[loopx-agent]

Goal And Delivered Outcome

An author could pass pr-review --check-result with an incomplete or wrong author-conclusion title, then have the same review rejected by merge readiness. Publication checks now use the same visible, exact title as published-review readback. Correct author approvals, author requests for changes and independent formal reviews retain their respective paths.

The existing instruction-semantics check also required authorization_basis for intentional_change_validated without listing that field in its structured contract. The contract now declares the conditional field and the checker consumes that declaration. Policy revision 24 requires a fresh result under the current contract; published review history is not rewritten.

Author Declaration

• Written by: model_agent · GPT-6.1 Sol · OpenAI.
• Implemented against the existing author-conclusion contract and instruction-semantics contract.

Existing criterion	Disposition	Implementation and decisive check
Author COMMENTED conclusions carry the matching exact title and complete bilingual body	Implemented	Shared visible-title matcher; the same synthetic bodies go through the public result-check CLI, published-conclusion reader and readiness owner
Intentional instruction changes supply authorization; equivalent and inapplicable assessments retain their own requirements	Implemented	fields_by_verdict; public CLI contract readback and conditional-field regressions

Scope And Continuation

Complete within this scope. The existing review-body module owns the title constants and matcher for both callers, removing duplicate knowledge. Hidden comments and fenced examples cannot supply a visible conclusion. Independent formal-state, exact-head, evidence, reviewer-identity and merge-authority checks remain in their existing owners. No new module, state owner, provider or permissions.

This closes the bounded publication/readback defect. Broader research-page language and visual-quality qualification remains separate; consistent checks do not prove that an analysis is correct or a page is readable.

Validation

• Tested source: d9db6d8348c069f0fbc48ac03c71adbe2d46d7db, target main.
• Run state: finished. Inputs: synthetic and existing public fixtures.

Check	Result	Evidence and boundary
Failing-before regression	Passed	The new cases expose 8 failures on the baseline, with 4 valid-path cases already passing; includes near/missing/opposite/hidden titles and the undeclared conditional field
Review domain regression	Passed	374 tests across result checks, published body, execution contract, GitHub readback/readiness and approval closeout
Public CLI / parity	Passed	The same bodies preserve valid author and independent review paths; incorrect titles fail before publication; original input files remain unchanged
Standard premerge	Passed	6 direct checks and 15 selected executable checks; zero failures or manual holds, including compilation, semantic drift, caller contracts and public-boundary scan
RFC diff and scoped semantic advisory	Passed	No paired RFC changes. No supported vocabulary carrier detected; dynamic semantics still reviewed manually

No remote CI was fetched, polled or awaited under the current Goal's wait_for_ci=false. The unchanged head has a published author-owned COMMENTED approval, a clear approval closeout and ready merge readback. Installed CLI validation adopted merged commit 7f01e00fa8a7951fbb7fe83e877e75c66391e204: policy 24, 15 valid/invalid fixture cases, all four changed production files equal to reviewed source, and live readback of the merged review as valid without a duplicate review action. The existing installer warning about stale promotion-readiness evidence remains; full native App/platform/release qualification is not claimed. Model understanding, independent semantic judgment and visual quality are not inferred from these tests.

Frontend / Visual Evidence

UI impact: none. The changed surface is the existing review CLI contract and its validators; no rendered page, controls, stylesheet or native window changed. Browser screenshots are not applicable to this slice.

Type And Boundary

Bug fix, documentation and focused regression coverage in the existing pull-request-review capability; roadmap S8/S12 engineering delivery. Shared-authority migration/promotion: not applicable.

• Public text and changed files exclude private state, credentials, raw traces and local machine paths.
• Only the bounded review defect and its active callers changed; no benchmark work duplicated.
• No paired RFC changed; bilingual diff check passed.
• Visual impact is explicitly none.
• The commit includes DCO sign-off.

