[loopx-agent]

Existing Lark conversations were followed by an always-expanded setup form and long command instructions. Existing conversations now remain the primary view; “Connect an assistant” opens the optional form, with two columns on desktop and one on narrow screens. An empty connection list keeps setup open.

Complete chat/commission guidance moves to the shared command-help glyph. Scope, source gaps, invalid grants, notification effects, operation errors and the refresh action remain visible. Empty Goal-connection help moves out of the table into the toolbar; populated tables receive explicit column-header and cell roles. No backend, permission, session, configuration or scheduler policy changes.

Validation at 71780e8beb4473de1214a018250d0386b76655f2:

• Production Chat/TypeScript build, changed browser-script syntax, CSS tokens and bilingual diff check passed.
• Four existing packaged scenarios passed: private-steward-scope, progressive-loading, steward-group-trigger and lark-cli-missing. Includes exact-target scope failure/retry, explicit read-only refresh, late/failed catalog recovery, focus/tap/Escape, empty setup, long mixed connections and 390px layout.
• Same frozen browser harness: immutable base presentation fails the default-readability invariant; the restored exact head passes the complete scenario.
• 308 existing Lark HTTP/Core/provider tests passed in237.54s, using disposable state and synthetic provider IO.
• Exact-scope change-quality receipt passed. Standard premerge: 4 direct and 8 selected checks passed. The raw result retains 10 raster manual holds; all 10 captures were inspected visually and for metadata. Shared-registry diagnostics retain two unrelated existing Goal-authority errors; their public-source scan is clean.

The source change is complete and reviewable. Installed WebView accessibility/adoption and live-account fault injection are not certified by the packaged screenshots; monitor/report configuration and the broader chat/research journeys remain outside this slice.

Visual evidence

Production Chat compiled from the named revision; all data is synthetic. Before captures use baseline a9a720db5ef77f52ac9a3c1e59be7df324149aa6. Desktop 1512×982; narrow 390×844. The first view now exposes existing scope, state and the original-session action before optional setup. Full permission effects remain beside connection creation; complete command paragraphs stay readable on demand. Three mixed connections retain partial coverage, unconfirmed delivery, a long bilingual title and an invalid workspace without hiding the failures.

Default desktop before	Default desktop after

Default narrow before	Default narrow after

Expanded setup and complete command help

Source gap and populated mixed-state review

[loopx-agent]

Reviewer: model_agent; gpt-6.1-sol; OpenAI; runtime_reported; reasoning_effort=max (latest host-recorded Turn; active_turn_verified=false).

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

动机

在设置中管理已有飞书会话的用户，需要先确认连接范围、状态和继续原会话的入口。 原页面先铺开六项新建配置和两段命令说明，已有连接下方的操作被拉长；现在已有连接先出现，新连接按需展开，空列表则直接展开配置。 打包页面已验证现有范围和错误保持可见，刷新不依赖展开配置，完整命令说明可通过聚焦、点击和 Escape 读取或关闭。 本次仅整理 Lark 设置的会话与连接展示，不改变后端权限、会话归属、配置默认值或调度策略。 已安装 WebView 的最终采用与辅助入口仍须独立回读；完整双端聊天、历史工具、研究用量和监控报告编辑不由本 PR 宣告完成。

改动思路

现有组件已经拥有连接读取和授权操作；用原生 details 组织可选配置，并复用 ContextHelp，足以修复展示层，不需要新增持久状态或并行权限所有者。 本 PR 交付已有会话优先、响应式配置、完整命令帮助和有效表格可访问语义，保留现有错误与恢复路径。 不能把整张连接卡一起收起，否则已有范围、原会话入口、收件缺口和失败恢复也会消失；只隐藏文字又仍留下六行无关的新建表单。因此保留每个已有连接，把配置放在一个原生折叠区，空连接时直接展开。标题旁的来源和命令图标分别沿用共享帮助组件。刷新移到卡片标题，读取或恢复状态无需先展开配置。没有新的连接缓存、权限决策或后台执行入口。

