Fix clipped status-only conversation labels

[loopx-agent]

Goal And Delivered Outcome

The active “Status only” label used the same fixed 36px grid as icon-only controls. Its text wrapped vertically beyond the button on desktop and phones. It now fits beside the icon on one line, without another interaction.

• Related to fix(chat): consolidate model controls and retain native reply parity #6756 and fix(chat): show real model identity and simplify Agent conversations #6699; target: main, tested base 105c4bc15091.
• Before: 56–57px of content in a 34px content box. After: the Chinese label fits in a 98px control; English also fits at 390px. Both Manager and Goal conversation headers were checked.

Author Declaration

Written by: model_agent · OpenAI Codex, GPT-6 family; exact model variant and reasoning effort unreported.

Implemented against

Specification: LoopX design contract at the tested base.

Criterion	Disposition	Implementation	Evidence
Earn The User’s Attention	implemented	Active mode remains directly readable	Packaged before/after viewports below
Typography / Responsive Behavior	implemented	Existing body tokens; content-sized active label	Chinese 1512px/390px; English 390px
Accessibility	implemented	Existing button, label and aria-pressed retained	Keyboard focus/Enter and refresh readback

Self-check: reviewed the complete six-path diff, existing header switch handler and responsive rules. Two local CSS rules reuse the existing state and prevent icon shrinkage. A new component, setting or state owner would add unnecessary structure here.

Scope And Continuation

Complete within this label-layout scope. The inactive icon control and original executor-selection handler are retained. Broader conversation, native IME, shared-session and installed-App acceptance remains with the existing work; this repair does not certify it.

Validation

Tested source matches committed CSS blob dda47243dafaa3990e20c30538d6335ffad730d7 at 5a6a5ad5c046553e63082b3543a6878b44369025.

• Packaged Chat build/typecheck and source verification passed; typography and existing workspace contracts passed.
• Actual production ChatHTTPServer, isolated empty workspace and synthetic File-authority Goal: Chinese desktop/phone and English phone, refresh, keyboard focus/Enter, no label or document overflow. No sessions or Turns were created. Browser preview only; no native App installation or model execution.
• Premerge: 14 selected checks and four direct checks passed. Its docs/public-boundary selection is supplemented by the actual CSS/browser checks above.
• Required validation remains non-passing: root-module tests report 1 failed, 5 passed, with 146 modules against the pinned 145 limit (workflow_skill_install.py). No Python inputs change here. The existing repair is tracked in fix(packaging): restore top-level module budget #6768; the exact-scope quality receipt remains failed, without a baseline waiver. No merge is requested as already qualified.
• Public text and PNG metadata scans passed; all five images were visually inspected.

Additional native qualification at the unchanged head

A release-mode Tauri QA App was built from 5a6a5ad5c046553e63082b3543a6878b44369025 and connected to a genuine privately installed wheel from that source. The App identity and updater endpoint were overridden only for the isolated QA bundle. Production Rust, packaged frontend and backend sources were unchanged; no installed user App or default CLI was replaced. The read-only status/Chat service identities and disposable registry routes were checked before native input.

• Physical macOS Pinyin input in the actual Tauri Goal conversation: candidate-confirmation Enter leaves the draft in place, both idle and busy. A subsequent Enter sends normally or creates one real next-Turn queue entry.
• Shift+Enter preserves the exact Chinese multiline message; Command+Enter steers the precise active upstream Turn and falls back to ordinary send when idle. The queued message actually executes after the original Turn ends.
• Status refresh retains the exact queue entry. Normal quit/reopen returns to the same Goal and canonical Session with all saved messages, no added Turn or steering, and no remaining active/queued work. Reopen performs one thread/resume reattachment to the same saved ordinary Chat thread; this is not evidence for the separate attached-provider no-resume boundary.
• The executor is a scripted stdio fixture: five Turn starts and one steering request, no actual model inference or private conversation test. The native shell's production minimum width remains960px; prior390px frontend/WebKit evidence remains separate. This adds desktop native qualification, not native390px, live Lark/CS-Notes, installed-user-App adoption or parent-task completion.
• The unqualified source/editable CLI was first correctly rejected. A genuine wheel and canonical isolated global registry repaired the QA setup; no runtime identity was manufactured or gate bypassed. The QA App and its owned service ports were closed afterwards.

The required module test was rerun to reconcile the earlier description's count: 1 failed /5 passed,146>145. The existing failed quality receipt and merge hold remain; the exact-head layout approval does not waive them.

Visual Evidence

Actual first viewports from the packaged preview. The before/after pairs use the same isolated backend and unavailable-executor state. Model metadata shown in the empty preview is unexecuted configuration.

View	Before	After
Desktop, Chinese
390px, Chinese

Goal conversation, English at 390px (candidate):

[loopx-agent]

Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · reasoning_effort=max

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

审查绑定 5a6a5ad5c046553e63082b3543a6878b44369025。无阻塞发现；此 COMMENTED 承载代码批准结论，不代表正式 GitHub 自批准、required quality 或合并资格已通过。

动机

在 Manager 或 Goal 对话中选择「仅查状态」的用户，需要直接看清当前模式及“不调用模型”的含义。基线把图标和文字塞进固定36px图标按钮，中文在桌面与手机越出按钮，英文手机还会挤压图标；候选按内容扩展按钮，原位置即可读清当前模式。独立的准确头打包预览已在真实 ChatHTTPServer 与合成 File-authority Goal 上读回：中英文标签单行完整，键盘切换与焦点可见，刷新保留选择，未选中图标布局与基线一致。验收只覆盖状态按钮的布局及既有操作；不认证正式 App 安装、模型对话、原生输入法、飞书或共享会话，也不关闭这些已有产品验收缺口。

