[loopx-agent]

The 2.2 notes now teach the supported default-on context path, progressive reads through the original recovery reference, exact-cycle deltas, and explicit App/benchmark opt-out. Existing entries also cover the optional subscription-first API fallback and independent conversation recipients, with activation, verification, disable/rollback and authority/privacy guidance in both languages.

Community attribution is reconciled through the current main range: 180 merged PRs from 21 non-founder contributors plus 12 verified direct commits. Every original English/Chinese credit row is retained; the added claims cite their actual merged PRs.

Merge dependency: #7815 must be merged before this documentation PR. This proposal updates release instructions; it does not publish a tag, package, release or stable channel. Formal release qualification still runs on the final frozen source.

Validation at 9bc6ffe10eb5a01998cb873d39f0af06e84310b2: all 16 bilingual lifecycle entries pass the exact-body validator; the contributor delta is checked against merge metadata and Git ancestry; public scans, diff and bilingual gates pass. Standard Goal premerge passes 4 direct checks and 17 selected checks with no failure or manual hold. Exact-scope receipt cqr_c21b3e0e2164d16837e8 is valid. No runtime behavior or configuration is changed.

Final publication addendum — 2026-11-01

LoopX 2.2.0 is published from 2c3524857e8a3dd553335b79444333786f492c43. The original proposal and validation above remain tied to their recorded head. The final tag range expands community attribution to 185 merged PRs from 22 non-founder contributors plus 12 verified direct commits; the complete bilingual release body carries every credited person and concrete PR attribution.

The final cut includes the reviewed sessionless recovery and scoped configuration-read fixes #7833, #7838. Testing-helper #7831 and the separate blog proposal #7829 are excluded. The final source passed 89 critical tests with 8 explicit native opt-in skips, isolated 2.1.1-to-2.2.0 upgrade preservation and installed Host checks. Complete local pytest/model matrices are unqualified; no earlier failure is reclassified as passing. Canonical package/PyPI and desktop channels retain independent workflow/readback status in RELEASE-QUALIFICATION.json.

正式版本已发布。最终范围为 22 位非创始社区贡献者、185 个已合并 PR、12 个另行核验的直接提交；逐位双语归因见正式发布说明。保留原提案 head 的历史验证；最终版本采用有界验证，未完成的全量／模型矩阵及此前失败不计为通过。安装制品、PyPI 与桌面渠道的实际就绪状态分别读回。

Channel closeout — 2026-11-01

Publication recovery #7847 resumed the original validated build without moving the product tag or replacing any of its nine immutable assets. The publication-only workflow succeeded, including PyPI Trusted Publishing. Public PyPI wheel/source hashes match the original build; isolated public-index installation, version, deep doctor, Host skills and frontend-resource checks passed. The source stable channel is at the same frozen product commit and its public installation was verified. Desktop installers and the Darwin signed stable feed were verified separately. The bounded validation and earlier unqualified/failed matrix results above remain unchanged.

渠道已收口：GitHub、PyPI、桌面安装包、macOS 签名更新与源码 stable 均已发布并核验。恢复发布复用原制品，产品 tag 和九个不可变附件保持不变；公开安装和实际版本回读通过。历史失败与完整矩阵未合格范围继续保留。

[loopx-agent]

Reviewer: model_agent — gpt-6.1-sol / OpenAI — runtime_reported, effort=max.