具体改动

规范基准为 docs/development/design.md 和 docs/development/frontend-delivery.md，版本 a9a720db5ef77f52ac9a3c1e59be7df324149aa6。explanations-on-demand 已实现：完整命令与委托说明经共享图标读取，授权后果留在连接按钮旁。populated-page-review 已实现：检查默认、展开、空态、来源缺口和三个混合连接，包含长中英文标题及390px布局。task-first-delivery 已实现：从已有连接进入原会话，显式升级失败后保留原范围，经只读刷新和确切目标重试读取结果。规范本身未在本PR修改。

关键代码讲解

1. apps/presentation/dashboard/src/features/personal-workspace/private-conversation-panel.tsx:12 的 PrivateConversationPanel 仍从原有类型化读取获得连接及版本。rows.length===0 只决定新建配置的初始展开；用户手动展开由浏览器管理，不写入新配置。原行中的授权范围、监听、收件缺口、通知效果、无效授权和错误保持在折叠区外。连接、升级、解绑仍使用原目标和原条件；帮助及展开不会授予权限。
2. apps/presentation/dashboard/src/features/personal-workspace/lark-settings-page.tsx:183 的 LarkSettingsPage 将空连接帮助放入工具栏，空集不再构造没有数据的表格；有记录时五个表头和每行五个单元格具有明确角色。原查询过滤、连接编辑、解绑、目录核验与取消逻辑不变。此处修正结构所有权，不复制另一套帮助组件。
3. apps/presentation/dashboard/src/features/personal-workspace/private-conversation.css 只约束既有会话区域：正文从说明小字回到正文字号，已有行之间用间距和分隔线区分；配置桌面两列、窄屏一列，窄屏长标题和继续按钮上下排列。完整文本没有通过全局隐藏或缩小字体来压缩。
4. examples/personal-workspace-browser/private-steward-scope.mjs:5 与 examples/personal-workspace-browser/progressive-loading.mjs:5 扩展既有场景，保留原500失败、范围重试、目录晚到、未知与空集区别，再验证完整帮助、可见刷新和混合状态。examples/personal-workspace-browser/steward-group-trigger.mjs 在既有群连接路径中核对表头/单元格，未另建重复测试框架。十张合成截图直接放在PR描述，数据与安装边界明确。

对主干的风险

主要风险是收起配置时一并隐藏真实失败、授权后果或恢复入口，或者仅凭不溢出和测试绿灯判断可读。相同冻结测试输入下，主干展示代码以“Existing connections are buried under optional setup or command instructions”退出1，恢复本head后完整场景通过，工作树回到确切提交。首次窄屏长标题仍把按钮挤在正文侧边，实际查看后改为上下排列。键盘重试测试还曾错误等待全局就绪；现等待确切升级控件可操作，保留中间失败记录。

四项既有打包场景通过；308项原生Core/提供方/真实HTTP处理器测试通过，提供方IO为合成数据。生产构建、TypeScript、脚本语法、31份CSS、双语差异检查通过。标准premerge的4个直接检查及8个选择检查通过，原始结果仍为manual_review_required，十张PNG逐一检查像素和元数据后未见私有信息。独立共享registry诊断仍有两条无关的既有Goal权威错误，未改写为全绿；本差异公开边界扫描干净。确切范围CQR通过；依当前Goal策略不查询或等待CI。无存储迁移，回滚本提交即可恢复展示。

我的整体评价

本次为 justified_increment：已有会话读取和可选配置得到一个完整、可独立验证的展示改进；更大的原会话与研究体验仍沿原任务继续。long_horizon 为 preserved，没有测得长程效率收益的声明；user_experience 在本次设置页范围为 improved。默认首屏能先找到当前连接及原会话入口；390px 长标题与按钮上下排列，完整命令帮助保持正文字号，展开配置的权限后果仍紧邻连接按钮。 合成 API 证明编译页面的交互与布局，不等同真实账号故障注入或已安装 WebView 验收；当前没有这两项完成声明。

English verdict: APPROVE at 71780e8beb4473de1214a018250d0386b76655f2. Existing conversations remain primary, complete help and consequential states stay accessible, and exact-target recovery semantics are preserved. Paired base/head UI evidence, four packaged journeys,308native/HTTP tests and declared premerge checks support this bounded presentation change; installed adoption remains separately unverified.