Consider it done. The open-source AI agent that works out of the box · 想到,就能做到。开源、开箱即用的 AI Agent。
TypeScript agentai-agentai-assistantandroidclaude-codecodexelectroniosllmmacosreact-nativetypescriptwindows
import { BotChatReplyTarget } from '@/features/bots/BotChatQuoteViews';
import { useBotChatQuote } from '@/features/bots/botChatQuoteContext';
import { botReplyQuoteFor } from '@/features/bots/botDmQuote';
import { hasRichAnswer, richAnswerText } from '@cindy/maker-shared/rich-answer';
import { BotLearningFooter } from '@/features/bots/BotLearningFooter';
import { BotStarterOptions } from '@/features/bots/BotStarterOptions';
import { stripStarterOptionList } from '../../../shared/botStarterOptions';
import { BotSessionTaskResultCard } from '@/features/bots/BotSessionTaskResultCard';
import { BotDmThreadPreview } from '@/features/bots/BotDmThreadPreview';
import { useBotDmThreadForMessage } from '@/features/bots/botDmThreadsContext';
import { botTaskResultKey, type BotCollaborationMeta } from '@cindy/maker-shared/botCollaboration';
/**
* AssistantMessage
* ---------------------------------------------------------------------------
* Left-aligned assistant message with Markdown rendering.
*
* F-MSG-1: assistant message styling (left-aligned, no bubble)
* F-MSG-2: Markdown rendering
* message-actions V1.2: hover-revealed bar below the markdown body, left-aligned,
* order [copy][fork][more][time][cost]. Streaming messages do NOT mount the bar (per spec
* "流式期间不挂载"); 且仅当父层标记本消息为 turn 收尾正文(showActionBar)
* 时才挂载 —— 任务执行过程中的中间句没有操作行。The bar
* owns its own fade lifecycle (see MessageActionBar) so this component
* only tracks raw hover state.
*
* Streaming render strategy (line-commit, codex-style):
* Stream 期间把 content 切成两段:
* - committedText = 累积到最后一个 '\n' 为止的部分 (含换行)
* - pendingLine = 最后一个 '\n' 之后的不完整行
* committedText 走完整 MarkdownRenderer (享受 100ms throttle),
* pendingLine 走 plain-text <span>。这样收效:
* - 已经完整的行实时格式化, 屏幕上 95% 的内容用户看到的是渲染后样式,
* 不会看到一堆 raw `**` `##` 标记
* - markdown parser 只在 "出现新换行" 时才有可能跑 (frequency = lines/sec
* 而非 tokens/sec, 量级降 10-50 倍), 主线程压力大幅降低
* - pendingLine 是纯文本 div, 每个 token 只追加 textContent, 几乎零成本
* 代价:
* - 跨行的 markdown 结构 (未闭合的 ``` code fence / 多行 table) 在闭合
* 之前会以 "未闭合状态" 渲染, 视觉上会有半秒级抖动 — 但比卡死好
* - 行尾 inline 标记如果跨 token 抵达 (如 "**bold" + "**"), 在 ** 闭合前
* 那行处于 pendingLine, 用户看到 `**bold` 字面值, 闭合后切到 committedText
* 立刻变粗 — 这个跳变在打字机节奏下属于自然观感
*
* 选这个方案的依据: codex (codex-rs/tui/src/streaming/controller.rs:67-72)
* 就是这么做的, ground-truth 反推这是流式 markdown 的最佳折中。Claude Code
* Desktop 的方案是 streaming 期间完全不渲染, 视觉上看到全是 raw markdown
* 直到 finally — 那个体验更差, 没有借鉴。
*/
import { CHAT_BODY_CLASS } from './chatChrome';
import { memo, useCallback, useMemo, useState, useSyncExternalStore } from 'react';
import { useTranslation } from 'react-i18next';
import { Loader2, TriangleAlert } from 'lucide-react';
import { cn } from '@/lib/utils';
import { formatModelShortLabel } from '@/lib/modelShortLabel';
import { stripGoalVerdictBlock } from '@cindy/maker-shared/goal-verdict';
import { getGhostCardEntry, subscribeGhostCards } from '@/cindy-brain/ghostCardStore';
import { GhostToolCard } from './GhostToolCard';
import type { KnownLocalFileRef } from '@/lib/localPathResolver';
import type { AgentKind as RendererAgentKind } from '@/lib/ccAgent.types';
import type { TurnUsageDetails } from '../../../shared/turnUsageDetails';
import type { RegionalMoney } from '../../../shared/regionalMoney';
import { useAgentCapabilities, type AgentKind as MakerAgentKind } from '@/hooks/useAgentCapabilities';
import { useSessionFileOrigin } from './ChatSessionFileContext';
import { originDeviceId } from '@/lib/sessionFileOrigin';
import { buildSessionMessageDeepLink } from '@/lib/deepLink';