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

动机 / Motivation

准备升级 2.2 的 Codex App／benchmark 用户，以及负责发布与贡献者归属的维护者。

旧说明把 context delivery 写成默认关闭，并让省略开关充当完整交付对照；新 heartbeat 默认开启后，这会误导希望关闭的用户。新版明确要求显式 off，并解释删除本机配置会恢复默认开启。

同一份中英说明现在可以找到默认值、聚焦首读、原 recovery 引用、显式关闭，以及 16 项能力的启用、验证、回退和权限说明。

本 PR 仅修订发布文档，不改变运行时、不启用付费 API，也不发布包或移动 stable。

正式发布仍须包含 #7815 和本 PR 的同一冻结源码，完成原定完整资格验证、安装升级与远端读回。

改动思路 / Approach

继续使用既有发布说明和 16 项能力条目，不建立第二套配置或发布指南。默认值、聚焦首读、原 recovery 引用、批量选读和显式关闭放回各自的使用路径；API 按量付费仍是独立的默认关闭选项。仅改一句默认值会留下回退和对照实验说明的冲突，因此这次同步修改对应的中英文启用、验证和停用段落。

我通过真实 CLI 核对了文中的读取方式：已知地址一次批量读取，陌生父节点展开两层目录。对固定选择的续页改地址、以及替换捕获字节，都会拒绝；这些检查没有执行 guard 或重新准入。它们证明命令与完整性边界，没有证明模型保留或任务收益。

具体改动 / Changes

关键内容讲解

审查 head 为 9bc6ffe10eb5a01998cb873d39f0af06e84310b2。修改仅在 docs/product/releases/v2.2.0.md，分类为 docs，+102/-40。规范依据为 docs/product/release-readiness.md，固定 revision d430c0396017bc9d53350e22126407f30b8bd653，按原有条款逐项核对：

条款	结论与依据
release-decision	implemented：升级对象、默认变化、验证和维护者在中英摘要一致。
product-groups	implemented：保留产品优先的既有分组，新增能力描述均链接对应 PR。
contributor-attribution	implemented：旧范围证据加六个实际合入 PR，核对 merge OID 祖先关系；180 个社区 PR、21 人和 12 个直接提交，原 42 行中英归属全部保留。
bilingual-lifecycle	implemented：16 对条目具备启用、验证、停用／回退、权限／隐私和版本化文档；精确正文检查通过。
final-publication	out_of_scope：这是文档提案；最终正文、冻结源码和远端发布读回仍由原发布工作执行。

来源校验引用仍保留完整摘要，页面不重复显示摘要不等于省略校验。更换选读地址要回到原 recovery 引用；删除本机配置会恢复新的开启默认值，不能用来关闭。API key 的添加、连接检查、付费启用和撤销也分别说明，避免把连接成功写成推理额度证明。

对主干的风险 / Risks

主要风险是说明先于实现合入，或字段齐全但命令及贡献者归属错误。PR 已明确 #7815 必须先合入；这是维护者需要执行的合入条件，描述本身不会形成 GitHub 自动阻断。本次审查没有发现需修改当前文档的阻断项。

PASS：精确正文的 16 项双语生命周期检查、真实 reader 正反例、贡献者范围与旧行保留检查、公开边界扫描、diff／双语门均通过。Goal premerge 的 4 个直接检查和 17 个选中检查全部通过，无 manual hold；CQR cqr_c21b3e0e2164d16837e8 匹配当前范围。CI 按 wait_for_ci=false 未查询。原始验证入口及观察假设的失败保留在记录中，没有改产品或放宽预算来取得通过。

我的整体评价 / Verdict

这是 justified_increment：完整修正可审阅的使用说明和归属，并保留原发布任务的冻结源码验证与发布读回。long_horizon 为 preserved：当前来源、原 Turn 和恢复责任保留，不把文件存在当成保留或收益；user_experience 为 improved：已知路径和关闭路径都能直接找到，减少误操作与重复发现。

本 PR 只修改 docs/product/releases/v2.2.0.md 的 Markdown 正文、链接和命令示例，没有组件、样式、导航或渲染实现变化。

正式 GitHub Release 页面的展示及已发布客户端不由本次文档审查验收。

单文件边界可独立回退，没有新增运行时词表、状态、付费调用或权限。文档审查结论为通过，合入顺序与正式发布资格仍须按上述条件分别核验。

English verdict: APPROVE — 9bc6ffe; bilingual guidance, attribution and reader boundaries verified. Merge after #7815; formal release qualification remains separate.