[loopx-agent]

PR review bodies currently concentrate English in the final verdict. The shared checker also rejects headings such as 动机 / Motivation, so readable bilingual titles cannot pass the existing publication check.

This change keeps the five canonical Chinese section IDs and adds exact bilingual presentation aliases in their existing owner. The template and host adapter use English structure, concise Chinese explanation, informative finding titles, Problem / Evidence / Impact / Fix, and comparable PASS / FAIL / UNVERIFIED observations throughout a complete review. This is writing guidance; evidence floors, head binding, full coverage, hidden-content checks and verdict/merge authority stay intact. Policy revision advances from 24 to 25.

中文锚点和原有完整证据要求保留；不按语言比例打分，不为排版重复同头评审。

Compatibility and bounded scope

• Existing Chinese headings remain valid. Only the five explicit bilingual aliases are accepted; arbitrary suffixes, English-only headings, mixed duplicate sections and hidden/empty content remain rejected.
• One shared title map serves the existing parser and template. No new renderer, configuration namespace or language heuristic is introduced.
• The adapter documents GitHub's body-only edit route for one's own writing-only corrections, with original evidence preserved and unchanged ID/commit/state/text readback. Substantive conclusions and changed heads still use the normal review workflow.
• The thin-adapter regression ceiling changes from 180 to 190 lines: the measured source grows from 165 lines / 11,818 UTF-8 bytes to 181 / 13,018 to preserve edit authority and old-checker compatibility clauses. It is not a hard host-loading budget; no obligation was removed to fit it.

Validation

Tested proposal revision: 1accf9366538cebdd8b9886abc939e5d241fa929.

Check	Result
Focused body, contract, check-result and GitHub scan tests	PASS — 356
Complete historical bilingual body, public CLI	Immutable base rejects the aliases; changed source accepts their canonical sections. The fixture is historical, not a current-head re-review.
Legacy Chinese body / bilingual alias parity	PASS — identical check result and prose floors; duplicate, hidden, arbitrary-title and empty-risk negatives covered
Public PR-review canary	PASS — replaced obsolete Chinese-only wording assertions with actual template/parser semantics
Premerge on latest target 99921829f2c5fa2ad35a024c72abb88717003ef3	PASS — 6 direct, 8 catalog, 8 risk-selected checks and public boundary scan; no manual holds
Ruff / configured mypy / public scan	PASS — all changed Python paths, 17 configured typecheck sources and all 8 changed source paths

The initial premerge run failed on the stale 详细中文评审 assertion. That failure is retained; the repaired canary checks canonical sections and alias parity instead of restoring obsolete wording. Semantic inventory found no supported new vocabulary carriers; manual assessment classifies the exact aliases as local presentation of existing section IDs.

Actual rendered review evidence

Existing review #7338 was revised through a body-only edit. Its review ID, f3d3e6aa8e5acbf51e0a038fcae01131db29067b commit, COMMENTED state, findings and APPROVE conclusion are unchanged. Independent GitHub readback matches the checked full body. The current native inventory correctly disallows a new execution for this concluded head; the original publication plan validates the writing correction, without a force-audit override.

These are actual GitHub desktop/narrow captures of that revised review, not packaged LoopX App or installed parser evidence. Legacy Chinese headings plus English substructure work with the current installed checker; preferred slash headings become available after this proposal is merged and installed. The first viewport keeps the required identity/author-account conclusion and short Chinese motivation; the populated risk table separates pass, conditional skips, manual hold and unverified scope. At 390px the native page wraps within the viewport. No claim is made about a custom responsive UI or installed adoption.

Before — desktop

After — desktop

Populated risk and evidence table

After — narrow viewport

This proposal does not merge, upgrade an installation, dismiss findings or launch additional reviews merely to demonstrate layout.

[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)

Reviewed exact head: 1accf9366538cebdd8b9886abc939e5d241fa929. No blocking finding. 本评审按 policy 24 的规范中文锚点组织，英文子结构保持可读。

动机

Motivation. 维护者和评审 Agent 需要从整份评审中快速找到问题、实现依据和验证限制。

此前合法的“动机 / Motivation”标题会被检查器当作缺失章节；现在五个指定双语标题复用原章节，中文标题与旧正文仍可使用。

本提案让双语标题经过公开 CLI 和隔离安装包完整校验，并提供英语结构与简短中文解释的指导。

本次不证明模型一定遵从新写法、读者理解或长程效率收益，也不授权修改他人的评审、改变结论或执行合并。

改动思路

Approach — one map, existing evidence owner. 保留一个标题映射供解析器和模板共同使用，证据与权限继续由原 owner 判断；不增加语言分类器、渲染器或第二套状态。

当前切片交付标题兼容、五段写作指导及自有正文纠正流程；现有公开 CLI 和包内 Skill 均可消费，不要求新的产品页面。

没有改动完整评审的先后顺序：先规格与完整提案、再正反例和验证、检查精确正文、最后发布及回读。原“完整中文主体”调整为英语结构与简短中文解释，是明确的写作指导变化；没有语言比例打分。保持原状会继续拒绝合法双语标题；只改说明又不能修复解析，所以这组小改动有完整的现有调用方。交付的经验提示仅用于核对这条实际阅读路径，不被当作新的恢复门或收益证明。

