[cocolord]

The standalone core package lane ran the distribution suite but omitted the installed context suite, and a context-test-only change did not trigger the lane. Add that existing suite to both PR/main path filters and run both suites against the separately built and installed core wheel. The paired RFC checkpoint records the coverage repair and the limits of the complete-Turn loading/network observations.

Validation:

• Built and independently installed core 2.3.0 with Python 3.11.15. All 1,885 tracked runtime files match the source bytes.
• Ran the exact combined pytest selection against that installed interpreter: 20 passed in 107.79 seconds. The source test driver used Python 3.13.5; the remote lane retains Python 3.11 and Node 26.
• An independent parsed-workflow check rejects the unchanged base for both missing triggers and the omitted suite; it accepts the repaired filters and command.
• Staged bilingual RFC and diff checks passed. The public-boundary scan completed cleanly for all three changed files; the broader local Goal check reports unrelated legacy canonical-authority diagnostics.
• Exact-head standard premerge passed: four direct checks and eleven selected checks, with no failures or tracked side effects.
• Current remote kernel-static-checks fails the unchanged F811 at tests/test_lark_private_conversations.py:2816. Exact base/head source blobs and the same Ruff command's diagnostics match the remote failure. Existing test(lark): restore lint for the imported transcript fixture #8657 owns that repair; other remote gates and merge readiness remain separately recorded holds.

The runtime, context protocol and defaults are unchanged. Native context events and the installed task Host remain fixtures: these checks do not establish real model retention/adoption, full R1 qualification or complete cost improvement. The local network observation identifies loopback Effect RPC traffic in the fixture, with observer controls; it is not an OS-wide or ordinary-telemetry guarantee. No UI changes or release/install rollout are included.

Adjacent refactor: unnecessary for this correction. The existing single core job and installed-context owner already provide the required boundary.

[cocolord]

Reviewer: model_agent · GPT-6 · OpenAI · self_reported

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

评审提交：8720bdf71d47ae124dd0ceff4945c8f3579a754c。结论：APPROVE。安装上下文测试的触发与执行遗漏已被独立对照证实，这个提交在原有 package job 中补齐了完整覆盖。

动机 / Motivation

维护独立 core 安装与上下文恢复的贡献者。此前仅修改安装上下文测试时，package lane 不会触发；其他变更触发时，它也只运行 distribution suite。修复后，两种触发过滤器包含该文件，独立 wheel lane 同时执行两套完整测试。安装上下文断言的变化进入独立 wheel 的自动回归。本 PR 的交付边界是安装协议的 CI 覆盖。真实宿主的初始、修订、压缩和丢 ACK 采用及完整 R1 成本资格仍待验收。

改动思路 / Approach

复用现有独立 core job、构建安装顺序与两套已交付的完整断言。改动仅增加一个文件的两处触发匹配与一次 pytest 参数，不另建安装器、上下文 owner 或状态记录。保持单一 CI 入口的方案相称；只补运行参数会遗漏 context-only 变更，只补触发器则仍不执行上下文断言，所以三个点需要一起修复。测试失败继续由原 pytest、Actions 和 merge gate 处理，十分钟 job 预算及产品默认值保持原状。

既有 workflow 与测试已提供完整边界，无需新增抽象。本次边界为 CI 覆盖加配对证据 ledger。

具体改动 / Changes

依据 docs/architecture/rfcs/monorepo-distribution-split-v0.md，spec_revision 为 f6cde42。V5 的完整宿主验收继续 deferred：此处只将已交付安装协议的回归接入 CI。V10 完整配对成本属于 out_of_scope；新增观察明确限制为合成任务与关闭遥测的本地诊断。R1 完整退出条件继续 deferred，由既有真实宿主、D2/D5 组合与成本资格推进，不能用本次检查通过替代。

关键代码讲解

• .github/workflows/package-smoke.yml:10：PR 和 main push 的路径列表均加入安装上下文测试，覆盖只有该文件变化的输入；已存在的其他包路径和事件保持原匹配。
• .github/workflows/package-smoke.yml:62：同一 core job 在独立构建与安装之后，用同一个 LOOPX_CORE_TEST_PYTHON 执行 distribution 与 context 两套完整测试，保持源码测试驱动和实际安装产物的分离。

解析 workflow 的独立对照在原 base 上报出三处遗漏，在 exact head 上通过。重新构建的 Python 3.11.15 core 有 1,885 个 tracked runtime 文件，与源码字节一致；使用该安装解释器执行两套组合测试，20 项通过，107.79 秒。源码测试驱动为 Python 3.13.5，远端 lane 继续使用 Python 3.11/Node 26。测试中的真实 File/SQLite、独立产物验证和重放断言有效；原生事件与 Host 仍为协议夹具。

对主干的风险 / Risks

主要风险是让测试名字出现在 CI 里，却仍从源码环境调用运行时，或让 context-only 变更逃过触发。既有 fixture 的隔离解释器、实际 wheel 文件回读以及本次 workflow 对照分别覆盖这两条路径。产品 runtime、tests、packages、apps 和 skills 在 base/head 无差异，关闭上下文功能时的原值、部分交付/丢 ACK、修订、PostCompact 与捕获字节变化拒绝均由原安装测试保护。

PASS：组合安装测试 20 项；exact-head premerge 的 4 项直接检查和 11 项选定检查；双语、DCO、diff 与公开边界。完整 Goal 检查另外报告旧 canonical-authority 诊断，公开文件扫描自身为通过，未修改这些私有状态。远端 kernel-static-checks 的 F811 位于未改动的 tests/test_lark_private_conversations.py:2816；base/head 同一 Ruff 命令诊断完全一致，文件 blob 同为 b3e1d8cc598ec63b4bd24d5b1bf28f2a3367f2a9，匹配当前远端日志。其他 gate 失败与待完成项目仍单独记录，不把远端失败写成通过。既有 #8657 负责该 lint 修复。增加的一项上下文测试沿用现有 job，回退可直接 revert 本提交。

我的整体评价 / Verdict

这是完整、可独立验证且可回退的 CI 覆盖修复，采用 justified_increment。long_horizon 与 user_experience 均为 preserved：运行时和用户操作轮次未改变，原已交付的安装恢复检查开始自动运行，未声称真实模型效果提升。视图检查为 not_applicable；检查过的 diff 只涉及 CI 与非渲染 RFC ledger，apps/产品代码未变，视觉未验证范围为 none within this CI change。新语义复用既有测试与局部 workflow 选择，不引入共享词汇或第二决策 owner。相邻重构无需额外机制；完整 R1、真实宿主采用与完整成本继续开放。

English verdict: APPROVE at 8720bdf. Both context-test triggers and the installed pytest selection are corrected; 20 independent installed checks and premerge pass. Preserve the separate CI/readiness hold and real-host qualification boundary.