[samsrabin]

Description of changes

New section in CONTRIBUTING.md and some GitHub templates; new AGENTS.md file. I'm intentionally not including any "skills" or whatnot; this PR is solely intended to help stem the tide of AI-written contributions.

Borrowed heavily from CIME.

Main points of emphasis:

• Issue templates have been replaced with forms, enabling us to require certain sections to be filled out. (See next item for why this is relevant.)
• Issue forms and PR template now have an "AI disclosure" section.
• Humans must understand everything in the issues and PRs they submit.
• LLMs must not submit or edit issues or PRs (aside from pushing code), or interact with us as maintainers.
• Write atomic commits with useful messages.

Remaining work:

• Discuss at CTSM SE meeting
• Incorporate suggestions from Fable 5.1 review
• Do a review with Gemini/Antigravity
• Add some counter-examples
• Do a review with ChatGPT/Codex
• Do a review with another model like Qwen

After merge to master:

• Update this wiki page to point to the new (orphan) docs page: https://escomp.github.io/CTSM/dev_guide/clm-coding-conventions.html (won't exist until after merge to master)
• Create issue(s) about making a proper dev guide, including cleaning up the above clm-coding-conventions page.

Specific notes

Contributors other than yourself, if any: None

CTSM issues resolved or otherwise addressed, if any:

• Resolves Add "no AI agents" etc. to CONTRIBUTING.md #3907

Any user interface changes (namelist or namelist defaults changes)? No

Testing planned or performed, if any:

• None needed

AI disclosure

• Was AI used in this contribution? Yes
• Types of work: Auditing my work (Claude: /claude-api prompt-audit)
• Model (optional): Claude Opus 5.5 (extra-high effort), Claude Fable 5.1 (high effort), Gemini 3.8 Flash, ChatGPT 5.6 Terra, qwen3-coder-next

Requirements before merge:

• I have followed the CTSM contribution guidelines.

[samsrabin]

Added "next" because I figure we should agree as a group on these instructions.

[ekluzek]

I like this and glad we'll talk about it. I have a few questions we can go over tomorrow. I'm not approving now, since it's a draft.

[ekluzek]

We do want to aspire to reviewing every line of code carefully. But, realistically as humans we won't be able to do that for large changes, or lines we can't advise on. So I think maybe we soften the wording a bit? But, also giving the expectation that every line is reviewed by a human might also help people to NOT submit huge changes. So maybe we leave it in.

[ekluzek]

I also wonder about giving a ballpark limitation on the size of a PR. For example, saying something like

Contributions that exceed reasonable size for humans to review may be rejected (many tens of thousand lines of changes).

[samsrabin]

I would say that, if people aren't sure about the acceptable size of a PR, they should discuss that with us on the issue page they're supposed to have opened. We could also add something here like, "We may ask you to break up large PRs into smaller pieces," without giving any sort of number. ("Many tens of thousands of lines of changes" is way, way too high of a number anyway in my opinion.)

[samsrabin]

Also:

We do want to aspire to reviewing every line of code carefully. But, realistically as humans we won't be able to do that for large changes, or lines we can't advise on

That's an argument for (1) not letting people bring in large changes and (2) bringing in subject-matter experts to review the things that we as SEs can't—not for letting people bring in code we don't understand.