Skip to content

A local-first multi-agent app development platform, built on PenguinHarness

The messaging delivery options explain themselves once, and briefly (#520)

Yaowei Zheng9dadf40e557e
diff --git a/changelog/unreleased/2026-08-27-messaging-final-reply-only.md b/changelog/unreleased/2026-08-27-messaging-final-reply-only.md
index a0344e02..03312266 100644
--- a/changelog/unreleased/2026-08-27-messaging-final-reply-only.md
+++ b/changelog/unreleased/2026-08-27-messaging-final-reply-only.md
@@ -50,9 +50,11 @@ the previous behaviour.
spends that window on the run: a run that outlives it delivers nothing, where the
every-message relay would have sent whatever completed inside it. That is a different
outcome rather than a shade of the same one — silence on the other two channels, lost
- output here — so on QQ it is said twice rather than left to these notes: appended to the
- option's explanation, for the reader deciding, and stood under the row as a warning strip
- while the switch is actually on. `linePerMessage` needs no such warning: QQ clamps the split
+ output here — so on QQ it is appended to the option's own explanation rather than left to
+ these notes. Only there: a standing strip under the row would be one more line permanently
+ explaining a channel to someone who has already read it once, where what belongs on screen is
+ the switch and what belongs behind the "?" is why you might not want it.
+ `linePerMessage` needs no such warning: QQ clamps the split
to its own budget and the reply still arrives.
- The `approval_request` notice is not an assistant message and is unaffected — a run blocked on
approval is exactly when the chat has to hear something. The other fixed notices and the test
diff --git a/changelog/unreleased/2026-08-27-messaging-final-reply-only.zh.md b/changelog/unreleased/2026-08-27-messaging-final-reply-only.zh.md
index 4d611e3d..7e8c408f 100644
--- a/changelog/unreleased/2026-08-27-messaging-final-reply-only.zh.md
+++ b/changelog/unreleased/2026-08-27-messaging-final-reply-only.zh.md
@@ -39,8 +39,9 @@
只在为其提供额度的入站消息之后几分钟内被接受,而把回复扣到运行结束才发,等于把这个窗口花在了运
行本身上:运行时长超过窗口时什么都送不出去,而逐条转发至少能把窗口之内完成的部分发出去。
这与另外两个渠道是不同的结果,而不是同一件事的程度差别——那边的代价是安静,这边是输出丢失——所以在
- QQ 上这句提醒说了两遍,而不是只留在这份记录里:既附在该选项的说明中供正在做决定的人读到,也在开关
- 真正打开时以警示条的形式立在该行下面。「每行一条消息」不需要类似的提醒:QQ 会把拆分收敛到自己的额
+ QQ 上这句提醒附在该选项自己的说明里,而不是只留在这份记录里;也只在那里:在该行下面常驻一条警示条,
+ 等于在这张表单上再多一行永久向已经读过一遍的人解释某个渠道,而屏幕上该留的是开关本身,「?」后面该留
+ 的是「你为什么可能不想要它」。「每行一条消息」不需要类似的提醒:QQ 会把拆分收敛到自己的额
度内,回复仍然会到达。
- `approval_request` 提醒不属于助手消息,不受影响——运行卡在审批上,恰恰是聊天必须听到动静的时
候。其余固定提示与测试消息同理不受影响;压缩输出仍在这一切之前就被跳过;中途接入的连接依旧不
diff --git a/packages/web/src/features/messaging/messaging-binding-editor.tsx b/packages/web/src/features/messaging/messaging-binding-editor.tsx
index 91157a3e..b3c49bb0 100644
--- a/packages/web/src/features/messaging/messaging-binding-editor.tsx
+++ b/packages/web/src/features/messaging/messaging-binding-editor.tsx
@@ -50,7 +50,7 @@ import * as api from "../../api/endpoints";
import { S } from "../../lib/strings";
import { apiErrorText } from "../../lib/api-error";
import { formatDateTime } from "../../lib/format";
-import { toneInk, toneStrip, type Tone } from "../../lib/tone";
+import { toneInk, type Tone } from "../../lib/tone";
import { Button } from "../../components/ui/button";
import { FieldLabel } from "../../components/ui/field";
import { HelpFold } from "../../components/ui/help-fold";
@@ -943,10 +943,11 @@ export function MessagingBindingBody({ b }: { b: MessagingBindingEditorState })
is a passive reply anchored to an inbound message, and that anchor expires, so
holding the reply to the run's end loses a long run's output entirely instead of
merely delaying it. That is a different outcome, not a nuance of the same one, so it
- is said twice — appended to the explanation, for the reader deciding, and stood on
