[loopx-agent]

Supporting outcomes previously exposed long requirement rows before the owner chose an outcome, and repeated historical-read labels on individual rows. The overview now starts with up to three compact, closed outcome rows; opening one shows all its requirements, with complete rationale and evidence available in place.

The existing native reader refreshes quietly every 15 seconds, backs off to 60 seconds after read failures and pauses while hidden. Read freshness stays separate from verification validity and write permission: a fresh read never validates changed requirements or starts a check. Open reading, drafts and unconfirmed operations remain visible through preview changes. The design contract now explicitly requires whole-outcome disclosure and region-level freshness explanations.

Visual evidence

Tested product revision: d643ebf77; final review head: 6252293a4 (only validation/image additions after the product commit); screenshots stored at: 5ae77b06a91a4fea64c225a606598d350b495fcd. Compiled Chat UI with a real disposable SQLite/HTTP/CLI backend and public synthetic content, not installed App acceptance. Both default comparisons use six outcomes and 33 requirements, including six genuinely invalidated reviews and 27 unverified requirements. The earlier 4184c2b33 bundle is the before control.

Observed composition: the desktop default keeps current work visible beneath three closed outcome rows; the full collection remains reachable. Long Chinese/English content, a selected requirement and its original evidence, later outcomes and current work were inspected by scrolling the actual page. The 390px view wraps without horizontal overflow; its third outcome and subsequent regions continue by scrolling. Empty/sparse views retain their task anchor. Secondary explanations appear on hover/focus or click through the existing help control.

Desktop before	Desktop after

390px before	390px after

Complete content, hover help and the lower page

Selected outcome and complete requirement evidence:

The same disclosure at 390px:

Explanations appear on demand:

Later outcomes and current work remain reachable:

Validation and boundaries

• Strict TypeScript plus production dashboard and Chat build passed on main f1bd2e286 with this change.
• Existing packaged outcome journey passed with real SQLite/HTTP/CLI: default disclosure, complete content, preserved draft/scroll, invalidated review after quiet read, lost reply/replay, concurrency/rebase, read/delivery failures, lifecycle and same-ID remote isolation, keyboard and 390px touch open/close. The same disclosure oracle rejects the before bundle (zero complete-body disclosures for six outcomes).
• Packaged acceptance-contract regression passed: feature-off parity, six check states, opaque identity collision, locales, keyboard, narrow wrapping, revision replacement, read-only export and source mismatch.
• The existing write-authority fixture passed healthy/failure cadence, background pause/recovery and cancellation/read-failure protection against stale writes. Only explicit fresh retry emitted its intended check request.
• Bilingual RFC and whitespace checks passed. Added text and all eight synthetic screenshot pixels were inspected; PNGs contain no text/EXIF metadata. The semantic probe has limited supported carriers and is not exhaustive coverage.
• Exact-scope quality receipt cqr_3fd8b05bc6d2e90adb62 passes with one warning for independent screenshot review. Standard premerge completed: all direct checks plus 16 selected catalog/risk/public-boundary checks passed, with zero failed checks. A main-branch update made that run’s earlier quality receipt stale; the unchanged diff was requalified. A final quick policy check covers the subsequent test-only touch assertion commit. Exact-head pixel/privacy review of the eight PNGs remains open. The initial lifecycle-smoke failure was repaired by initializing its canonical SQLite fixture; no native fallback or assertion was relaxed.

Independent exact-head pixel/privacy review and installed App adoption remain unqualified. This author proposal does not change native Goal state, verification rules, write/check authority, model selection or scheduling.

Bounded refactoring: removed the redundant status-group disclosure and unused preview-key helper, reused the existing cached read owner and native test initialization, and kept full currentness/source isolation checks. The existing lifecycle smoke also initializes canonical SQLite and cleans up its isolated runtime; its original dry-run, actor and index assertions remain. No new configuration or read service was added.

[loopx-agent]

Reviewer: actor_kind=model_agent; model=gpt-6.1-sol; provider=OpenAI; declaration_source=runtime_reported; reasoning_effort=max

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

详细中文评审

动机

成果与要求很多时，用户还没选中一项成果，概览就已铺开长要求，并在多行重复提示历史读回；当前工作被推到后面，阅读也需要反复刷新。
同样六项成果、33条要求，旧页面直接露出要求正文；当前默认展示最多三项紧凑成果，整项正文关闭，选中后再读完整要求、依据与来源。读取能安静更新，不能把失效的自查变成通过。
当前打包页面配合真实一次性SQLite/HTTP/CLI，验证了整项开合、完整内容、草稿/展开/滚动保留、原生写入重放与失败恢复；独立取消竞争和15秒/失败60秒/后台暂停验证没有产生额外检查写入。
本次资格是源码打包与隔离后端、公开合成页面和受控竞争边界；不是安装版App的实际Goal采用，也不关闭整个研究工作区或证明模型/长程效果。
安装版实际Goal及父级工作区验收仍待原产品owner继续核验；390px的完整成果和当前工作列表仍需滚动，首屏保留Goal和当前工作摘要。

