[loopx-agent]

Goal And Delivered Outcome

The opening description does not identify LoopX as a semantic control plane for long-running agent teams across sessions and runtimes. Returning readers also lack a concise major-milestone entry near the start of the README, while Releases mixes stable versions with frequent desktop preview builds.

Add a curated 📰 News / 📰 最新动态 section after the two-step quick start in both READMEs. Cover the project’s evolution through LoopX 2.0, Agentic Software, EdgeBench feedback research, long-horizon bootstrapping/RSI, LHTB, LoopX 1.0, and the early 200+ hour trajectory showcase and X community attention, newest first. Each entry has a distinct reader takeaway and a primary public source; ordinary version bumps or PR merges alone do not qualify. Keep the existing hero navigation and installation flow. Fold the older 2.0 background into an expandable section while retaining its migration warning, source links and existing anchors.

Per review direction, the News list does not add the vLLM case or the source-freshness postmortem. The vLLM delivery remains available in the existing Evidence showcase; this News section stays focused on the broader project and research evolution.

Use the approved English positioning, with corresponding Chinese copy:

Turn long-horizon goals into verifiable results.
A local-first semantic control plane and research workbench for long-running agent teams, across sessions and runtimes.

Link semantic control plane to the core architecture article’s definition, research workbench to the README’s existing usage section, and the ownership/collaboration benefit to the native Kanban walkthrough. Clarify that the semantic state preserves goals, acceptance, decisions, ownership and evidence; models and people use it to reassess the next step, while the runtime supplies models/tools and human authorization still governs protected actions.

Research and decision

• Core architecture: what a semantic control plane preserves directly explains the opening concept; its matching Chinese section is linked from the Chinese README.
• Native Kanban: one complete collaboration explains responsibility, evidence, takeover and result return. The link sits on the corresponding benefit instead of adding another navigation row.
• Microsoft VibeVoice (54,954 stars at observation) uses 📰 News, then small 🚀/⚡/📣 icons on individual releases.
• OpenBMB ChatDev (34,586 stars) uses 🎉 News, keeps 🚀/🎉 item cues, and moves older entries behind a disclosure.
• HKUDS LightRAG (40,312 stars) uses 🎉 News with one repeated 🎯 cue for feature updates.
• ACE-Step 1.5 (13,995 stars) uses 📰 News, a 🎉 item cue, and emoji-bearing table-of-contents anchors.
• OpenHands leads with quick start and architecture without a dedicated News heading. News is an editorial choice, not a prerequisite for a useful README.
• GitHub's README guidance prioritizes usefulness and getting started. Keep onboarding ahead of this new section.

Expected value: make major product and technical milestones easier to scan and give returning users a direct upgrade/evidence path. The adopted pattern is a single 📰 chapter marker with one semantic icon per item: 🚀 release, 🤖 engineering approach, 📊 feedback research, 🧠 long-horizon method, 🧪 benchmark, 🎉 product milestone, and ⏱️ elapsed trajectory. The heading uses the neutral newspaper metaphor because News includes mixed product and research entries, not only celebratory launches. No experiment establishes increased stars, downloads or conversion. Setup is this static documentation change; ongoing work is selecting and verifying substantial milestones and updating both languages alongside significant public releases. It adds no scheduled execution, model calls, service subscription or separate archive. Releases remains the complete history. Release dates were checked against the published UTC metadata for v1.0.0 and v2.0.0; the LHTB date is the first public research-brief merge (#4703, 2026-09-18T16:03:13Z), rather than a new benchmark run. Routine minor/patch releases and prerelease desktop snapshots stay in the full Releases history. The research entry links the published task-level results and method/limitations; it makes no equal-budget efficiency or causal-uplift claim.

Repository-wide selection

Surveyed the release timeline and stable release metadata, the overall roadmap’s product/authority/collaboration/runtime/research/adoption streams, capability and technical-direction indexes, the Showcase catalog, published benchmark notes, public blog reading routes and relevant publication history. The selection follows the project’s evolution rather than release numbering alone.

