Re: AGENTS.md
Jean Louis <[email protected]> Fri, 07 Aug 2026 21:19:38 +0300
| Newsgroups | gmane.emacs.devel |
|---|---|
| Organization | GNU Support |
| Message-ID | <[email protected]> |
On 2026-08-07 20:08, Dr. Arne Babenhauserheide wrote: > For AGENTS.md in Emacs, the support by harnesses in Emacs should be the > most relevant, otherwise that would not help people who use Emacs to > work on Emacs. Me sorry doctor, just to try to clarify things, AGENTS.md is not related to harnessing, it is simply extra context to the prompt. After prompt comes the response from the LLM. And software agent can decide what to actually do based on the response. If the response says "Let us delete full directory and make new one", the software may decide to ask user about it. That is how I see where the harnessing stage comes. Entering anything into the prompt is not for me harnessing, yet some may disagree, because that decision has already been made, so there is nothing to harness. > I asked to know whether it could be useful to write to the devs of > those > packages whether they could add support for it, but I didn’t expect > that > that hadn’t been checked at all. I find it fundamentally misleading to ask packages to do that as that would mean increasing cost in their spending when accessing proprietary Service as Software Substitute Large Language Models. The so called "standard" was invented by those making money on it. It serves them. I believe that some of those packages have concept of memories which is basically just extra text inserted into the prompt to be given to the LLM. Personally I am using LLM memories which is just concatenated to the prompt to LLM. -- Jean Louis