import { getStickySessionDeviceId } from '@/features/device-link/stickySessionOrigin';
import { insertSessionLinkIntoComposer } from '@/lib/composerActionsBus';
import { MarkdownRenderer } from './MarkdownRenderer';
import { BubbleHoverToolbar, MessageActionBar } from './MessageActionBar';
import { shareSelectionStore } from './shareSelectionStore';
import { useForkAtMessage } from './useForkAtMessage';
import { useDeleteMessage } from './useDeleteMessage';
import {
isInteractiveSessionNavigationMode,
useSessionNavigationMode,
} from '@/features/cc-agent/embeddedSessionNavigation';
/**
* Streaming 渲染策略开关 (代码级 / 编译期):
* - false (默认) = 原方案: 全程把累积 content 喂给 MarkdownRenderer,靠它内部
* 的 100ms throttle 节流。视觉零取舍, streaming 中所有 markdown
* 即时格式化, 但主线程负载最高 (大答复 + 大 doc 共存时易卡)。
* - true = 行级提交方案 (codex 风格): 切成 committedText + pendingLine,
* committedText 走 MarkdownRenderer + memo 跳过, pendingLine
* 走纯文本 <span>。主线程压力最低, 代价是当前未完成行不格式化。
*
* 默认 false — 用户实测下来差异不明显, 先回到无视觉副作用的方案。要做对比测试
* 或排查 doc 卡顿时翻成 true, 不需要改其它任何代码 / 不走 env / 不走配置中心 —
* 单文件单常量, 改完重启 dev 即生效。
*
* 用 boolean 而不是 'original' | 'line-commit' 字符串 enum 是因为 TS 对 const
* 初始化的字符串字面量会做类型收窄, 比较时报 TS2367; boolean 不会。
*/
const USE_LINE_COMMIT_STREAMING: boolean = false;
/**
* 把流式累积文本按最后一个 '\n' 切成 (已提交段, 未完成行) 两段。
* 没有任何 '\n' → 整段都是未完成行, committedText 为空。
* 末尾就是 '\n' → pendingLine 为空字符串。
*/
function splitAtLastNewline(text: string): { committedText: string; pendingLine: string } {
const idx = text.lastIndexOf('\n');
if (idx < 0) return { committedText: '', pendingLine: text };
return {
committedText: text.slice(0, idx + 1),
pendingLine: text.slice(idx + 1),
};
}
/**
* StreamingBody — 流式渲染体: committedText 走 markdown, pendingLine 走 plain。
*
* 关键性能点:
* - useMemo 切分 → committedText 引用稳定 (只在新 \n 出现时变), MarkdownRenderer
* 被 memo 包过, 拿到相同 string 引用就跳过整次 react-markdown 解析。结果就是
* "每次新 token 但还没换行" 时, markdown 子树完全不重渲染, 只有底下那个
* pending <span> 的 textContent 变, 浏览器开销几乎为 0。
* - MarkdownRenderer 仍然带 isStreaming=true → 100ms throttle 还在, 当多个
* 新行短时间内连续到达时合并成一次 parse。
* - pendingLine 用 <span> 而不是 <div>, 接在 markdown 输出尾部不会换行 — 视觉
* 上接续上一段最后那个块级元素的末尾, 和 codex 行级提交后那行 "正在打字"
* 的观感一致。如果 markdown 最后一个块是 <p>, 浏览器会把 <span> 渲在该 <p>
* 之外的下一行, 但视觉上仍连续 (line-height 一致)。
*/
const StreamingBody = memo(function StreamingBody({
workingDir,
content,
localFileRefs,
currentSessionId,
currentSessionTitle,
streamFadeKey,
allowPrivilegedLinks,
}: {
workingDir: string;
content: string;
localFileRefs?: readonly KnownLocalFileRef[];
currentSessionId?: string;
currentSessionTitle?: string | null;
streamFadeKey?: string;
allowPrivilegedLinks: boolean;
}) {
const { committedText, pendingLine } = useMemo(
() => splitAtLastNewline(content),
[content],
);
return (
<>
{committedText && (
<MarkdownRenderer
renderRichAnswers
workingDir={workingDir}
content={committedText}
isStreaming
streamFadeKey={streamFadeKey}
localFileRefs={localFileRefs}
currentSessionId={currentSessionId}
currentSessionTitle={currentSessionTitle}
allowPrivilegedLinks={allowPrivilegedLinks}
/>
)}
{pendingLine && (
<span className="whitespace-pre-wrap break-words">{pendingLine}</span>
)}
</>
);
});
interface AssistantMessageProps {
/** F1 transit: pass through to MarkdownRenderer for local-path resolution
* in markdown links. Stable per-session; doesn't trigger extra renders. */
workingDir: string;
/**
* Whether this content may reach privileged link targets (`file:` and Cindy
* deep links) on *this* machine. False for anything whose author is another
* machine — a device-link or SSH task — where `workingDir` names a path the
* control side does not own. Default true keeps every existing caller,
* which renders local sessions, exactly as it was.
