[cocolord]

Goal And Delivered Outcome

The synthetic delegation Host inherits a locale decoder on Windows CPython 3.11 even though its existing parent transport sends UTF-8 JSON. Unicode input can therefore fail the Host before the independent artifact validator runs, leaving the original-task recovery cases without task_failure.

Configure only the fixture Host's stdin as UTF-8. A File/SQLite regression simulates CP1252 in that child, rejects an invalid artifact, repairs it and revalidates the cached result with the original turn key and exactly one Host invocation. Related to #8360; target main, author baseline 4e949f405528fd8087f0623526c62f890c1ce982.

Author Declaration

• Written by: model_agent — TRAE / GPT-6.
• Specification: docs/architecture/rfcs/monorepo-distribution-split-v0.md §9 V4 and the existing generic Host transport / original-task revalidation contracts at the author baseline.

Criterion	Disposition	Path	Evidence
Consume the existing UTF-8 pipe protocol	implemented	tests/test_local_delegation.py::HOST	Same-environment immutable baseline Host fails the new oracle; corrected Host passes.
Preserve original task identity and avoid a second Host execution during artifact repair	implemented	New File/SQLite regression and tests/test_delegation_task_revalidation.py	Source and isolated-wheel recovery checks below.
Preserve failure and remaining acceptance boundaries	implemented	Paired RFC non-normative checkpoint	Original Windows and Frontstage failures retained; no performance or full-kernel claim.

• Self-check: inspected the actual transport and validation owners, baseline sensitivity, complete recovery suites, package provenance, paired translation and public boundary. The adjacent refactoring pass found no additional extraction warranted: the encoding fix belongs to the existing source-owned Host fixture. No production rule, persisted state, protocol vocabulary or default Host environment changes.

Scope And Continuation

Complete fixture correction with a durable regression and paired qualification record. The parent minimal-kernel program stays open for complete provider-first transaction, PostgreSQL host/wheel, complete costs and adoption acceptance. Retain I3 for at least one minor release and all historical/frozen failures. #8360 retains its own exact-head review/CI gates. Native Windows verification of this new head remains pending; the prior failed run is retained. This proposal does not merge or upgrade an installed Host.

Validation

• Tested fixture commit: a4ce613585fdb329ba3600f0ede768864214ed84; final paired-doc head: 107c4fab80406a96bb0f6ccfd62b47230034d2fc.
• Run state: finished locally; CI pending after submission.
• Input classes: synthetic, public_fixture.

Check kind	Result	Public-safe evidence / limitation
regression_parity	passed	CPython 3.13.5 / Node 22.22.3 on macOS arm64: replacing only the corrected fixture HOST with the immutable baseline makes the Unicode task-validation oracle fail. The corrected File/SQLite cases both pass. Earlier invalid probe preparations are retained separately.
integration / real_entrypoint	passed	tests/test_local_delegation.py plus tests/test_delegation_task_revalidation.py, CPython 3.13.5 / Node 22.22.3, 81 cases pass using temporary File/SQLite stores and synthetic subprocess Hosts.
integration / real_entrypoint	passed	CPython 3.11.15 / Node 22.22.3 source: new two provider cases plus all 12 existing original-task recovery cases pass, including MCP disconnection and lock-timeout recovery.
integration / real_entrypoint	passed	Fresh isolated monolithic wheel, CPython 3.11.15 / Node 22.22.3: the same 14 cases pass. Wheel imported before source-owned fixtures; all 1,920 present packaged tracked runtime entries match the source. The same 17 source-only exclusions remain: canary baseline, host plugin/config, fault-test helper and documentation/media.
static	passed	Ruff and paired staged RFC check; diff advisory introduces no shared vocabulary. Public-boundary subcheck passes for the three candidate files. Full local loopx check still fails on unrelated registered-Goal authority errors; private diagnostic contents are excluded.
static / integration	passed	Standard diff-derived loopx canary premerge --from-git-diff --git-diff-base origin/main: all five direct and 11 selected checks complete, zero blocking failures. The unchanged module-metric ceiling finding in loopx/extensions/lark/private_conversations.py is reported as an inherited advisory. The first attempt failed because this new worktree lacked TypeScript dev dependencies; npm ci --ignore-scripts prepared them, and the unchanged-head rerun passes. No gate was relaxed.
integration	not_run	New exact-head native Windows CI pending. The original #8360 Windows arm has six recovery failures. Its independently retried Frontstage manifest request still returns HTTP 504. #8360 also retains separate digest-declaration and selected-owner loading failures; this fixture correction does not resolve those gates.

The changed fixture is exercised through the real generic Host subprocess transport and real temporary File/SQLite providers. No model or real-agent execution, PostgreSQL host/wheel qualification, performance campaign, split installation or installed adoption is claimed. Existing recovery and failure classification remain intact.

