docs: optional AI-agent skill for CTSM (community contribution, no code changes)

[ktwu01]

Hi CTSM maintainers,

I'm Koutian Wu (ktwu@utexas.edu), a researcher working on AI-assisted scientific
computing workflows at UT Austin. I write agent "skills": structured
markdown documentation that lets an AI coding agent (e.g. a CLI coding agent or
AI IDE) install, build, configure, run, debug, and contribute to a model with
less hand-holding than a cold prompt. I'm doing this for a family of
Earth-system models; CTSM is one of them, so I'll note up front that the
shape of this message is shared across that family. The CTSM skill itself is
specific to CTSM, and lives here:

https://github.com/earth-space-ai/ctsm-skill

This PR is a single small, optional pointer, concretely, adds one line under a 'Community resources' section in README.md linking the skill repo, noting
that an unofficial, community-maintained agent skill exists. It changes no model
code, no build system, and no scientific defaults. If you'd rather not link it,
that's completely fine: close this and I'll keep maintaining the skill on my side.
The offer stands either way.

What a "skill" is

A directory (SKILL.md + reference/*.md): a routing hub the agent reads
first, then deep-dive docs it loads on demand. It encodes the procedural
knowledge usually picked up by working alongside an experienced developer.

Honest disclaimers (stated up front in the skill)

• It's a helper for new users, not a gold-standard reference. Upstream
  remains the source of truth; where the skill and upstream disagree, upstream
  wins and I treat that as a bug.
• It was assembled with AI assistance, and AI makes mistakes (wrong paths,
  drifted line numbers, stale fields, hallucinated flags). The skill says so and
  tells the agent to cross-check upstream before acting.
• It is MIT-licensed on my side; CTSM stays governed by its own upstream
  license, which the skill links to and does not relicense.

Why it might be worth a look

It helps onboarding. The most-exercised skill in this family so far is the Noah-MP one, so I'll give that as the data point and flag plainly that it's a sibling skill, not the CTSM one: I used it to help an undergraduate go from zero to a scoped, runnable regional simulation plan (Texas, 12.5 km, NLDAS-2 forcing, on TACC) in about 3 DAYS, spin-up reasoning and a budget/smoke-test plan included. The CTSM skill is built the same way, drawing on your community's own docs.

Credit where it's due

The skill draws on the ESCOMP/CTSM repository, Sphinx docs, and wiki maintained by the NCAR/CGD CTSM software-engineering and science teams; it restructures that material for agent use. The acknowledgments section names the upstream maintainers and the
specific docs it draws on. Any errors or oversimplifications in the skill are
mine, not the upstream community's.

Happy to collaborate

I'd genuinely enjoy collaborating on AI-assisted workflows for CTSM. If
there's an official place such a skill should live, conventions to follow, or
checks to enforce, I'm glad to adapt it.

Either way, thank you for maintaining CTSM.

Koutian Wu
ktwu@utexas.edu
UT Austin
Google Scholar: https://scholar.google.com/citations?user=s9w1k-cAAAAJ
LinkedIn: https://www.linkedin.com/in/ktwu01/