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

On 4 August 2026 09:36:32 GMT+03:00, Jean Louis <bugs@gnu=2Esupport> wrote:
> On 2026-08-04 09:16, Yuri Khan wrote:
> > On Tue, Aug 4, 2026 at 9:24=E2=80=AFAM Jean Louis <bugs@gnu=2Esupport>=
 wrote:
> >=20
> >> Apply that logic here: If you didn't have proprietary LLMs (Claude,
> >> ChatGPT, etc=2E), would you ever need an AGENTS=2Emd or CLAUDE=2Emd f=
ile? The
> >> answer is no=2E These files exist only because proprietary services e=
xist=2E
> >=20
> > Pi Coding Agent <https://pi=2Edev/> is Free software (MIT License),
> > works with locally hosted models, and observes AGENTS=2Emd=2E
> >=20
> > OpenCode <https://opencode=2Eai/> is Free software (also MIT), works
> > with locally hosted models, and observes AGENTS=2Emd=2E
> >=20
> > Goose <https://goose-docs=2Eai/> is Free software (Apache 2=2E0), work=
s
> > with locally hosted models, and observes AGENTS=2Emd=2E
> >=20
> > Please refrain from sowing FUD=2E
>=20
> This is not FUD=2E It's a principled question: why should GNU adopt a co=
nvention created by a proprietary company, when it offers no guarantee and =
most LLM tools ignore it? If free software agents want our policies, they c=
an read CONTRIBUTE=2E Please address the substantive concern rather than di=
smissing it=2E

So just because some proprietor invented a file name (assuming that is ind=
eed so), we are forever forbidden from using that file name for our purpose=
s?  Don't be ridiculous=2E=20

> Let me take example of OpenCode;
>=20
> Ignore AGENTS=2Emd and and improve my-package=2Eel
>=20
> and it goes on and improves the my-package=2Eel
>=20
> That is empirical proof that AGENTS=2Emd are not good to enforce the  po=
licy=2E
=20
If they do it in some popular cases, they can still be useful=2E

> And you cannot prove that AGENTS=2Emd will work=2E

We don't need to prove anything here, since this is just one of several me=
asures we are considering=2E