改动思路

复用既有原生成果投影和同一React Query读取owner，只把折叠层级移到整项成果，并把读取新鲜度与可写资格分开；无需新增配置、状态服务或另一份验收事实。
当前16文件交付整项阅读、安静只读恢复和原写入门，保留完整集合、所有原要求/负例和八张公开图；安装采用与更大研究界面继续由既有产品工作承接。
只缩小字体、折叠依据或隐藏几条要求，仍然没有给用户一个先选成果再读全文的入口。现在的开合是自愿阅读导航，不增加确认、配置或重新输入已知上下文的轮次；写入继续使用原有明确操作。完整列表收起后，已经打开、编辑或结果未确认的项仍可见，避免为了整齐丢失当前工作。

具体改动

依据 docs/development/design.md，规范修订 f1bd2e2865bb6fa4b7f82578f08aa97392f65db3，并核对同源frontend-delivery。Earn The User's Attention 的有界重复内容、真实状态和下一主要区域落实到三项紧凑成果；Explanations On Demand 保留完整正文、来源、依据与区域帮助；Populated Page Review 用六成果33要求的整页、异常恢复及390px验证。新增设计文字作为本次整项折叠/安静更新的明确纠正审查，没有把改后的规范当作独立验收来源。

关键代码讲解

GoalOutcomeRequirements 从完整原生要求派生状态数量和优先显示的成果ID，默认至多三项，并保留按原声明顺序显示及完整集合入口。Outcome 用整项details承载全部要求；稳定的JSON Goal/outcome键保留同来源开合和草稿，超出预览但仍在阅读、编辑或待确认的项不消失。GoalAcceptanceContractSection 继续使用同一queryKey，正常15秒、失败60秒、后台暂停；fresh只描述同Goal读取，writeCurrent仍同时要求canManage。异步取消之后还核source、generation和当前可写引用，再发送原POST。displayState 继续从原生status/basis_current投影，读取成功不会把旧自查改成独立通过。

对主干的风险

最强反例是折叠看似存在却仍默认展示全文，或一次GET覆盖草稿、修复了显示却放开旧写入。精确基线f1bd的正常Chat资产，在同一当前原生后端和未改断言下得到整项折叠0/6；当前路径通过。实际SQLite/HTTP/CLI覆盖新声明、失效自查、草稿与滚动、丢回复后同操作重放、并发版本冲突/明确rebase及来源/生命周期隔离。打包合同回归覆盖默认off、六状态、opaque ID、中英/键盘/移动、版本替换与只读导出。独立Vite竞争/时钟场景验证失败退避、隐藏暂停、可见恢复，取消期间读取失效不发旧POST，明确新鲜重试仍能发出唯一预期检查。部分传输和生命周期投影受控，不能把它们称成安装版或真实模型验证。

八张提交图的实际像素、PNG空文本/EXIF元数据和合成来源均已独立检查；另看了本轮打包真实fixture的桌面与390首屏。桌面三项标题、状态、Goal锚点和当前工作可同时理解；窄屏无横向溢出，完整第三项和任务列表继续向下滚动。展开后的长要求、原依据和下方成果可读，帮助在原位按需出现。图示产品d643的三个生产模块与当前625逐字节相同；before的4184同三个模块与实际base相同，实际base整包仍另外重建作反例，不冒称整个旧bundle与新版本相同。图片来自公开合成内容，不含私人消息、路径或实时Goal上下文。

严格TypeScript、基线和当前正常Chat构建通过。当前标准预检5直接、7 catalog、8 risk及1公共边界检查全部通过；生命周期smoke用隔离SQLite准备修复了旧fixture，原dry-run、actor和index断言保留。完整命令仍exit1/manual_review_required，八图人工门原样保留，独立像素检查不改写机器收据。第一次基线资产参数误选web外层目录，在产品断言前失败；仅改为实际web/chat后才取得0/6反例，失败输出保留。没有查询、轮询或等待CI。

我的整体评价

当前完整16文件没有验证出的阻断发现。long_horizon为preserved：原任务、操作身份、重放和权限继续由原生owner管理，没有静默执行或重复效果；user_experience为improved：先选整项成果再读全文，刷新不会打断同来源阅读和草稿。复用既有原生成果投影和同一React Query读取owner，只把折叠层级移到整项成果，并把读取新鲜度与可写资格分开；无需新增配置、状态服务或另一份验收事实。
当前16文件交付整项阅读、安静只读恢复和原写入门，保留完整集合、所有原要求/负例和八张公开图；安装采用与更大研究界面继续由既有产品工作承接。
删除重复状态分组与无用预览键helper是合适的局部简化；新读服务、额外配置或框架迁移没有当前必要。批准这一可用源码切片，不把作者自查、公开图片或一次测试当成正式App采用、整个父目标完成或合并授权。

英文简短结论

English verdict: APPROVE — 6252293. The immutable-base UI fails the unchanged whole-outcome disclosure oracle (0/6); the current native SQLite/HTTP/CLI path, packaged contract regression and controlled cancellation/cadence tests pass. All eight images were independently inspected, with source/metadata limits retained. Automated premerge checks pass while its eight image manual holds and installed-App/parent-outcome qualification remain explicit.