Re: AGENTS.md

--0000000000005f31bb06582417da
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Mon, Aug 3, 2026 at 8:47=E2=80=AFAM Eli Zaretskii <[email protected]> wrote:

> > From: Sean Whitton <[email protected]>
> > Cc: [email protected], [email protected]
> > Date: Mon, 03 Aug 2026 13:41:10 +0100
> >
> > Eli Zaretskii [03/Aug  3:32pm +03] 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 this
> > >> > prevent what we want to avoid with likelihood that is close enough
> to
> > >> > 100% for us to consider it the "solution"?  Because if not, we'd
> need
> > >> > this _and_ the hooks as Plan B.
> > >>
> > >> I would be very surprised if an LLM just ignored the AGENTS.md, unle=
ss
> > >> the user specifically told them to do that.  So I would say it's
> pretty
> > >> close to 100%.
> > >
> > > I'm not worried by the file being ignored.  But the text is rather
> > > general, so my bother is whether "instead of generating any code" is
> > > enough to completely prevent code generation, especially if the user
> > > asks for it, explicitly or implicitly.
> >
> > We can say it repeatedly in more words to make it more salient.
> >
> > I could also run some tests on it at work.
>
> That would help, but I was actually asking about some evidence that
> someone already provided.  Perhaps some investigation into the effects
> of this is available somewhere?
>

I tried to find evidence for other work and found none.  There seems to be
no anti-LLM file like NO-AGENTS.md and no text I could find that has
measurable anti-llm efficacy when present in AGENTS.md.

--0000000000005f31bb06582417da
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div dir=3D"ltr"><div class=3D"gmail_default" style=3D"fon=
t-family:monospace"><span style=3D"font-family:Arial,Helvetica,sans-serif;b=
ackground-color:transparent">On Mon, Aug 3, 2026 at 8:47=E2=80=AFAM Eli Zar=
etskii &lt;<a href=3D"mailto:[email protected]">[email protected]</a>&gt; wrote:</spa=
n></div></div><div class=3D"gmail_quote gmail_quote_container"><blockquote =
class=3D"gmail_quote" style=3D"margin:0px 0px 0px 0.8ex;border-left:1px sol=
id rgb(204,204,204);padding-left:1ex">&gt; From: Sean Whitton &lt;<a href=
=3D"mailto:[email protected]" target=3D"_blank">spwhitton@spwhitton.=
name</a>&gt;<br>
&gt; Cc: <a href=3D"mailto:[email protected]" target=3D"_blank">jporter=
[email protected]</a>, <a href=3D"mailto:[email protected]" target=3D"_blank=
">[email protected]</a><br>
&gt; Date: Mon, 03 Aug 2026 13:41:10 +0100<br>
&gt; <br>
&gt; Eli Zaretskii [03/Aug=C2=A0 3:32pm +03] wrote:<br>
&gt; &gt;&gt; From: Sean Whitton &lt;<a href=3D"mailto:spwhitton@spwhitton.=
name" target=3D"_blank">[email protected]</a>&gt;<br>
&gt; &gt;&gt; Cc: <a href=3D"mailto:[email protected]" target=3D"_blank=
">[email protected]</a>, <a href=3D"mailto:[email protected]" target=
=3D"_blank">[email protected]</a><br>
&gt; &gt;&gt; Date: Mon, 03 Aug 2026 13:04:44 +0100<br>
&gt; &gt;&gt;<br>
&gt; &gt;&gt; Eli Zaretskii [03/Aug=C2=A0 2:40pm +03] wrote:<br>
&gt; &gt;&gt; &gt; Do we have any idea how effective is this in practice?=
=C2=A0 Does this<br>
&gt; &gt;&gt; &gt; prevent what we want to avoid with likelihood that is cl=
ose enough to<br>
&gt; &gt;&gt; &gt; 100% for us to consider it the &quot;solution&quot;?=C2=
=A0 Because if not, we&#39;d need<br>
&gt; &gt;&gt; &gt; this _and_ the hooks as Plan B.<br>
&gt; &gt;&gt;<br>
&gt; &gt;&gt; I would be very surprised if an LLM just ignored the AGENTS.m=
d, unless<br>
&gt; &gt;&gt; the user specifically told them to do that.=C2=A0 So I would =
say it&#39;s pretty<br>
&gt; &gt;&gt; close to 100%.<br>
&gt; &gt;<br>
&gt; &gt; I&#39;m not worried by the file being ignored.=C2=A0 But the text=
 is rather<br>
&gt; &gt; general, so my bother is whether &quot;instead of generating any =
code&quot; is<br>
&gt; &gt; enough to completely prevent code generation, especially if the u=
ser<br>
&gt; &gt; asks for it, explicitly or implicitly.<br>
&gt; <br>
&gt; We can say it repeatedly in more words to make it more salient.<br>
&gt; <br>
&gt; I could also run some tests on it at work.<br>
<br>
That would help, but I was actually asking about some evidence that<br>
someone already provided.=C2=A0 Perhaps some investigation into the effects=
<br>
of this is available somewhere?<br></blockquote><div><br></div><div class=
=3D"gmail_default" style=3D"font-family:monospace">I tried to find evidence=
 for other work and found none.=C2=A0 There seems to be no anti-LLM file li=
ke NO-AGENTS.md and no text I could find that has measurable anti-llm effic=
acy when present in AGENTS.md.</div></div></div>

--0000000000005f31bb06582417da--