Losita/smartmate
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Version: 0.9.45.dev.260427

Browse Source 
后端：
1. execute 主链路重构为“上下文工具域 + 主动优化候选闭环”——移除 order_guard，粗排后默认进入主动微调，先诊断再从后端候选中选择 move/swap，避免 LLM 自由全局乱搜
2. 工具体系升级为动态注入协议——新增 context_tools_add / remove、工具域与二级包映射、主动优化白名单；schedule / taskclass / web 工具按域按包暴露，msg0 规则包与 execute 上下文同步重写
3. analyze_health 升级为主动优化唯一裁判入口——补齐 rhythm / tightness / profile / feasibility 指标、候选扫描与复诊打分、停滞信号、forced imperfection 判定，并把连续优化状态写回运行态
4. 任务类能力并入新 Agent 执行链——新增 upsert_task_class 写工具与启动注入事务写入；任务类模型补充学科画像与整天屏蔽配置，粗排支持 excluded_days_of_week，steady 策略改为基于目标位置/单日负载/分散度/缓冲的候选打分
5. 运行态与路由补齐优化模式语义——新增 active tool domain/packs、pending context hook、active optimize only、taskclass 写入回盘快照；区分 first_full / global_reopt / local_adjust，并完善首次粗排后默认 refine 的判定

前端：
6. 助手时间线渲染细化——推理内容改为独立 reasoning block，支持与工具/状态/正文按时序交错展示，自动收口折叠，修正 confirm reject 恢复动作

仓库：
7. newAgent 文档整体迁入 docs/backend，补充主动优化执行规划与顺序约束拆解文档，删除旧调试日志文件

PS:这次科研了2天，总算是有些进展了——LLM永远只适合做选择题、判断题，不适合做开放创新题。
This commit is contained in:
Losita
2026-04-27 01:09:37 +08:00
parent 04b5836b39
commit 66c06eed0a
60 changed files with 9163 additions and 1819 deletions
Download Patch File Download Diff File Expand all files Collapse all files

195
frontend/src/components/dashboard/AssistantPanel.vue
View File

@@ -153,6 +153,10 @@ interface DisplayAssistantBlock {
event?: ToolTraceEvent
statusEvent?: StatusTraceEvent
schedulePreview?: SchedulePreviewData
/** 所属的源消息 ID，用于状态查询 */
sourceId?: string
/** 所属的源消息引用，用于渲染辅助信息 */
source?: AssistantMessage
}
interface AssistantContentBlock {
@@ -223,6 +227,7 @@ const statusTraceEventsMap = reactive<Record<string, StatusTraceEvent[]>>({})
const toolTraceExpandedMap = reactive<Record<string, boolean>>({})
const assistantReasoningSeqMap = reactive<Record<string, number>>({})
const assistantContentBlocksMap = reactive<Record<string, AssistantContentBlock[]>>({})
const assistantReasoningBlocksMap = reactive<Record<string, AssistantContentBlock[]>>({})
const assistantTimelineLastKindMap = reactive<Record<string, 'content' | 'tool' | 'status' | 'reasoning' | 'other'>>({})
const conversationContextStatsMap = reactive<Record<string, ConversationContextStats | null>>({})
const conversationContextStatsLoadingMap = reactive<Record<string, boolean>>({})
@@ -502,6 +507,11 @@ function appendToolTraceEvent(
const eventSeq = nextAssistantTimelineSeq()
const eventId = `${messageId}:tool:${eventSeq}`
// 如果上一个阶段是推理，则结束并折叠它
if (assistantTimelineLastKindMap[messageId] === 'reasoning') {
finishCurrentReasoningBlock(messageId)
}
toolTraceEventsMap[messageId].push({
id: eventId,
seq: eventSeq,
@@ -536,6 +546,11 @@ function appendStatusTraceEvent(
}
const eventSeq = nextAssistantTimelineSeq()
// 如果上一个阶段是推理，则结束并折叠它
if (assistantTimelineLastKindMap[messageId] === 'reasoning') {
finishCurrentReasoningBlock(messageId)
}
statusEvents.push({
id: `${messageId}:status:${eventSeq}`,
seq: eventSeq,
@@ -554,6 +569,11 @@ function appendAssistantContentChunk(messageId: string, chunk: string) {
const blocks = assistantContentBlocksMap[messageId]
const lastKind = assistantTimelineLastKindMap[messageId]
// 如果是从推理切换到正文，则结束并折叠推理块
if (lastKind === 'reasoning') {
finishCurrentReasoningBlock(messageId)
}
if (lastKind === 'content' && blocks.length > 0) {
blocks[blocks.length - 1]!.text += chunk
return
@@ -568,6 +588,41 @@ function appendAssistantContentChunk(messageId: string, chunk: string) {
