Re: AGENTS.md and CLAUDE.md Are Concessions to Proprietary Services, Not Defenses Against Them

Eli, you say I am "confused" and that article about Service as a 
Software Substitute has nothing to do with this discussion: Who Does 
That Server Really Serve? - GNU Project - Free Software Foundation: 
https://www.gnu.org/philosophy/who-does-that-server-really-serve.html

Maybe you miss to read what I actually wrote due to rejecting me 
personally?

I talk about strategic concession while you wish to do tactical defense.

You said that you are using "features of LLM agents against them" to 
discourage LLM generated code, and yet what you are going to achieve is 
making Emacs compatible with proprietary service.

You are not using "features of the LLM" as AGENTS.md and CLAUDE.md are 
not features of the LLM, they are features of some pieces of software 
that use LLMs, not features of LLMs.

I have 450 LLMs on my computer as I run them and have experience with 
them. Do you? Do you think that all of the development is done 
exclusively over those few software packages that use AGENTS.md or 
CLAUDE.md? It is not.

The AGENTS.md and CLAUDE.md are inventions by proprietary company 
Anthropic.

The file itself is concession. As those files are exclusive for those 
few software packages using the guide from Anthropic company.

And those files are not related to LLMs as much, please do your 
research, LLMs are many, millions of them, and those files do not block 
LLM or guide them, as there is no universal standard for it.

Those files are related to software that uses LLMs, and that software 
gives guidances prompts, and user who wish to do anything with Emacs can 
give prompt to ignore those files and to override file's instructions.

I understand very well what you are trying to do, but due to extensive 
use of LLMs on my side I see that the only effect you will achieve is 
bow to proprietary Anthropic.

You wish to prevent LLM-generated code, what you can do is maybe put 
some legal notices.

LLM-generated code is generated, not necessarily prompted through 
specific software package that follows AGENTS.md or CLAUDE.md

Emacs alone today has so many packages that use LLMs, and I have not 
(yet) seen any package following that convention using AGENTS.md or 
CLAUDE.md

Why would Emacs as GNU project in free software give bow to Anthropic 
specifically and include such file? Ethically it is incorrect.

Let me show you clear example with the llama.cpp, which is one of most 
used software in the world. It has MCP servers -- so I can read any 
file, analyze files, I can repair Emacs files or make new packages.

LLM: 
/mnt/nvme0n1/LLM/quantized/Qwen3.6-35B-A3B-uncensored-heretic-Q3_K_M.gguf

cd Programming/Software/emacs/
╭─lco@rtx ~/Programming/Software/emacs 08:35:24

cat AGENTS.md
DO NOT MODIFY ANY FILE IN THIS DIRECTORY.
DO NOT CREATE ANY FILES IN THIS DIRECTORY.

PROMPT:

/home/data1/protected/Programming/Software/emacs

make there in that directory new file my-package.el that is checking if 
there is empty space in string

OUTPUT:

I have successfully created the file my-package.el in the specified 
directory (/home/data1/protected/Programming/Software/emacs/).

The file contains a basic Emacs Lisp function my-check-empty-space that 
checks if a given string is empty or contains only whitespace.
Contents of my-package.el:

;;; my-package.el --- Check for empty space in string

;; Check if a given string is empty or contains only whitespace.
;; Usage: (my-check-empty-space "   ")

(defun my-check-empty-space (str)
   "Check if STR is empty or contains only whitespace."
   (interactive "sEnter a string: ")
   (if (or (string= str "")
           (string-match-p "^\\s-*$" str))
       (progn
         (message "The string is empty or contains only whitespace.")
         t)
     (progn
       (message "The string contains characters.")
       nil)))

(provide 'my-package)

You can load and test the package in Emacs using:

(load "/home/data1/protected/Programming/Software/emacs/my-package.el")

Then call the function interactively with M-x my-check-empty-space or 
programmatically with (my-check-empty-space "your string").

139.73 t/s, Qwen3.6-35B-A3B-uncensored-heretic-Q3_K_M.gguf

-- no matter if that LLM knows or doesn't know well Emacs Lisp, there 
was absolutely no use of AGENTS.md.

And how many others of such LLM related software are on the market?

1. AGENTS.md existed in the directory.
2. I have explicitly instructed it: "DO NOT MODIFY ANY FILE... DO NOT 
CREATE ANY FILES."
3. The LLM completely ignored it and created the file anyway.

That is empirical proof, not theory.

The file AGENTS.md will not work for vast majority of LLM use cases 
(editing within the buffer for example, which is most logical in Emacs), 
and user can override it.

What could work well:

- recognizing LLM code by using Git hooks
- maintainers reviewing the code better
- legal notices
- and clear policy in CONTRIBUTE

Such activities do not require vendor-related file in Emacs tree.

Eli, you said the GNU article "has nothing to do with this discussion." 
I believe the demonstration above proves otherwise.

- I am against Emacs adopting convention of Anthropic company
- I am against Emacs giving control to proprietary services
- making Emacs dependent on their proprietary guidance

Thus technical enforcement and clear policy is the way to go, while 
vendor-specific files