- screen under the row while the switch is on, for the one who already has. Left to
- the "?" alone, the user learns it from an empty chat. `linePerMessage` needs nothing
- similar: QQ clamps the split to its own budget and the reply still arrives. */}
+ is appended to this option's explanation — and only there. A standing strip under the
+ row would be the third thing on this form permanently explaining a channel to someone
+ who has already read it once; what belongs on screen is the switch, and what belongs
+ behind the "?" is why you might not want it. `linePerMessage` needs nothing similar:
+ QQ clamps the split to its own budget and the reply still arrives. */}
<DeliveryOptionRow
label={S.messaging.finalReplyOnly}
help={
@@ -962,16 +963,6 @@ export function MessagingBindingBody({ b }: { b: MessagingBindingEditorState })
checked={delivery.finalReplyOnly}
onChange={(v) => patchDelivery({ finalReplyOnly: v })}
/>
- {/* Rendered by the state it describes, so it is on screen exactly while the loss it
- names is possible and nowhere else. */}
- {channel === "qq" && delivery.finalReplyOnly && (
- <p
- role="alert"
- className={`rounded-md border px-2.5 py-1.5 text-xs ${toneStrip.attention}`}
- >
- {S.messaging.finalReplyOnlyQQWarning}
- </p>
- )}
<DeliveryOptionRow
label={S.messaging.linePerMessage}
help={S.messaging.linePerMessageHelp}
diff --git a/packages/web/src/lib/strings-en.ts b/packages/web/src/lib/strings-en.ts
index 64dffccd..61e313a2 100644
--- a/packages/web/src/lib/strings-en.ts
+++ b/packages/web/src/lib/strings-en.ts
@@ -1985,23 +1985,19 @@ Scenarios:
linePerMessage: "One message per line",
/** Its disclosure, beside the label: what the option does to a reply, and its two edges. */
linePerMessageHelp:
- "Each non-blank line of a reply is sent as its own message and blank lines are dropped, so an answer written as several spoken lines arrives as several messages. Past a per-reply limit the remaining lines are combined into one last message rather than dropped, and a line longer than the channel allows is still split. The messages go out about a second apart, the pace a chat platform allows.",
+ "Each non-blank line of a reply is sent as its own message, so an answer written as several spoken lines arrives as several messages. Past a per-reply limit the remaining lines are combined into one last message rather than dropped.",
/** The saved delivery option: hold a run's working notes, send its last reply only. */
finalReplyOnly: "Final reply only",
/** Its disclosure, beside the label: what the option changes, and what it costs. */
finalReplyOnlyHelp:
- "Only the last thing the assistant says in a run is sent, when the run ends. The notes it writes between tool calls stay in the web app instead of arriving as their own messages, so the chat receives the answer rather than the work — at the cost of hearing nothing while a long run is under way. The approval reminder is not a reply and still arrives immediately. With one message per line also on, that final reply is the one that is split.",
+ "Sends only the last thing the assistant says in a run, when the run ends; the notes it writes between tool calls stay in the web app. The cost is hearing nothing while a long run is under way. The approval reminder is not a reply and still arrives immediately.",
/**
- * QQ only, and said twice there: appended to the option's explanation for the reader
- * deciding, and stood under the row as a warning strip while the switch is on. Not a
- * nuance of the same trade but a different outcome: QQ can only send a passive reply
- * anchored to an inbound message, and that anchor expires. On the other two channels the
- * cost of holding the reply is silence; here a run that outlasts the window delivers
- * nothing at all — too much to rest on a "?" the user need never open, which would leave
- * them to discover it from an empty chat.