[loopx-agent]

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

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

动机

维护者或作者在提交完整评审、准备合并时会遇到这项检查。

此前，作者评审缺少准确结论标题也能通过提交前检查，直到合并时才被拒绝；现在同一份正文在发布前就得到一致的反馈，可以改正标题后继续原评审。

相同输入已通过真实结果检查命令、发布结论读取和合并检查验证；正确作者结论和独立正式评审仍可使用。

本次范围是既有评审规则及其声明的一致性；检查通过不证明分析语义正确、页面美观、评审独立性或合并授权。

评审基线是 0d55a3d8a22f3dece6a135aa817ed5ec04a26a0a 的 loopx/capabilities/pr_review_queue/README.md，逐项核对 author-owned conclusions 与 decision_text_assessment 两个既有要求。本次完整审查 d9db6d8348c069f0fbc48ac03c71adbe2d46d7db 的七个文件，没有沿用早先批准。

改动思路

复用现有正文 owner 足以统一两个检查入口；保留一份精确标题规则和原正式状态判断，比增加新的评审状态或配置更直接。

本 PR 完成发布前与读回规则一致性及条件字段声明，不扩展 UI、模型语义判断、审批或合并权限。

原有 Python 正文解析器已经处理注释与代码块，两个实际调用方共用它的标题匹配；契约统一声明有意改变指令含义时所需的授权字段。实际 GitHub 身份、正式评审状态和 typed history/readiness authority 保持在原 owner，没有新增状态、provider、配置或通用框架。

同一输入	原提交前检查	当前提交前与读回
作者标题缺失、近似、相反或藏在注释/代码块	可以放过	拒绝；改正原正文后继续
正确作者结论或独立正式评审	各有合法路径	合法路径保持
有意改变指令含义	要求授权字段却未列入契约	原要求明确声明

此表概括合成输入经过真实命令和生产读取方的结果，不证明评审内容真实。

具体改动

• author_owned_review_title_matches 只匹配与英文 verdict 对应的可见精确标题，沿用现有注释及 fenced code 解析；标题常量随规则放入原正文模块。
• check_review_result 对 packet 声明的作者分支检查标题；pr_review._review_conclusion 复用同一规则，仍要求 COMMENTED、当前 head 和完整双语正文。独立评审的正式状态要求未放松。
• _check_decision_text 消费 DECISION_TEXT_ASSESSMENT.fields_by_verdict 中原有的 authorization_basis 要求。equivalent 不额外要授权字段，not_applicable 仍需要范围和理由，其他证据要求没有删除。
• 既有公开 CLI 测试核对契约回读，新增成对场景覆盖两类作者结论、独立评审和错/缺/隐藏标题。README 披露拒绝变化和限制，policy 从23升至24，旧保存结果按当前契约重新准备，已发布历史不被改写。

对主干的风险

最强反例是修复作者标题后误拦独立评审，或只改提交前检查却仍在读回接受隐藏示例。回归直接使用真实 result-check 命令及生产 conclusion/readiness：原生产代码上8个新反例失败、4个合法场景通过；相同 fixture 与测试在本 head 全部12个通过。五个相关套件共374例通过，包含正文、正式状态、旧/错误 head、模型身份、历史和 closeout；输入文件字节保持，无外部发布或模型调用。六个变更 Python 文件的 Ruff 通过。

标准及带 Goal 策略的 premerge 均通过，无失败或人工挂起；包含编译、现有维护性与语义规则、调用方合同及公开边界。质量凭据 cqr_62ed10ba9f974484a35e 精确读回有效。既有 mypy 配置只覆盖未变更文件，本次不运行无关 TS/PG/完整平台矩阵。按当前 wait_for_ci=false 未查询、轮询或等待远端 CI。

本地保存的 packet 不能认证远端作者或证据真伪，发布和合并前仍需新鲜 head、正式状态、讨论和权限检查。本次没有声称模型理解、视觉效果或安装采用已经成立；它们各有独立验收。回滚为一份有界提交，没有数据迁移或自动重写历史。

我的整体评价

APPROVE。该有界修复使原规则在真实调用链中一致，并减少重复知识。未来重构检查已应用到同一正文 owner；进一步迁移语言或增加配置没有当前收益，不纳入本 PR。没有发现当前精确 head 的阻断项。

已检查本 PR 全部生产差异，仅涉及评审命令与正文校验器，没有改动渲染页面、布局、控件或样式。因此视觉评审不适用；研究页实际可读性及中文语义不由这次校验通过来认证。此为作者自己完成的评审，COMMENTED 记录批准结论，不构成独立验收或新授权。合并仍须既有明确授权及该未变 head 的 closeout/readiness。

English verdict: APPROVE — head d9db6d8348c069f0fbc48ac03c71adbe2d46d7db aligns author-conclusion checks before publication and during readback, and declares the existing conditional authorization field. Paired CLI cases fail on baseline and pass on this head; 374 domain regressions, scoped Ruff and premerge pass. Evidence truth, visual quality, independent acceptance and merge authority remain separate.