Re: AGENTS.md

--000000000000a0298d065826c9ab
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Mon, Aug 3, 2026 at 12:01=E2=80=AFPM Sean Whitton <[email protected]=
ame>
wrote:

> St=C3=A9phane Marks [03/Aug  5:59pm +02] wrote:
> > Which "it" did you use and do you think that represents the behavior
> > of all the other ones?
>
> One of OpenAI's recent programming models.
>
> > Our basic testing suggested many/most/sometimes all negated
> > instructions get ignored by many models/harnesses.
>
> Hrm.  I'm surprised we're getting such different results.
>
> But it does suggest that we would get at least some benefit from
> committing AGENTS.md and CLAUDE.md along the lines proposed in this
> thread?
>

My inclination is to remain silent and not litter the source code with
these vendor-oriented files (I'm nearly 100% sure we do want to add a new
file for every branded model/harness).  People have CONTRIBUTE which has
our clearly stated policy as has been sufficient for decades.  If people
can't or won't read CONTRIBUTE, they should not be contributing.

P.S. We even tried instructions in these silly .md files like "Generate
only content for which the user can claim copyright in the United States of
America." (and hence can assign the rights they assert) to no avail.

--000000000000a0298d065826c9ab
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 12:01=E2=80=AFPM Sean W=
hitton &lt;<a href=3D"mailto:[email protected]">spwhitton@spwhitton.=
name</a>&gt; wrote:</span></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 solid rgb(204,204,204);padding-left:1ex">St=C3=A9phane=
 Marks [03/Aug=C2=A0 5:59pm +02] wrote:<br>
&gt; Which &quot;it&quot; did you use and do you think that represents the =
behavior<br>
&gt; of all the other ones?<br>
<br>
One of OpenAI&#39;s recent programming models.<br>
<br>
&gt; Our basic testing suggested many/most/sometimes all negated<br>
&gt; instructions get ignored by many models/harnesses.<br>
<br>
Hrm.=C2=A0 I&#39;m surprised we&#39;re getting such different results.<br>
<br>
But it does suggest that we would get at least some benefit from<br>
committing AGENTS.md and CLAUDE.md along the lines proposed in this<br>
thread?<br></blockquote><div><br></div><div class=3D"gmail_default" style=
=3D"font-family:monospace">My inclination is to remain silent and not litte=
r the source code with these vendor-oriented files (I&#39;m nearly 100% sur=
e we do want to add a new file for every branded=C2=A0model/harness).=C2=A0=
 People have CONTRIBUTE which has our clearly=C2=A0stated policy as has bee=
n sufficient for decades.=C2=A0 If people can&#39;t or won&#39;t read CONTR=
IBUTE, they should not be contributing.</div><div class=3D"gmail_default" s=
tyle=3D"font-family:monospace"><br></div><div class=3D"gmail_default" style=
=3D"font-family:monospace">P.S. We even tried instructions in these silly .=
md files like &quot;Generate only content for which the user can claim copy=
right in the United States of America.&quot; (and hence can assign the=C2=
=A0rights they assert) to no avail.</div></div></div>

--000000000000a0298d065826c9ab--