+ * Appended to the option's explanation on QQ only. Not a nuance of the same trade but a
+ * different outcome — silence on the other two channels, lost output here — which the
+ * channel-neutral sentence above would leave the user to discover from an empty chat.
*/
finalReplyOnlyQQWarning:
- "On QQ this needs a second thought: QQ only accepts a passive reply anchored to an inbound message, and that anchor expires after about five minutes. Holding the reply until the run ends means a run that outlasts the window delivers nothing at all — with this off, at least whatever finished inside the window arrives.",
+ "QQ can only answer an inbound message, and that anchor expires after about five minutes: with this on, a run taking longer delivers nothing at all.",
/** Enabled-row indicator's tooltip / sr text (the small per-channel glyph on the session row). */
enabledIndicator: {
feishu: "Feishu connection enabled",
diff --git a/packages/web/src/lib/strings.ts b/packages/web/src/lib/strings.ts
index 0a7a0612..42e2b2f8 100644
--- a/packages/web/src/lib/strings.ts
+++ b/packages/web/src/lib/strings.ts
@@ -1939,23 +1939,19 @@ Benchmark:
linePerMessage: "每行一条消息",
/** Its disclosure, beside the label: what the option does to a reply, and its two edges. */
linePerMessageHelp:
- "开启后,回复中的每个非空行各自作为一条消息发出,空行忽略——写成多句台词的回复就按台词逐条到达。行数超过每条回复的上限时,余下的行合并为最后一条,内容不会丢失;过长的单行仍按渠道长度上限分段。多条消息之间相隔约一秒发出,这是聊天平台允许的速率。",
+ "回复的每个非空行各发一条消息,写成多句台词的回复就按台词逐条到达。超出每条回复的消息上限时,余下的行合并为最后一条,内容不会丢失。",
/** The saved delivery option: hold a run's working notes, send its last reply only. */
finalReplyOnly: "只发送最终回复",
/** Its disclosure, beside the label: what the option changes, and what it costs. */
finalReplyOnlyHelp:
- "开启后,一次运行中只有助手最后说的那段话会发出,且在运行结束时发送。它在工具调用之间写下的过程记录留在网页端,不再各自作为一条消息到达——聊天里收到的是结论而非过程,代价是长时间运行期间聊天中一片安静。审批提醒不属于回复,仍会立即到达。若同时开启「每行一条消息」,被拆分的就是这条最终回复。",
+ "只发送一次运行中助手最后说的那段话,在运行结束时发出;工具调用之间的过程记录留在网页端。代价是长时间运行期间聊天里一片安静。审批提醒不属于回复,仍会立即到达。",
/**
- * QQ only, and said twice there: appended to the option's explanation for the reader
- * deciding, and stood under the row as a warning strip while the switch is on. Not a
- * nuance of the same trade but a different outcome: QQ can only send a passive reply
- * anchored to an inbound message, and that anchor expires. On the other two channels the
- * cost of holding the reply is silence; here a run that outlasts the window delivers
- * nothing at all — too much to rest on a "?" the user need never open, which would leave
- * them to discover it from an empty chat.
+ * Appended to the option's explanation on QQ only. Not a nuance of the same trade but a
+ * different outcome — silence on the other two channels, lost output here — which the
+ * channel-neutral sentence above would leave the user to discover from an empty chat.
*/
finalReplyOnlyQQWarning:
- "在 QQ 上要多想一步:QQ 只接受锚定在某条入站消息上的被动回复,且该锚点约五分钟后失效。把回复押到运行结束,意味着运行超过这个窗口时一条也发不出去——关闭本开关时,至少窗口内完成的那几条能送达。",
+ "QQ 只能回复入站消息,锚点约五分钟失效:开启后,运行超过五分钟就一条也发不出。",
/** Enabled-row indicator's tooltip / sr text (the small per-channel glyph on the session row). */
enabledIndicator: {
feishu: "飞书连接已启用",