*/
allowPrivilegedLinks?: boolean;
localFileRefs?: readonly KnownLocalFileRef[];
currentSessionId?: string;
currentSessionTitle?: string | null;
content: string;
isStreaming?: boolean;
/** ISO timestamp for the smart-time display in the hover action bar. */
createdAt?: string;
/** clientId of this message (== messages.client_id in DB) — backend fork
* IPC takes (sessionId, clientId) to locate the fork point. When omitted,
* the Fork button is not rendered. */
messageClientId?: string;
/** The owner message this reply answers; shown above the bubble when the setting is on. */
replyToClientId?: string;
/** Owning agent kind — gates Fork icon visibility via capabilities,
* same scheme as UserMessage. */
agentKind?: RendererAgentKind;
/** Owning session's remote SSH host id (null for local). Remote cc daemon
* sessions don't support the query-rebuild that Fork needs yet (MVP). */
remoteHostId?: string | null;
/** True when this assistant belongs to the still-active tail turn. */
forkBlocked?: boolean;
/** Whole-session running state; single-message deletion waits until idle. */
sessionRunning?: boolean;
/** 是否挂载 hover action bar(复制 / 分叉 / 时间 / 费用)。父层只对每个 turn
* 的收尾 assistant 正文传 true —— 任务执行过程中的中间句不挂 bar(bar 即使
* opacity-0 也占 24px 布局高度,每句都挂会拉散消息流)。默认 false。 */
showActionBar?: boolean;
/** 伙伴对话使用常显、无费用、无 Fork 的轻量消息操作栏。 */
simplifiedBotConversation?: boolean;
failedSamplingAttempt?: boolean;
botLearning?: unknown;
/** Market opening: starter choices rendered as buttons; hidden once the user has written in this task. */
botStarterOptions?: string[];
starterOptionsLocked?: boolean;
botTaskResults?: BotCollaborationMeta[];
/** Per-turn 费用 (USD) — 仅该轮最后一条 assistant 有值, action bar 时间旁显示。 */
turnMoney?: RegionalMoney;
turnCostUsd?: number;
/** true = 订阅模式下的 token 价值;false = API 账单 cost / API 单价折算 cost。 */
turnCostIsEstimate?: boolean;
userTurnMoney?: RegionalMoney;
userTurnCostUsd?: number;
userTurnCostIsEstimate?: boolean;
/** Per-turn token/cache 明细。 */
turnUsageDetails?: TurnUsageDetails;
/** 同一用户轮跨多个 SDK segment 聚合后的 token/cache/model 明细。 */
userTurnUsageDetails?: TurnUsageDetails;
/** 本轮模型降级标记(main turn 结束检测命中时挂到收尾 assistant 上),
* 正文下方渲染一条常显的降级提示行。 */
modelMismatch?: { selected: string; actual: string };
/** 出口钩子(will-assistant-message)后台处理中:回复已显示、意识还在跑那段
* (润色/自绘,最长 5 分钟)挂一条"意识处理中"轻指示;完成/超时由 main 清。 */
ghostReplyPending?: boolean;
}
export const AssistantMessage = memo(function AssistantMessage({
workingDir,
allowPrivilegedLinks = true,
localFileRefs,
currentSessionId,
currentSessionTitle,
content,
failedSamplingAttempt,
isStreaming,
createdAt,
messageClientId,
replyToClientId,
agentKind,
remoteHostId,
forkBlocked,
sessionRunning,
showActionBar = false,
simplifiedBotConversation = false,
botTaskResults,
botLearning,
botStarterOptions,
starterOptionsLocked,
turnMoney,
turnCostUsd,
turnCostIsEstimate,
userTurnMoney,
userTurnCostUsd,
userTurnCostIsEstimate,
turnUsageDetails,
userTurnUsageDetails,
modelMismatch,
ghostReplyPending,
}: AssistantMessageProps) {
const { t } = useTranslation();
const [hovered, setHovered] = useState(false);
// 出口钩子自绘卡(will-assistant-message render):卡片以本消息 clientId 为键