Dimension	Selected reader-facing milestone	Why / evidence boundary
Product and architecture	1.0 Workspace → 2.0 TypeScript control plane and research workbench	Significant user-facing transitions; 2.0 already links its detailed engineering history
Engineering approach	Agentic Software article	Connects goals, evidence and human feedback to engineering delivery without promoting a separate user case into a project milestone
Scientific feedback	EdgeBench research note	Feedback validity, continuation, replanning and complete cost; observed progress, counterexamples and unproven mechanism gains remain distinct
Early technical/community history	200+ elapsed-hour trajectories shared on X (2026-08-02)	Original maintainer post and the README’s public-safe trajectory evidence; broad attention is a dissemination signal, not verified adoption or proof of unattended autonomy
Development method and history	Long-horizon bootstrapping / RSI perspective	Connects early self-development to the current project; a technical viewpoint, not proof of autonomous RSI or model-weight improvement
Benchmark outcomes	LHTB public 46-task study	Task-level comparison and methodological limits; no equal-budget or causal uplift claim

Effect Programs, transaction/authority migration, semantic checks and legacy retirement are covered by the 2.0 engineering story. Core architecture and native Kanban remain contextual concept links instead of duplicate announcements. SWE-Marathon and DeepSWE stay discoverable through Evidence: SWE-Marathon’s withdrawn comparison arms and DeepSWE’s non-success feature-coverage metrics do not justify a fresh blanket performance headline. Roadmap-only or partially qualified team/scale/provider work is not announced as a completed capability. Routine 2.1/patch releases remain in Releases.

The early entry uses the original X post, published 2026-08-02T13:14:26Z and read on 2026-10-24 (UTC). Its rendered view count was 386.1K at observation. This is a changing platform display, not unique people, verified adoption, or a controlled capability result. The README leaves the metric out and connects the public technical showcase with its community attention. The post’s 220.7/272.9 hours refer to elapsed trajectories containing execution, waits, human decisions and recovery; the entry and linked evidence retain that boundary. Routine founding/rename commits remain available in the existing project-history article rather than adding every repository event to News.

Publication dates: Agentic Software 2026-10-14 and RSI 2026-10-02 follow the bilingual published articles. EdgeBench uses its first public publication (#5649, merged 2026-10-05T05:13:39Z), matching the blog index. Its current article dateline and later research continuations retain their own dates; the News date does not relabel those later observations.

Author Declaration

• Written by: model_agent · GPT-6 · OpenAI (self-reported).
• Specification and revision: no written feature specification; the accepted README positioning, repository-wide milestone curation, early technical/community history and the current News refinement request are the basis. Presentation follows AGENTS.md and docs/development/design.md at baseline adaa46677e2ebbed75c65df0ed67a78dcdc86adb.

Criterion	Disposition	Path	Verification
Curate distinct milestones from repository-wide evolution	implemented	News in both READMEs and selection table	Roadmap/release/capability/showcase/research/blog survey, public article readback; original early X post and trajectory evidence
Assess reader benefit, alternatives and ongoing cost	implemented	This description	Primary-source comparison; benefits distinguished from unmeasured conversion
Use the approved positioning and provide contextual concept links	implemented	Both README introductions	Exact visible English wording; bilingual public blog section and local workbench anchors verified
Provide factual, scannable, bilingual recent milestones without displacing onboarding	implemented	Both READMEs	Seven aligned entries with semantic emoji cues, published release metadata, public research-brief merge and real GitHub rendering
Consolidate older supporting detail without losing migration guidance	implemented	Existing 2.0 section	Original links/anchors retained; keyboard expansion checked in both languages

Scope And Continuation

Complete within this documentation scope. Only the two READMEs change. Runtime, installation, release automation and analytics are outside this change; no successor or new recurring task is required.

Validation

• Tested revision: f33ce9b2288aca2f19d1fc547d11ab1bb0910a61.
• Run state: finished. Input classes: public_fixture, synthetic.

