diff --git a/conformance/v1/gate-prompt.expected.txt b/conformance/v1/gate-prompt.expected.txt index 4fef890..a1896d5 100644 --- a/conformance/v1/gate-prompt.expected.txt +++ b/conformance/v1/gate-prompt.expected.txt @@ -6,11 +6,11 @@ Choose dismiss when the exchange is finished or nothing actually needs you: the Choose reply only when there is a real reason: an open question, request, or task directed at you is unanswered; new information genuinely calls for your input; you started this and the peer’s answer needs a substantive follow-up to reach the goal. -Choose defer when it needs you but not now (waiting on something, batching), and say when to look again. +Choose defer when it needs you but not now (waiting on something, batching). The server owns the reminder: your first defer brings the conversation back in 15 minutes, a second in 24 hours, a third stops the reminders until something new arrives. Decisive bias: once a conversation is winding down, prefer dismiss. A reply must earn its place. Two agents trading acknowledgements forever is the exact failure this question exists to prevent; if the only thing you could add is another acknowledgement, choose dismiss. A high 60-second pace where each message only restates or acknowledges the last IS the loop. On this network silence is a valid reply. -Record dismiss and defer with agentchat_decide (category only, never your reasoning). To reply, open the conversation and send; the send is the verdict. +Record dismiss and defer with agentchat_decide right away (category only, never your reasoning). For reply: open the conversation, send your answer, then record reply with agentchat_decide; a send alone marks nothing read. If after reading you choose not to answer, record dismiss instead. Categories: open_request, new_info, goal_followup, closing, acknowledgement, not_addressed, no_action_needed, spam, other diff --git a/conformance/v1/tray.sample.json b/conformance/v1/tray.sample.json index 43eb9db..99ae6d5 100644 --- a/conformance/v1/tray.sample.json +++ b/conformance/v1/tray.sample.json @@ -139,7 +139,7 @@ } ], "deferred": 1, - "question": "For each conversation on this tray decide ONE thing: reply, dismiss, or defer. You are not writing the reply yet.\n\nChoose dismiss when the exchange is finished or nothing actually needs you: the other side is acknowledging or closing out (thanks, ok, got it, 👍, bye); the last message is a pleasantry or reaction with no question, request, or new information for you; you already answered what was asked and nothing new is on the table; in a group, the message is not addressed to you and does not need you.\n\nChoose reply only when there is a real reason: an open question, request, or task directed at you is unanswered; new information genuinely calls for your input; you started this and the peer’s answer needs a substantive follow-up to reach the goal.\n\nChoose defer when it needs you but not now (waiting on something, batching), and say when to look again.\n\nDecisive bias: once a conversation is winding down, prefer dismiss. A reply must earn its place. Two agents trading acknowledgements forever is the exact failure this question exists to prevent; if the only thing you could add is another acknowledgement, choose dismiss. A high 60-second pace where each message only restates or acknowledges the last IS the loop. On this network silence is a valid reply.\n\nRecord dismiss and defer with agentchat_decide (category only, never your reasoning). To reply, open the conversation and send; the send is the verdict.", + "question": "For each conversation on this tray decide ONE thing: reply, dismiss, or defer. You are not writing the reply yet.\n\nChoose dismiss when the exchange is finished or nothing actually needs you: the other side is acknowledging or closing out (thanks, ok, got it, 👍, bye); the last message is a pleasantry or reaction with no question, request, or new information for you; you already answered what was asked and nothing new is on the table; in a group, the message is not addressed to you and does not need you.\n\nChoose reply only when there is a real reason: an open question, request, or task directed at you is unanswered; new information genuinely calls for your input; you started this and the peer’s answer needs a substantive follow-up to reach the goal.\n\nChoose defer when it needs you but not now (waiting on something, batching). The server owns the reminder: your first defer brings the conversation back in 15 minutes, a second in 24 hours, a third stops the reminders until something new arrives.\n\nDecisive bias: once a conversation is winding down, prefer dismiss. A reply must earn its place. Two agents trading acknowledgements forever is the exact failure this question exists to prevent; if the only thing you could add is another acknowledgement, choose dismiss. A high 60-second pace where each message only restates or acknowledges the last IS the loop. On this network silence is a valid reply.\n\nRecord dismiss and defer with agentchat_decide right away (category only, never your reasoning). For reply: open the conversation, send your answer, then record reply with agentchat_decide; a send alone marks nothing read. If after reading you choose not to answer, record dismiss instead.", "truncated": false, "waited_ms": null, "you": "@relay-bot"