具体改动

Specification. 先读基线 99921829f2c5fa2ad35a024c72abb88717003ef3 的 loopx/capabilities/pr_review_queue/README.md，再审整份 8 文件 +140/-25 提案；实际 target 观察为 e20e4b684d04e4faed8ca3437880c9bbd70070a0，共同祖先仍为该基线。Review Flow 4、Review Flow 6、Review Flow 8 均为 implemented：完整证据、五段内容及精确版本/权限边界保留。

关键代码讲解

1. REVIEW_SECTION_TITLES 以一个五项映射派生原中文章节顺序；旧正文没有迁移要求，也没有第二套英文章节 ID。
2. _review_sections 仅接受完整匹配的旧标题或指定别名，再交回原正文可见性、重复检测和长度检查。中文加双语重复同一节仍报错，标题包含某个中文词并不足以通过。
3. _section 从同一映射生成建议标题，保留原节名、空白模板及 40/80/180/120/60 正文下限。build_review_template 在既有输出提示中加入可用的英语结构，policy 从24升25；不能拿旧 policy 结果直接改号充数。
4. Skill 纠正“已发布评审正文不能编辑”的错误说法。GitHub 官方接口 确有 body-only PUT：只为自己的写作纠正保留原文/证据，核同 commit/head，更新后核 ID、state、全文并重检；新 head 或实质结论仍走正常评审。旧检查器用中文锚点配英文子结构。openai.yaml 与能力 README 同步；三份测试/smoke 保护真实解析及模板，不再冻结旧“详细中文评审”字串。

Bounded refactoring. 已应用的收敛是原节名列表改由共享映射派生；无需新 helper、设置或渲染器。Skill 实测 165→181 行、11818→13018 UTF-8字节，回归上限 180→190 不是宿主硬限制，新增的编辑权限及旧检查器恢复条款未为预算删减。

对主干的风险

Risks — preserve evidence while accepting presentation aliases. 最强反例是孤立标题测试通过，但完整 check-result 的公开说明匹配仍拒绝别名，或混用别名掩盖重复/空白证据。我以同份完整合成评审走实际控制台与物理安装包，而非只调用新函数。

Check	Observation
PASS — repository suite	565通过；43项显式可选模型/provider用例跳过，未冒充模型遵从。
PASS — native CLI	同Python3.12的完整旧正文响应逐字段相同；五个单独别名及全别名在基线拒绝、新头通过，字数下限相同；18个重复、隐藏、空白、任意标题、错head、矛盾verdict反例仍拒绝。
PASS — physical package	1748个Py/TS成员及两份Skill资产与Git/wheel/隔离安装一致；9个实际包CLI正反例与源码完整响应一致，fixture packet为policy25。
PASS — local gates	changed/required Ruff、mypy17、模块预算6、RFC/diff、公开canary和构建通过；标准premerge六直接、八catalog、八风险及公共边界自动检查通过，无manual hold；精确CQR已记录并验证。
UNVERIFIED	正式安装升级、模型遵从、独立理解/阅读速度和长程净益；当前策略未查询、轮询或等待CI。

测试入口最初误写了一个不存在的测试文件；私有CLI探针先把shell包装器当Python执行，随后合成的Example Model声明被真实运行时正确拒绝。改用现有console入口及当前运行时声明后得到上述结果，未改产品、断言或丢弃失败。作者旧canary断言失败仅是其历史报告，本次当前canary通过不倒补它。

语义与 CI 对齐

Semantics. 这是原有五节 ID 的本地展示扩展，既有协议 schema、canonical字段和权限未变；新可接受输入及写作指导默认变化已公开。可选功能关闭不因Skill可用而激活；没有新语言启发式或消息/merge权。语义探针未发现支持的新词汇carrier，不能替代本次人工条款核对。专用Markdown处理留在原Python能力owner，未新造共享状态决策。

我的整体评价

Verdict — APPROVE, source/package scope. 该切片目标已实现，长期工作边界 preserved，具体CLI接受性和桌面定位体验 improved；没有把这些机械/视觉观察推成完整成本或模型质量收益。整套8文件与周边调用方已独立审过，未继承作者测试作为独立验收。

已检查实际 GitHub 评审的桌面首屏、下方混合验证表和 390px 页面；英文结构便于定位，窄屏三列表格仍偏挤且长词会断行。没有新增折叠或响应式组件。

演示使用原评审的既有证据和旧中文锚点；未证明正式安装已采用斜杠标题、模型遵从、移动阅读速度或独立理解收益。

旧中文标题和旧检查器路径兼容有真实历史正文及独立部署依据，应保留。桌面示例是已有review f3d3e6aa8e5acbf51e0a038fcae01131db29067b 的展示修订；不把它当本提案运行资格或再发同头评审的理由。分支仍落后主干，精确review、readiness、maintainer受保护决定各自保留；COMMENTED 是共享作者账号限制下的批准结论，不是GitHub formal批准或合并权限。

English verdict: APPROVE — 1accf9366538cebdd8b9886abc939e5d241fa929. Exact bilingual aliases pass the real CLI and isolated package while legacy full responses, evidence floors and 18 negative cases remain protected. 565 tests pass with 43 explicit optional skips; local premerge checks pass. Existing GitHub desktop/narrow rendering was inspected; narrow tables remain dense. No installed adoption, model-compliance or merge-authority claim.