Check kind	Result	Evidence / limitation
static	passed	git diff --check; python scripts/check_rfc_bilingual.py --base origin/main --head HEAD (no paired RFC changed)
static	passed	loopx check --scan-path README.md --scan-path README.zh-CN.md with an isolated empty registry/runtime: public boundary clean
manual	passed	Stable v1.0.0/v2.0.0 release dates and status checked through GitHub API; LHTB first public research-brief PR and the current public page/study source checked; tagged v1.0.0 Workspace guide exists (Chinese original); existing 2.0 migration and engineering links retained; both languages of the semantic-control-plane and Kanban sections opened successfully; workbench anchors resolve
manual	passed	All seven remaining bilingual article URLs were checked; first-public EdgeBench date checked against merged #5649 and the blog index; original 2026-08-02 X post date, full text and displayed view count read; bilingual article and trajectory anchors resolve; article viewpoints, reported outcomes and research limits preserved
real_entrypoint	passed	Actual GitHub Markdown rendering in English/Chinese at 1200×1100 and 430×1100; all emoji headings and seven icon cues render without horizontal overflow; GitHub's emoji-heading permalinks #-news and #-最新动态 resolve; compact hero navigation retained
manual	passed	Both 2.0 disclosures were checked with keyboard Space at the prior head; their markup, migration text, links and explicit anchors are unchanged by this milestone correction
unit	not_run	Documentation only; no runtime behavior changed

At the tested revision f33ce9b2288aca2f19d1fc547d11ab1bb0910a61, Release Artifacts / build failed in the installed Workspace browser smoke (goal-activity.mjs:437, “The empty running tile discloses the open turns”). The observed tile reads 正在执行 / 0 instead of the expected unknown-execution disclosure. This two-README proposal does not change that product surface or resolve the failure; merge qualification is not claimed from it. The screenshots establish documentation presentation, not installed product behavior or adoption/conversion gains.

Frontend / Visual Evidence

• UI impact: changed (public documentation).
• Before: f565d91fd36eef5687f9c250a19d5f20b5fc9fff.
• After: f33ce9b2288aca2f19d1fc547d11ab1bb0910a61.
• Source data: public_fixture — the public repository README rendered by GitHub, not an App/Goal or private operating session.
• Preview: English · 中文.
• Attention/craft review: onboarding remains first, and the hero navigation retains its existing links and narrow-screen line count. The hero names the semantic control plane, long-running teams and cross-session/runtime scope, with details available through contextual links. News earns its space through distinct product, architecture and research milestones; the older table and supporting 2.0 explanation become optional. Product/evidence illustrations remain in their existing default reading flow. No new cards, colors, animation, loading states or runtime controls are introduced. One 📰 heading marker and one semantic emoji per item improve scanning while staying below the title typography; all seven entries wrap readably at narrow width, and folded migration text remains fully readable on request.

First viewports

Language / viewport	Before	After
English · 1280×1000
English · 390×844
中文 · 1280×1000
中文 · 390×844

News section at the final head

The captures below show the complete section in the actual GitHub Markdown renderer at 1200×1100 and 430×1100. No image compositing was used.

English desktop	English narrow

中文桌面	中文窄屏

Type / Area

• Documentation update
• Public docs or presentation surface

Shared-authority RFC fixture impact

N/A — no shared authority, protocol, storage or runtime changes.

Boundary Checklist

• Diff, description and attachments contain only public documentation and public-safe presentation evidence.
• No benchmark runs or recurring automation added.
• Scope remains the accepted README maintenance request.
• No paired RFC changed; bilingual RFC diff check passed.
• Actual desktop/narrow before/after views are embedded above.
• All PR commits include DCO sign-off.

