a static snapshot, so re-run the setup
whenever you want to pull the latest.

Quick Start Prompt

Paste the prompt below into your coding agent:

(Cursor, Claude Code, Windsurf, Copilot, Cline, Codex/AGENTS.md, and others)

Set up a personalized copy of these data-science coding rules for me.
The rules live in this public repo: https://github.com/NCEAS-AI/Coding-Agent-Rules-for-Data-Science
Read the setup guide at https://github.com/NCEAS-AI/Coding-Agent-Rules-for-Data-Science/blob/main/agent-setup/SETUP.md
and follow it: figure out which IDE/tool I'm using and where its rules files go,
inspect (or ask about) my system, ask me the preference questions, then build the
ruleset from agent-setup/rules-template.md and show me the whole thing before you
write any file.

The agent will:

1. Ask how you want to run setup — let it probe your machine and auto-answer what it can, or answer everything yourself; fast path or full interview.
2. Figure out which tool you're using and where that tool expects rules files.
3. Inspect (or ask about) your system — cores, RAM, GPU, OS, installed languages — never assume anyone else's.
4. Ask the parts that are genuinely your call (your field, languages, geospatial/parallel/GPU work, server vs. laptop, emoji comments, strictness, layout, and more).
5. Assemble a personalized ruleset from agent-setup/rules-template.md, write it into the correct location and format for your tool, and print the whole thing back for you to review.

Prefer to do it by hand? agent-setup/SETUP.md is readable on its own — the
phases and the full questionnaire are right there. And if you just want a ruleset
without any of the setup dance, copy
personal-rules/data-science-rules.mdc
straight into your IDE's rules folder.

What are these rules? (Nothing to install, nothing runs)

These rules are just a text file. A rules file is a plain Markdown document
that your coding agent reads and keeps in its context to guide how it behaves —
the same way you might paste standing instructions at the top of a chat, except
your IDE loads it automatically on every request. There's nothing to install,
no software to run, and nothing that phones home. "Setting it up" just means
putting the Markdown file where your IDE looks for its rules.

Because it's only text, you can read every word before you use any of it. If
you're wary of pointing an agent at a stranger's repo, don't take our word for
it — open the files and see exactly what the agent would be told to do:

• personal-rules/data-science-rules.mdc —
  the author's own working ruleset. This is the best place to start: it's a real,
  filled-in example of what these conventions look like, and you can copy it as-is
  or just skim it to get the gist.
• agent-setup/rules-template.md — the full
  build source the setup reads from, with every section it can include.
• agent-setup/SETUP.md — the exact, human-readable
  steps the agent follows to tailor the rules to you.

What's in here

Design notes

• IDE-agnostic by construction. The setup detects the target tool and
  writes natively for it, falling back to the cross-tool AGENTS.md standard when
  you use several tools or aren't sure.
• Universal floor vs. your call. Safety and reproducibility items (real-data
  integrity, rebuildable environments, worker cleanup, CRS verification, guardrails)
  are included for everyone. Cosmetic and workflow choices (emojis, layout, README
  discipline, strictness) are asked, not assumed.
• Your machine, not mine. The system section is rewritten from a live inspection
  of your hardware, so utilization caps and worker sizing fit what you're actually on.
• Rules written as rules. Each bullet is one timeless instruction with a short
  reason — no dates, incident stories or current measurements, which go stale and
  then contradict reality. History belongs in a project's logs/DECISIONS.md. This
  follows Anthropic's, OpenAI's and Cursor's guidance: keep always-on rules short,
  state each rule once, explain why, and save emphasis for true invariants.
• Living source. This repo is actively maintained — the rules will be refined
  and expanded over time. It's the evolving source; see Static once set up
  below for how that interacts with a copy you've already generated.
• Static once set up. The generated ruleset is a snapshot. It does not phone
  home or auto-update from this repo — re-run the setup yourself if you want a refresh.

License

Released under CC0 1.0 Universal — public domain. Copy, modify, and use
it for any purpose, no attribution required.

Cite us!

Citation metadata lives in CITATION.cff — GitHub renders a "Cite this
repository" button from it, and a Zenodo DOI will appear at the top of this README once
archived.

📋 Copy-paste citation:

Broderick, C. (2026). Coding-Agent Rules for Data Science (Version v1.0.0)
[Software]. Zenodo. https://doi.org/10.5281/zenodo.21364495