assistantTimelineLastKindMap[messageId] = 'content'
}
/**
* 追加助理推理片段到特定消息的块映射中
* 1. 采用与正文相同的块化存储逻辑，确保推理片段能按 sequence 与工具等交错排序
* 2. 如果当前时间线最后一种类型就是 'reasoning'，则追加到最后一个块，避免碎片化
*/
function appendAssistantReasoningChunk(messageId: string, chunk: string) {
if (!chunk) {
return
}
if (!assistantReasoningBlocksMap[messageId]) {
assistantReasoningBlocksMap[messageId] = []
}
const blocks = assistantReasoningBlocksMap[messageId]
const lastKind = assistantTimelineLastKindMap[messageId]
if (lastKind === 'reasoning' && blocks.length > 0) {
blocks[blocks.length - 1]!.text += chunk
return
}
const seq = nextAssistantTimelineSeq()
const blockId = `${messageId}:reasoning:${seq}`
blocks.push({
id: blockId,
seq,
text: chunk,
})
// 记录块级别的起始时间和初始折叠状态
reasoningStartedAtMap[blockId] = Date.now()
reasoningCollapsedMap[blockId] = false
assistantTimelineLastKindMap[messageId] = 'reasoning'
}
function mapToolEventState(rawStatus?: string): ToolTraceState {
const normalized = `${rawStatus || ''}`.trim().toLowerCase()
if (normalized === 'start' || normalized === 'calling' || normalized === 'called') {
@@ -993,22 +1048,21 @@ function markReasoningStart(message: AssistantMessage) {
reasoningStartedAtMap[message.id] = Date.now()
}
function markReasoningFinished(message: AssistantMessage) {
const startedAt = reasoningStartedAtMap[message.id]
if (startedAt && !reasoningDurationMap[message.id]) {
reasoningDurationMap[message.id] = Math.max(1, Math.round((Date.now() - startedAt) / 1000))
function markReasoningFinished(blockId: string, messageId: string) {
const startedAt = reasoningStartedAtMap[blockId]
if (startedAt && !reasoningDurationMap[blockId]) {
reasoningDurationMap[blockId] = Math.max(1, Math.round((Date.now() - startedAt) / 1000))
}
thinkingMessageMap[message.id] = false
thinkingMessageMap[messageId] = false
}
function getReasoningDurationSeconds(message: AssistantMessage) {
const fixedDuration = reasoningDurationMap[message.id]
function getReasoningDurationSeconds(blockId: string) {
const fixedDuration = reasoningDurationMap[blockId]
if (fixedDuration) {
return fixedDuration
}
const startedAt = reasoningStartedAtMap[message.id]
const startedAt = reasoningStartedAtMap[blockId]
if (!startedAt) {
return 0
}
@@ -1016,13 +1070,28 @@ function getReasoningDurationSeconds(message: AssistantMessage) {
return Math.max(1, Math.round((reasoningDisplayNow.value - startedAt) / 1000))
}
function getReasoningStatusLabel(message: AssistantMessage) {
const durationSeconds = getReasoningDurationSeconds(message)
function getReasoningStatusLabel(block: DisplayAssistantBlock) {
const durationSeconds = getReasoningDurationSeconds(block.id)
if (durationSeconds > 0) {
return `已思考（用时 ${durationSeconds} 秒）`
}
return isStreamingMessage(message) && isThinkingMessage(message) ? '思考中' : '已思考'
const isThinking = block.sourceId === activeStreamingMessageId.value && thinkingMessageMap[block.sourceId]
return isThinking ? '思考中' : '已思考'
}
/**
* 结束当前消息正在进行的推理块
* 1. 计算耗时
* 2. 自动折叠
*/
function finishCurrentReasoningBlock(messageId: string) {
const blocks = assistantReasoningBlocksMap[messageId] || []
if (blocks.length === 0) return
const lastBlock = blocks[blocks.length - 1]
markReasoningFinished(lastBlock.id, messageId)
reasoningCollapsedMap[lastBlock.id] = true
}
function isReasoningCollapsed(messageId: string) {