[loopx-agent]

Reviewer: model_agent; model=gpt-6.1-sol; provider=OpenAI; reasoning_effort=max; declaration_source=runtime_reported; observation_id=cfa785f53bf653c307bef3fc4335bc533df059f1cd3a59a2f14fc45d884e8d43

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

动机

首次了解 LoopX 或回到仓库查找重大进展、升级依据的读者，是这次 README 改动的直接使用者。

原页只用通用控制面描述定位，读者需自行翻查混有频繁预览版本的 Releases 和多篇文章。现在先保留两步开始，再集中呈现八个有来源的产品、研究和社区里程碑；语义控制面与协作责任各有就近的解释入口。

当前双语 GitHub 页面能读到八条按日期倒序的动态，沿链接找到原始依据；原安装命令、旧锚点和升级警告仍可使用。 本次只交付公开文档的定位与发现路径，不改变运行时、权限、安装或调度，不把历史研究和 owner-run 案例重新宣布为新版本的独立效果证明。

没有阻塞性发现。这是一个完整、可撤销的文档改进：让读者判断是否值得深入并找到依据，不要求先理解仓库内部对象。收益限于实际可见的发现和阅读路径，不能由此推断更多安装、收入或更好的 Agent 成绩。

改动思路

沿用原有 README、公开文章、研究页和 Releases，各来源继续拥有自己的事实；News 只是人工选择的阅读投影。选择哪些历史事件值得呈现需要编辑判断，不能把每次版本或 PR 自动转成新闻。新增内容覆盖产品迁移、上游交付、反馈研究、自举观点、失败复盘、公开对照和早期轨迹，分别给出不同的读者收获。持续成本是核对来源、日期并同步两种语言，没有增加服务、抓取器、任务表或定时执行。

最有力的反对理由是八条内容会把“为什么需要 LoopX”推到更后面。只链接 Releases 更短，但仍无法从预览版本中识别这些技术与研究节点。当前方案把安装和第一个任务留在 News 前面，用简单列表承载信息，并将旧 2.0 背景表折叠，信息量与这次公开定位目标相称。没有理由为这一页新增状态 owner、后台或通用新闻框架。

以下是 1073ef6 的阅读路径说明，依据源码和实际页面观察，不替代运行证据：

读者任务	原路径	当前路径
第一次开始	两步安装与具体任务	保留原路径，News 在其后
查重大进展	自行翻查版本与文章	八条日期、主题与原始来源
查 1.x 升级限制	旧 2.0 段落	原锚点到折叠入口，键盘展开原警告

具体改动

表示层的既有规范是 docs/development/design.md，本次按不可变基线 158b4fa7bbb350e5da74cee0f6406e49cd44f10c 读取；没有另一个独立的 News 功能 RFC。下面的 README-ATTENTION、README-DISCLOSURE、README-WHOLE-PAGE 是本评审对应现有设计段落的标签，不是假定原规范有这些编号。

关键内容讲解

1. 英文开头与中文开头把语义控制面、长程团队及跨会话/运行时范围写清楚。概念链接的双语 control-plane 和 walkthrough 目标实际存在；本地研究工作台入口也完成原生锚点定位。下文仍明确由运行时提供模型和工具，人和 Agent 依据事实重新判断，受保护操作仍受人的授权约束。README-ATTENTION 已实现：链接在相应概念上，没有增加一排导航或第二个设置入口。
2. News与最新动态各有八条。1.0/2.0 日期对照稳定 release 的公开 UTC 元数据；LHTB 的 9 月 18 日与 #4703 首次发布合并一致，EdgeBench 的 10 月 5 日对应 #5649 首发，不是文章后来续作的日期。10 月 14 日是 Agentic Software 文章日期，vLLM #54951的实际合并日为 10 月 11 日，两者没有混用。RSI 与复盘分别标明观点和负面经验，未将历史待修复项写成已完成能力。
3. 8 月 2 日 X 原帖的时间、两条 200+ 小时轨迹及执行/等待/人工决策/恢复说明均已读回；当前可见公开展示量支持“广泛传播”的有限描述。新条目明确 owner-run、自然经过时长，并链接原有轨迹边界。它没有继承原帖更强的自治口号。LHTB 页面也保留单次有效试验、替代运行、预算不齐和非因果排名的限制，News 没有把它升级为新实验或同预算收益。
4. 2.0 背景与迁移采用原生 details/summary，旧的 loopx-20-1、英文旧标题和中文旧标题均可定位到关闭状态的摘要。README-DISCLOSURE 已实现：原表、链接和“安装不会静默转换状态”的整段警告与基线文本一致，Space 可展开再关闭。README-WHOLE-PAGE 已实现：检查了真实长列表、展开项、下方 Evidence 和 Current Status，以及四种语言/宽度组合。