Frontend / Visual Evidence

UI impact: none. No UI or documentation chrome changes; screenshots and preview are not applicable.

LoopX Area

• Host or runtime integration
• Public docs or presentation surface (RFC evidence only)

Shared-authority RFC fixture impact

N/A: no TypeScript migration or shared-authority production change.

Boundary Checklist

• Public-safe diff/body; no private state, credentials, raw traces, internal links or local paths.
• Scoped fixture repair; no competing implementation or benchmark work.
• Both English/Chinese RFC files change together.
• UI impact marked none.
• Every commit includes a DCO sign-off.

[mergify]

This pull request has merge conflicts with main and cannot be merged
until they are resolved. Please rebase or merge the base branch, @cocolord.

Choose the remote for the base repository, not an out-of-date fork.
For a fork clone, first inspect git remote -v; upstream must point
to https://github.com/loopx-project/loopx.git. If it is absent, add it
with git remote add upstream https://github.com/loopx-project/loopx.git.
Then run:

git fetch upstream
git rebase upstream/main
# Resolve each conflict, git add the resolved files, then git rebase --continue.
git push --force-with-lease origin HEAD

For a same-repository clone whose origin points to
https://github.com/loopx-project/loopx.git, use origin instead of
upstream for fetch/rebase. If you prefer merging the base, use
git merge <base-remote>/main and push normally.

Keep the DCO Signed-off-by trailer on every commit when you rebase.
https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[loopx-agent]

Reviewer: model_agent | gpt-6.1-sol | OpenAI | runtime_reported | reasoning_effort=max

APPROVE。已验证合成 Host 的两条恢复路径，整合时仍需处理双语 RFC 冲突。

精确 head 107c4fa

动机 / Motivation

维护原任务恢复的开发者，需要让测试 Host 先正确读到请求，再验证无效产物能否在原轮次重验。父进程通过管道发 UTF-8 JSON。若测试 Host 用 CP1252 读取，Unicode 内容会在验证前出错。显式设定 UTF-8 后，同一请求可到达产物验证。File 和 SQLite 均验证了失败分类、修复产物后接受原轮次结果，以及 Host 只执行一次。本切片覆盖合成 Host 和本地恢复验证。完整 V4、原生 Windows、拆分安装与成本仍待独立验收。目标 main 的两份 RFC 发生冲突，需保留两边的证据条目后重新验证新提交。

改动思路 / Approach

真实 generic Host 管道已明确发送 UTF-8。修复落在现有测试脚本中，避免改动生产解码和失败分类。回归先把合成子进程切成 CP1252，再送入原来的 UTF-8 请求和有意无效的产物。恢复沿用既有 validator 和缓存结果入口。

具体改动 / Changes

完整差异只有双语 RFC 与既有测试。测试 Host 在 json.load 前配置 stdin 编码。File、SQLite 用例分别断言拒绝中有 revalidate_cached_result，修复原产物后接受结果，轮次标识和 Host 调用次数不变。两份 RFC 保留原 Windows 和 Frontstage 失败，也明确了作者 source/wheel 证据的范围。

关键内容讲解

1. 合成 Host 在读取 JSON 前设置 UTF-8。父进程原有的 UTF-8 编码和真实子进程边界保留。 合成 Host L43

2. 双 provider 回归先制造可识别的失败，再走缓存重验。把旧 Host 字符串换回同一套测试时，两个用例均在缺少 task_failure 的断言处失败。 双 provider 回归 L104

规格映射 V4 的完整验收 deferred，Fixture-UTF8 implemented。。依据不可变版本 4e949f4。 docs/architecture/rfcs/monorepo-distribution-split-v0.md

对主干的风险 / Risks

本地 CPython 3.11.17 两套完整回归 81 passed。仅置换不可变 base 的合成 Host 时，两条新用例按预期失败。候选的测试覆盖了真正的进程传输和 File/SQLite 状态，测试没有代填 validator 结果。原生 premerge 通过。Lark 模块上限的继承 advisory 在相同 base/head 输出一致，原阈值保持。

实际合并基线为 4e949f4。对本次获取的目标 main 做 Git 三方检验，两份 RFC 均有内容冲突，测试文件无冲突。原生 Windows 和独立 wheel 资格仍属于各自验证范围。

我的整体评价 / Verdict

这个切片可独立恢复合成 Host 的测试覆盖。#8360 当前也携带相同的 Host 修复和测试，整合时应只保留一份。#8360 还包含独立的加载与浏览器中断变更。双语文档冲突属于整合门。 关联提案 #8360 长程影响判为 improved，恢复验证能覆盖原用例。用户体验不适用，既有 CLI 与页面未改。变更限于测试和证据，无页面展示改动。未验证原生 Windows、安装采用及长期成本。

English verdict: APPROVE — 107c4fa. The synthetic Host keeps UTF-8 task recovery reachable.