Re: AGENTS.md
Jorge Javier Araya Navarro <[email protected]> Tue, 04 Aug 2026 00:47:56 +0000
| Newsgroups | gmane.emacs.devel |
|---|---|
| Message-ID | <[email protected]> |
I read somewhere on =F0=9D=95=8F that the creator of HashiCorp have an=20 AGENTS.md file in one of his FLOSS projects, essentially it tells any LLM=20 that it should write code comments with something along the lines of "I'm=20 a loser and I don't review generated code before sending a patch".=20 That way, he can quickly identify and veto low quality patches sent to him.[1] So, yeah, there is no chance of any respectable harness to ignore=20 an AGENTS.md and its instructions. If people want to avoid code=20 generation being put on GNU Emacs, having an AGENTS.md with such instructions=20 is the best way to go. [1]: https://www.youtube.com/watch?v=3Dbh6S4N8TnYQ On lun, ago 03 2026, Eli Zaretskii wrote: >> From: Sean Whitton <[email protected]> >> Cc: [email protected], [email protected] >> Date: Mon, 03 Aug 2026 13:04:44 +0100 >> >> Eli Zaretskii [03/Aug 2:40pm +03] wrote: >> > Do we have any idea how effective is this in practice? Does=20 >> > this >> > prevent what we want to avoid with likelihood that is close=20 >> > enough to >> > 100% for us to consider it the "solution"? Because if not,=20 >> > we'd need >> > this _and_ the hooks as Plan B. >> >> I would be very surprised if an LLM just ignored the AGENTS.md,=20 >> unless >> the user specifically told them to do that. So I would say=20 >> it's pretty >> close to 100%. > > I'm not worried by the file being ignored. But the text is=20 > rather > general, so my bother is whether "instead of generating any=20 > code" is > enough to completely prevent code generation, especially if the=20 > user > asks for it, explicitly or implicitly. --=20 Jorge Araya Contacto: Telegram: t.me/shackra =C2=B7 Signal: Shackra.28