PR 描述的首屏截图来自父提交 0e1bf756106d0b72814533d179db2404aa72c674。当前提交只再加第八条动态；两种语言在 News 前的字节和迁移 disclosure 均与该父提交一致。我仍重新检查了当前精确 head 的首屏、完整动态与折叠内容，没有沿用先前较小范围的审批结论。

对主干的风险

主要风险是把文章观点、历史观察或传播量读成已经验证的产品效果。实际来源仍区分上游作者报告、研究中的混杂因素、尚未释放的历史修复和 owner-run 案例；新摘要的限定语与这些边界一致。这里的“保留状态”也没有替代模型判断或扩大执行权限。较长列表的编辑成本真实存在，应继续只选择有独立收获的重大公开事件；全量历史仍由 Releases 负责。

当前精确 head 的标准 loopx canary premerge --from-git-diff 通过：4 项直接检查、4 项目录 canary、8 项风险 smoke 和公共边界检查均通过。另以空的隔离 registry/runtime 扫描两份源文件，公共边界干净；临时 runtime 尚未创建的提示不影响该扫描。两种语言、1280×1000 与 390×844 的实际 GitHub 页面没有文章或文档横向溢出；关闭/展开迁移段正文为 16px，老锚点、工作台和轨迹入口都等待到真实可见位置。

只用字符串或目录测试会漏掉折叠后无法到达的警告、窄屏长内容和错误日期，因此另做了实际浏览及原始发布记录对照。一个窄屏长截图首行出现捕获缺失，保留该捕获问题，改用同 head 的普通窄屏截图核对首条、长图核对后续七条和桌面完整列表交叉检查；没有把失败截图当作产品通过证据。未新增镜像实现的测试或运行任何研究作业。按当前 wait_for_ci=false，未查询、等待或轮询远端 CI；作者所述历史 installed Workspace smoke 失败未由这份文档改动修复，也未据此宣称安装产品资格。

我的整体评价

在本次文档范围内判断 goal_achieved，long_horizon 为 preserved，user_experience 为 improved。持续工作、身份、权限和结算规则由现有系统保留；读者现在更容易识别重大进展及其证据，安装无需增加一次选择或确认。可见定位更具体，英语窄屏描述多占部分行高，仍在既有层级内；八条动态占用更多纵向空间的成本由不同主题与来源支撑，不据此宣称转化提升。

已对照实际 GitHub 的完整双语页面：首屏保持目标与开始入口的层级，八条动态以日期和主题组织，完整证据及失败边界继续在下文；迁移细节可按需展开，窄屏正文与展开表格仍可阅读。

未验证已安装 Workspace、阅读理解、采用率或转化收益；历史研究未重跑，X 展示量只说明公开传播，不能证明长期自治或效率。

bounded refactoring 的处理是复用原列表、现有来源与原生 disclosure，保留独立书签兼容，避免复制归档或引入 UI 框架；无须额外跟踪任务。语义控制面一词沿用已有架构文章，不引入新枚举、协议、CLI 或共享状态。回滚仅需还原两份 README；评审结论不代替独立合并与安装授权。

English verdict: APPROVE — 1073ef6. No blocking finding. Both READMEs preserve onboarding and migration semantics; eight dated milestones were checked against public primary sources. Actual bilingual GitHub desktop/narrow rendering, disclosure/anchor recovery, exact-head local premerge and public-boundary checks passed. This qualifies documentation presentation, not installed Workspace, causal benchmark gains or adoption. Merge authority remains separate.