这是当前模式可读性的修复：保留原状会让用户误读选中状态，隐藏文字或缩小字号又会丢掉直接反馈。原有操作已经表达用户意图，所以修复增加零个必需陈述、选择、确认或准备步骤。

改动思路

复用 ChannelHeader 的现有 statusOnly、aria-pressed 和切换 handler，只在同一 stylesheet 中让选中标签按内容布局；无需第二个组件、状态 owner 或设置。该切片完成已复现的标签越界修复；原执行器可用性、会话创建、发送和后端权限继续由已有 owner 负责。只改变选中状态的排版，没有把译文、图标或 CSS 变成第二个状态 owner。

同一真实页面与后端	基线 CSS 对照	准确头
中文 Manager，1512/390px	36px 控件内文字越界	98px，单行完整
英文 Manager/Goal，390px	文字越界，Manager 图标可被压扁	113.8125px，图标16px，单行完整
未选中的 Goal 图标	36px、16px SVG	同位置、同尺寸、同字体，测量结果一致

此表来自独立执行。使用准确头构建的打包 Chat、生产 ChatHTTPServer 和合成 File-authority Goal；对照在同一真实页面只移除两条新增 CSSOM 规则。源码核验表明移除它们即可逐字还原不可变基线 CSS，全部 JS/后端也未改动。它是明确标识的历史样式反例，不冒称另一次完整基线 App 构建。

具体改动

完整差异六个路径：生产 CSS 仅新增两条规则，另有五张公开合成首屏，共562878字节；一个提交带 DCO。修改前规范为 docs/development/design.md，revision 105c4bc15091afac832495167cb98a1389951a04，并结合该版本 task-first delivery；已在读差异前确认相关条款。

• Earn The User’s Attention：implemented。整体首屏能直接读清当前模式；Manager 与 Goal 两个真实入口均验证。PR 内五张中英文/桌面/手机图片已逐图检查。
• Typography / Responsive Behavior：implemented。沿用 body/leading token，实际仍为14px/20px；选中内容横排、留6px间隔和9px边距。中文98px、英文113.8125px，实际文字范围在按钮内，页面无横向溢出。
• Accessibility：implemented。既有 button、可访问名称、title、aria-pressed 与键盘入口保持；真实 Enter 切换、可见焦点和刷新/重载已读回。没有可用执行器时保持仅查状态。

关键代码是 personal-status-mode 的选中规则：以准确布尔属性选择内容尺寸 flex，未选中状态继续使用既有图标按钮。下一行 SVG 的 flex: none 防止标签挤压眼睛图标。实际手机覆盖规则与 Goal 导航换行也已检查，不能只看这一处 CSS。

ChannelHeader 的原 handler 未变化：选择状态由现有 selectedAgentId 决定；离开仅查状态时找第一个可用执行器，否则保持原状态。独立用仅响应版本探测的合成 CLI 检查 true→false→true，未执行 Agent；再从真实会话 API 读到空列表，磁盘无 Turn。没有引入新设置、持久化字段、公共协议或跨语言决策 owner。

对主干的风险

最强反例是内容尺寸按钮挤占窄屏导航、图标变形或焦点消失，而构建和静态合同仍全绿。本轮用现有生产入口做独立文字范围判断：四个历史 CSS 对照均失败，准确头均通过；未选中图标的完整位置/尺寸/样式测量一致。可用探测执行器的键盘切换、不可用状态、Goal 刷新和重载都保留真实反馈，没有发送消息或调用模型。

准确头的 packaged build/typecheck、source verification、font 与现有 workspace contracts 通过；标准 premerge 四项 direct 和十四项选择检查通过。其 selector 偏 docs/public，所以这些检查不能代替实际浏览器验证。五张 PNG 的 CRC、文本/EXIF 元数据与像素已检查，无私有操作内容。独立浏览器夹具及服务已清理，没有正式 App 安装或默认 CLI 变更。

required 根模块检查仍为 1 failed / 5 passed，146 > 145，workflow_skill_install.py。相同不可变基线和本头的完整错误只差末尾耗时；检查器、预算 fixture、offender 字节及完整146模块清单相同，且本 PR 不改 Python。保留这个原有质量失败，没有放宽预算或扫描范围。作者正文报告的通过数不替代这里独立观测的5；其精确 scope CQR fail 仍归作者，reviewer 工作区的原生核验为 receipt_missing，不把它说成可复用的 passing receipt。按当前策略没有查询、轮询或等待 CI。

我的整体评价

APPROVE；goal_achieved 仅限已复现的标签布局切片。 user_experience=improved：原位置直接读清当前模式，零额外交互轮次；long_horizon=preserved：选择与会话 owner、任务/调度/权限未改动，重载和真实零会话读回支持这个有界判断，不宣称 Agent 效果收益。邻近改善是同 owner 的内容尺寸与 SVG 约束；进一步组件或框架重构没有必要。

已经采用召回经验中“真实任务、整体首屏与失败后读回”的方法，没有继承旧案例结论，也没有把记忆投递、图片或本次代码批准当作产品采用。required quality、正式安装、原生 IME、飞书、共享会话和上层产品验收仍各有边界，审查不授予合并或发布权限。

English verdict: APPROVE 5a6a5ad5c046553e63082b3543a6878b44369025. Independent source-packaged Chat on the real production backend confirms readable Chinese/English Manager and Goal labels, keyboard operation, reload and inactive geometry parity. Removing exactly the two CSS rules reproduces the historical defect. Build and focused checks pass; the independently identical pre-existing146/145required failure and CQR qualification gap remain. CI was not consulted; no merge, installation or parent-product completion is implied.