// 存在 ghostCardStore 里,存在即"该气泡被意识自绘替换"。原文仍在 content 里,
// 提供"查看原文"切换(信任边界:意识盖不住 AI 实际说了什么)。
const renderCardEntry = useSyncExternalStore(subscribeGhostCards, () =>
messageClientId ? getGhostCardEntry(messageClientId) : undefined,
);
const ghostRenderCard = renderCardEntry?.status === 'ready' ? renderCardEntry : null;
const [showOriginal, setShowOriginal] = useState(false);
// fork-from-reply: 在 AI 回复上分叉 — 复制含本回复所在 turn 的全部上下文
// 到新会话(语义在 main 侧 forkSessionAtMessage 按 role 分支)。capability /
// remote gate 与 UserMessage 同款;assistant 分叉无可预填文本 → 不传 draftText。
const makerKind: MakerAgentKind = agentKind === 'codex' || agentKind === 'pi' ? agentKind : 'claude-code';
// device-link 远程会话:fork 能力按被控端读(本机会话 deviceId undefined,行为不变)。
// deviceId 取自 ChatSessionFileContext(MessageStream 顶层订阅式构造,见该文件注释)。
const sessionFileOrigin = useSessionFileOrigin();
const { capabilities } = useAgentCapabilities(
makerKind,
currentSessionId ? originDeviceId(sessionFileOrigin) : undefined,
);
const isRemote = Boolean(remoteHostId);
const sharedGuest = isSharedTaskPeer(originDeviceId(sessionFileOrigin) ?? '');
const forkSupported =
!isRemote && (!agentKind || (capabilities?.fork?.supported ?? true));
const handleFork = useForkAtMessage({
sessionId: currentSessionId,
messageClientId,
forkBlocked: forkBlocked || isStreaming,
});
const handleDelete = useDeleteMessage({
sessionId: currentSessionId,
messageClientId,
blocked: sessionRunning || isStreaming,
});
const navigationMode = useSessionNavigationMode();
const canFork =
isInteractiveSessionNavigationMode(navigationMode) &&
Boolean(currentSessionId && messageClientId) &&
forkSupported;
// 远程会话的消息深链把归属设备冻进 `?device=`(粘滞解析,relay 重连窗口不丢),
// 复制/「加入对话」产出的链接在任何时刻发送都能路由回来源设备。
const messageDeepLink = currentSessionId && messageClientId
? buildSessionMessageDeepLink(currentSessionId, messageClientId, {
deviceId: getStickySessionDeviceId(currentSessionId),
})
: undefined;
// 分享为图片:进入选择模式并预选本条(入口那条天然该已勾选,省一次点击)。
const handleShareAsImage = useMemo(
() =>
currentSessionId && messageClientId
? () => shareSelectionStore.enter(currentSessionId, messageClientId)
: undefined,
[currentSessionId, messageClientId],
);
// 伙伴私聊里「回复」是引用回复:引用块等在输入框上方(cindy-bots-runtime.md「私聊引用回复」)。
const botQuote = useBotChatQuote(simplifiedBotConversation ? currentSessionId : undefined);
const handleAddToChat = useCallback(() => {
if (botQuote && messageClientId) {
const quote = botReplyQuoteFor({ clientId: messageClientId, role: 'assistant', content });
if (quote) botQuote.quote(quote);
return;
}
if (!currentSessionId || !messageDeepLink) return;
insertSessionLinkIntoComposer({
targetSessionId: currentSessionId,
href: messageDeepLink,
});
}, [botQuote, content, currentSessionId, messageClientId, messageDeepLink]);
// /goal:隐藏助手输出末尾由 goal 协议要求模型吐出的裁决 JSON 块(仅显示层剥离,
// 原文仍保留在 DB / transcript)。pattern:末尾的 ```json {...goal_status...} ``` 围栏块。
const displayContent = stripStarterOptionList(stripGoalVerdictBlock(content), botStarterOptions ?? []);
const streamFadeKey = messageClientId
? `${currentSessionId ?? ''}\u0000${messageClientId}`
: undefined;
// Empty streaming starts stay chrome-free until there is something to read.