@@ -1086,6 +1155,8 @@ function getDisplayAssistantBlocks(dm: DisplayMessage): DisplayAssistantBlock[]
type: 'tool',
seq: event.seq,
event,
sourceId: source.id,
source,
})
}
@@ -1096,6 +1167,32 @@ function getDisplayAssistantBlocks(dm: DisplayMessage): DisplayAssistantBlock[]
type: 'status',
seq: statusEvent.seq,
statusEvent,
sourceId: source.id,
source,
})
}
// 从推理块映射中提取所有独立的推理片段
const reasoningBlocks = assistantReasoningBlocksMap[source.id] || []
if (reasoningBlocks.length > 0) {
for (const rb of reasoningBlocks) {
blocks.push({
id: rb.id,
type: 'reasoning',
seq: rb.seq,
text: rb.text,
sourceId: source.id,
source,
})
}
} else if (source.id === activeStreamingMessageId.value && thinkingMessageMap[source.id]) {
// 流式过程中尚未有实质文本产出时的“思考中”占位块
blocks.push({
id: `${source.id}:reasoning:streaming`,
type: 'reasoning',
seq: assistantReasoningSeqMap[source.id] || 10,
sourceId: source.id,
source,
})
}
@@ -1108,6 +1205,8 @@ function getDisplayAssistantBlocks(dm: DisplayMessage): DisplayAssistantBlock[]
type: 'content',
seq: contentBlock.seq,
text: contentBlock.text,
sourceId: source.id,
source,
})
}
continue
@@ -1121,6 +1220,8 @@ function getDisplayAssistantBlocks(dm: DisplayMessage): DisplayAssistantBlock[]
type: 'content',
seq: fallbackSeq,
text: source.content,
sourceId: source.id,
source,
})
}
}
@@ -1135,16 +1236,6 @@ function getDisplayAssistantBlocks(dm: DisplayMessage): DisplayAssistantBlock[]
})
}
if (shouldShowDisplayReasoningBox(dm)) {
const reasoningSeq = getDisplayReasoningSeq(dm)
blocks.push({
id: `${dm.id}:reasoning`,
type: 'reasoning',
seq: reasoningSeq > 0 ? reasoningSeq : 10,
text: dm.reasoning,
})
}
if (!hasContentBlock && dm.content) {
fallbackSeq += 1
blocks.push({
@@ -1180,38 +1271,16 @@ function getToolTraceStateLabel(state: ToolTraceState): string {
return '已完成'
}
function shouldShowDisplayReasoningBox(dm: DisplayMessage): boolean {
if (dm.role !== 'assistant') return false
return dm.sources.some(m =>
Boolean(m.reasoning?.trim()) ||
(m.id === activeStreamingMessageId.value && thinkingMessageMap[m.id] === true),
)
}
function shouldShowDisplayAnsweringIndicator(dm: DisplayMessage): boolean {
return isDisplayStreaming(dm) &&
dm.sources.every(m => thinkingMessageMap[m.id] !== true) &&
!dm.content.trim()
}
function isDisplayReasoningCollapsed(dm: DisplayMessage): boolean {
return dm.sources.every(m => reasoningCollapsedMap[m.id] === true)
}
function toggleDisplayReasoningCollapse(dm: DisplayMessage): void {
const newCollapsed = !isDisplayReasoningCollapsed(dm)
dm.sources.forEach(m => { reasoningCollapsedMap[m.id] = newCollapsed })
}
function getDisplayReasoningStatusLabel(dm: DisplayMessage): string {
const totalSeconds = dm.sources.reduce(
(sum, m) => sum + (reasoningDurationMap[m.id] ?? 0), 0,
)
if (totalSeconds > 0) return `已思考（用时 ${totalSeconds} 秒）`
const hasActiveThinking = dm.sources.some(
m => m.id === activeStreamingMessageId.value && thinkingMessageMap[m.id] === true,
)
return hasActiveThinking ? '思考中' : '已思考'
// 此函数已废弃，推理状态现已下沉到各 source 块处理。
// 仅保留空实现以防意外调用。
return '已思考'
}
function isMessageViewportAtBottom(viewport: HTMLElement) {
@@ -1576,7 +1645,8 @@ function rebuildStateFromTimeline(conversationId: string, events: TimelineEvent[
if (reasoningChunk) {
currentAssistantMessage.reasoning = oldReasoning + reasoningChunk
// 记录推理块的 seq 环境
// 时序化存储推理内容
appendAssistantReasoningChunk(mid, reasoningChunk)
