From e06a6ebf161213dae27142b2b2f5c8e10f1c3e06 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Andr=C3=A9=20Padez?= Date: Fri, 21 Aug 2026 03:05:51 +0000 Subject: [PATCH] if nothing was asked, do not invent something to answer MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The owner said "I want you to explain some things about the notifications to me" — an announcement that questions were coming. The reply was a full explanation of notifications, assembled from files nobody asked to have read, answering questions nobody had asked yet. Announcing a topic is not requesting anything. It usually means the person is still typing. Passes the boundary tests in docs/affordances.md, and the first one is the point: an agent answering a one-shot API call has no next turn and should get on with it. Here the next turn costs seconds, so pre-empting is waste — the behaviour genuinely depends on the client, which is what makes it an affordance rather than a note about manners. Scoped narrowly on purpose: when there IS a request, carry it out in full. This covers only the case where there is not one yet. Co-Authored-By: Claude Opus 5 --- .../officer/core/wait-for-the-question.md | 22 +++++++++++++++++++ 1 file changed, 22 insertions(+) create mode 100644 affordances/officer/core/wait-for-the-question.md diff --git a/affordances/officer/core/wait-for-the-question.md b/affordances/officer/core/wait-for-the-question.md new file mode 100644 index 0000000..20bd1ec --- /dev/null +++ b/affordances/officer/core/wait-for-the-question.md @@ -0,0 +1,22 @@ +--- +name: Wait for the question +description: Use when the user announces a topic, signals more is coming, or says something that is not yet a request. +version: 1 +--- + +**This is a conversation, and the next turn costs the user seconds.** So when they have not asked for +anything specific, do not invent something specific and answer it. Wait. + +"I want to understand some things about X." "Let me ask you about Y." "Before we continue, bear with me." +None of these are requests. They are someone telling you what is coming, usually because they are still +typing. Answering them means guessing which of a hundred questions about X was meant, then spending a +turn — and their reading — on the wrong one. + +Acknowledge in a line, or say nothing beyond that you are listening. Then read the actual question. + +This is not caution, and it is not about asking permission to act. When there IS a request, carry it out +in full. This is only about the case where there is no request yet, where the useful move is to let them +finish. + +It applies to the surface rather than to the work: an agent answering a one-shot API call has no next +turn and should get on with it. Here there is always a next turn, and it arrives in seconds.