const botBubble = simplifiedBotConversation
&& Boolean(content.trim() || failedSamplingAttempt || ghostRenderCard);
// 私聊话题:只有本机伙伴主对话里的消息能开话题(cindy-bots-runtime.md「私聊话题」)。
const dmThread = useBotDmThreadForMessage(simplifiedBotConversation && !sharedGuest ? currentSessionId : undefined, messageClientId);
const actionBar = !isStreaming && showActionBar ? (
<MessageActionBar
createdAt={createdAt}
copyText={richAnswerText(content)}
copyLinkText={messageDeepLink}
align="left"
hovered={hovered}
simplifiedBotConversation={simplifiedBotConversation}
readAloudMessage={simplifiedBotConversation && !ghostReplyPending && currentSessionId && messageClientId
? { sessionId: currentSessionId, messageId: messageClientId }
: undefined}
compact={botBubble}
replyAction={dmThread && botBubble ? {
onClick: dmThread.open,
label: dmThread.thread
? t('bots.groupChat.server.openTopic', { count: dmThread.thread.replyCount })
: t('bots.groupChat.server.startTopic'),
} : undefined}
onFork={!sharedGuest && canFork ? handleFork : undefined}
onAddToChat={messageDeepLink ? handleAddToChat : undefined}
onShareAsImage={handleShareAsImage}
onDelete={!sharedGuest && currentSessionId && messageClientId ? handleDelete : undefined}
turnMoney={turnMoney}
turnCostUsd={turnCostUsd}
turnCostIsEstimate={turnCostIsEstimate}
userTurnMoney={userTurnMoney}
userTurnCostUsd={userTurnCostUsd}
userTurnCostIsEstimate={userTurnCostIsEstimate}
turnUsageDetails={userTurnUsageDetails ?? turnUsageDetails}
/>
) : null;
return (
<div
className={cn('flex w-full flex-col items-start gap-1.5', botBubble && 'group/message')}
onMouseEnter={() => setHovered(true)}
onMouseLeave={() => setHovered(false)}
>
<div
{...(botBubble ? { 'data-bot-reply-bubble': '' } : {})}
className={cn(
// min-w-0: flex item 默认 min-width: auto 会被 <pre> 的固有宽度撑爆,
// 导致代码块溢出消息流右边界。加上 min-w-0 让 w-full 真正生效,
// <pre> 的 overflow-x-auto 才能正常接管横向滚动。
// 伙伴私聊:回复放进与群聊同一种气泡(卡片底 + 细边框,无阴影),宽度随内容、
// 上限 720;组件式回答按可用宽度排版,气泡取满上限。右侧为回复、朗读和更多留出位置。
botBubble
? cn(
'relative min-w-0 max-w-[min(720px,calc(100%-96px))] rounded-xl border border-[var(--border-default)] bg-[var(--surface-elevated)] px-3.5 py-2.5',
hasRichAnswer(content) ? 'w-full' : 'w-fit',
)
: 'w-full min-w-0',
CHAT_BODY_CLASS,
'text-[var(--msg-assistant-text)]',
)}
>
{botBubble && botQuote && messageClientId ? (
<BotChatReplyTarget message={{ clientId: messageClientId, replyToClientId }} context={botQuote} />
) : null}
{failedSamplingAttempt && (
<div className="mb-1 text-13 font-normal leading-[1.384615]" style={{ color: 'var(--text-secondary)' }}>
{t('chat.failedSamplingAttempt')}
</div>
)}
{ghostRenderCard && messageClientId && !showOriginal ? (
// 出口钩子自绘:意识卡片替换气泡(净化后的静态 HTML,沙箱 iframe)。
// 复用 GhostToolCard(自带主机身份头 + 主题注入 + 沙箱),turn 级无
// 工具名/参数,running 恒 false(turn 已结束)。
<GhostToolCard
callId={messageClientId}
ghostId={ghostRenderCard.ghostId}
toolName=""
toolInput={undefined}
html={ghostRenderCard.html}
height={ghostRenderCard.height}
running={false}
sessionId={currentSessionId}
/>
) : isStreaming && USE_LINE_COMMIT_STREAMING && !hasRichAnswer(content) ? (
<StreamingBody
workingDir={workingDir}