if (!assistantReasoningSeqMap[mid]) {
assistantReasoningSeqMap[mid] = event.seq
}
@@ -1867,8 +1937,8 @@ async function submitConfirmRejectMessage() {
requestExtra: {
resume: {
interaction_id: interactionId,
type: 'ask_user',
action: 'reply'
type: 'confirm',
action: 'reject'
}
}
})
@@ -2107,10 +2177,9 @@ function processSseBlock(block: string, assistantMessage: AssistantMessage) {
if (payload === '[DONE]') {
if (isThinkingMessage(assistantMessage)) {
markReasoningFinished(assistantMessage)
finishCurrentReasoningBlock(assistantMessage.id)
}
activeStreamingMessageId.value = ''
reasoningCollapsedMap[assistantMessage.id] = true
// 整个 SSE 流结束信号
void loadConversationContextStats(selectedConversationId.value, true)
return
@@ -2150,27 +2219,23 @@ function processSseBlock(block: string, assistantMessage: AssistantMessage) {
if (!assistantReasoningSeqMap[assistantMessage.id]) {
assistantReasoningSeqMap[assistantMessage.id] = nextAssistantTimelineSeq()
}
assistantTimelineLastKindMap[assistantMessage.id] = 'reasoning'
appendAssistantReasoningChunk(assistantMessage.id, delta.reasoning_content)
assistantMessage.reasoning = `${assistantMessage.reasoning || ''}${delta.reasoning_content}`
}
if (!shouldSuppressVisibleDelta && typeof delta?.content === 'string' && delta.content) {
appendAssistantContentChunk(assistantMessage.id, delta.content)
if (isThinkingMessage(assistantMessage)) {
// 1. 一旦正文开始回流，立刻结束“思考中”阶段，避免两个等待动画同时出现。
// 2. 这样视觉上始终保持“先思考，再输出正文”的单阶段感知。
// 3. 若后端偶发交错发送 reasoning/content，也以前端阶段机兜底，优先保证阅读一致性。
markReasoningFinished(assistantMessage)
finishCurrentReasoningBlock(assistantMessage.id)
}
assistantMessage.content += delta.content
}
if (finishReason) {
if (isThinkingMessage(assistantMessage)) {
markReasoningFinished(assistantMessage)
finishCurrentReasoningBlock(assistantMessage.id)
}
activeStreamingMessageId.value = ''
reasoningCollapsedMap[assistantMessage.id] = true
// 单条消息结束标志
void loadConversationContextStats(selectedConversationId.value, true)
}
@@ -2681,18 +2746,18 @@ onBeforeUnmount(() => {
/>
</svg>
</span>
<span class="chat-message__reasoning-status">{{ getDisplayReasoningStatusLabel(dm) }}</span>
<span class="chat-message__reasoning-status">{{ getReasoningStatusLabel(block) }}</span>
</div>
<button
type="button"
class="chat-message__reasoning-toggle"
:aria-label="isDisplayReasoningCollapsed(dm) ? '展开深度思考' : '折叠深度思考'"
@click="toggleDisplayReasoningCollapse(dm)"
:aria-label="isReasoningCollapsed(block.id) ? '展开深度思考' : '折叠深度思考'"
@click="toggleReasoningCollapse(block.id)"
>
<span class="chat-message__reasoning-chevron">
<svg
class="chat-message__reasoning-chevron-icon"
:class="{ 'chat-message__reasoning-chevron-icon--expanded': !isDisplayReasoningCollapsed(dm) }"
:class="{ 'chat-message__reasoning-chevron-icon--expanded': !isReasoningCollapsed(block.id) }"
width="14"
height="14"
viewBox="0 0 14 14"
@@ -2709,11 +2774,11 @@ onBeforeUnmount(() => {
</button>
</div>
<div v-if="!isDisplayReasoningCollapsed(dm)" class="chat-message__reasoning-body">
<div v-if="isReasoningCollapsed(block.id) === false" class="chat-message__reasoning-body">
<div
v-if="block.text"
class="chat-message__markdown chat-message__markdown--reasoning"
v-html="renderMessageMarkdown(block.text)"
v-html="renderMessageMarkdown(block.text || '')"
/>
<div v-else class="chat-message__streaming chat-message__streaming--reasoning">
<div class="thinking-indicator">