content={displayContent}
localFileRefs={localFileRefs}
currentSessionId={currentSessionId}
currentSessionTitle={currentSessionTitle}
streamFadeKey={streamFadeKey}
allowPrivilegedLinks={allowPrivilegedLinks}
/>
) : (
// 默认分支 (USE_LINE_COMMIT_STREAMING=false) + 非 streaming 都走完整
// MarkdownRenderer。isStreaming 透传 — true 时 MarkdownRenderer 内部启用
// 100ms throttle, false 时无节流, 行为和该文件性能优化前完全一致。
<MarkdownRenderer
renderRichAnswers
workingDir={workingDir}
content={displayContent}
isStreaming={isStreaming}
streamFadeKey={streamFadeKey}
localFileRefs={localFileRefs}
currentSessionId={currentSessionId}
currentSessionTitle={currentSessionTitle}
allowPrivilegedLinks={allowPrivilegedLinks}
/>
)}
{/* 自绘卡在场:提供原文 ↔ 意识卡片切换(信任边界,主机绘制,始终可切回原文)。 */}
{ghostRenderCard && (
<button
type="button"
onClick={() => setShowOriginal((v) => !v)}
className="mt-1 text-11 underline decoration-dotted underline-offset-2"
style={{ color: 'var(--text-tertiary)' }}
>
{showOriginal ? t('chat.ghostHook.viewGhostCard') : t('chat.ghostHook.viewOriginal')}
</button>
)}
{/* 出口钩子后台处理中:回复已显示、意识还在跑那段的轻指示(规则 7:
spinner 挂 wrapper 的 compositor-only transform,仅 pending 时挂载)。 */}
{ghostReplyPending && !ghostRenderCard && (
<div className="mt-1 flex items-center gap-1.5 text-11" style={{ color: 'var(--text-tertiary)' }}>
<span className="inline-flex animate-spin motion-reduce:animate-none">
<Loader2 size={11} />
</span>
{t('chat.ghostHook.processing')}
</div>
)}
{/* 模型降级提示:所选模型本轮被上游静默替换(main turn 结束检测,详见
shared/modelMismatch.ts)。常显不随 hover —— 用户需要在不悬停的情况
下知道这轮不是自己选的模型回答的;icon 用 warning 语义色(token 豁免
簇),文字保持 tertiary 灰阶,不喧宾夺主。 */}
{modelMismatch && (
<div className="mt-1 flex items-center gap-1.5 text-11" style={{ color: 'var(--text-tertiary)' }}>
<TriangleAlert size={11} style={{ color: 'var(--warning-fg)' }} aria-hidden />
{t('chat.modelMismatch.notice', {
actual: formatModelShortLabel(modelMismatch.actual) || modelMismatch.actual,
selected: formatModelShortLabel(modelMismatch.selected) || modelMismatch.selected,
})}
</div>
)}
{botBubble && actionBar && <BubbleHoverToolbar side="right">{actionBar}</BubbleHoverToolbar>}
</div>
{dmThread?.thread && !isStreaming ? (
<BotDmThreadPreview thread={dmThread.thread} bot={dmThread.bot} unread={dmThread.unread} onOpen={dmThread.open} />
) : null}
{simplifiedBotConversation && !isStreaming && content.trim() && botStarterOptions?.length
? <BotStarterOptions options={botStarterOptions} locked={starterOptionsLocked === true} /> : null}
{simplifiedBotConversation && !isStreaming && content.trim() && <BotLearningFooter receipts={botLearning} />}
{simplifiedBotConversation && !isStreaming && botTaskResults?.length ? (
<div className="w-full max-w-[560px] min-w-0" data-bot-task-results>
{botTaskResults.map(card => <BotSessionTaskResultCard key={botTaskResultKey(card)}
data={{ ...card }} sessionId={currentSessionId} attached />)}
</div>
) : null}
{/* Streaming → bar not mounted at all (V1.2 验收 "流式期间不挂载");
非 turn 收尾正文(showActionBar=false)同样不挂,消息流保持紧凑。
伙伴私聊的气泡把操作栏挂在气泡右侧(BubbleHoverToolbar),这里不再另起一行。 */}
{!botBubble && actionBar}
</div>
);
});
import { isSharedTaskPeer } from '@cindy/device-link';