[CTYQAQ]

Goal And Delivered Outcome

• Outcome basis / optional anchor: no written specification; the request in this PR is the basis. loopx/extensions/lark/event_collector_runtime.py already read the provider's stdout lines by these prefixes, so it is the owner the other two sites now import from.
• Goal/source and gap: goal_topic_runtime.py (_EVENT_READY_PREFIX, _EVENT_DIAGNOSTIC_PREFIX) and team_plan_confirmation.py (_EVENT_DIAGNOSTIC_PREFIX) compiled the same two bodies locally. A wire-format change had three sites to find, and nothing failed if one of them drifted.
• Observable before → after, with the validation row that proves it: a literal census over loopx/**/*.py finds "[event] ready " in 2 modules and "[event] " in 3 before, and only in the owner after; re-adding a local copy fails test_only_the_owner_states_each_event_line_prefix.
• Issue/task and intended base: no issue; base main at 53dcb03082985bf894b1fca8813ab9f6e92503c5.

Author Declaration

• Written by: model_agent (Qoder); operator CTYQAQ

Implemented against

• Specification and revision: no written specification; the request in this PR is the basis. The owner module and both parse blocks are read at 53dcb03082985bf894b1fca8813ab9f6e92503c5.
• Criteria:

Criterion (spec clause)	Disposition	Symbol / path	Test or command
each event-line prefix is stated once, in the owner	implemented	event_collector_runtime.EVENT_READY_PREFIX, .EVENT_DIAGNOSTIC_PREFIX	literal census over loopx/**/*.py
every consumer holds the owner's object	implemented	goal_topic_runtime, team_plan_confirmation	test_every_consumer_holds_the_owners_object
a new local copy fails closed	implemented	new census test	mutation row below
fold the "[event] exited " token into the owner too	out_of_scope	goal_topic_runtime._EVENT_EXIT_REASON	one consumer, paired with a reason regex in the same block; left for a change that needs it

• Self-check before submission: read both parse blocks and preserved their branch order (ready, exited, then diagnostic); changed no wire value; ran the affected suites, the architecture gates and ruff; re-added a local copy to prove the census fails.

Scope And Continuation

• Completed scope and remaining work: complete within this scope.
• Slice boundary / successor: the exited token stays local per the out_of_scope row; no follow-up is needed unless that line gains a second consumer.

Validation

• Tested revision: 351412e
• Run state: finished
• Input classes: synthetic, public_fixture

Check kind	Result	Public-safe evidence / limitation
static	passed	python -m ruff check on the two changed files and the new guard
unit	passed	tests/extensions/test_lark_goal_topic_runtime.py, test_lark_realtime_dispatch.py, test_lark_event_collector_runtime.py — 128 passed
unit	passed	tests/extensions/test_lark_team_plan_confirmation.py — 6 passed; tests/test_steward_team_plan_preview.py + tests/architecture/test_lark_identity_shape_owner.py — 102 passed
unit	passed	tests/test_lark_private_conversations.py — 234 passed
regression_parity	passed	re-adding _EVENT_DIAGNOSTIC_PREFIX = "[event] " in team_plan_confirmation.py fails the census (1 failed, 1 passed); restored tree passes
integration	passed	tests/architecture/test_top_level_module_budget.py + tests/test_stage2c_dependency_lock.py + the new guard — 9 passed

• Coverage and gaps: the changed blocks are the stdout line filters exercised by the goal-topic and collector suites; no new branch was introduced.
• Rebase note: rebased onto b6db7f7c415b356ac94ebfc643977d381a314eb9 after submission; the branch patch-id is unchanged (afe06181…), and the suites were re-run on the rebased tree (245 passed together). The first rebased run of tests/test_lark_private_conversations.py reported 233 passed + 1 failure in test_app_long_return_keeps_all_parts_and_does_not_repeat_identical_parts; a repeat run of the same file passed 234/234, that case passes 3/3 in isolation, and no commit in the rebase range touches loopx/extensions/lark/ or that file. It is a pre-existing timing flake, also red in the 2026-10-27 scheduled run on main (run 40604536330).

Frontend / Visual Evidence

N/A — no user-visible surface changed.

[loopx-agent]

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

未发现当前精确 head 351412e 的阻断问题，结论为 APPROVE。

动机 / Motivation

维护 Lark 消息和团队计划回调监听器的开发者。

同一提供方的就绪与诊断行前缀原来分散在三个模块；现在两个监听器读取既有共同定义，原就绪判断、消息路由和失败恢复保持。

已验证真实子进程事件流的就绪、断线和回调恢复不变，并删除三处重复字面量。

本次不改变 Lark wire format、身份权限、回调批准权或外部消息发送。 基线已有正确wire值；本次价值在于消除会随格式更新漂移的重复知识，不冒称修复了线上事件丢失或完成了真实账户接入。

改动思路 / Approach

这两个值是既有提供方 stdout framing 的知识，复用现有 event_collector_runtime 即可；路由、确认与重试继续由原 owner 决定。

本 PR 完成既有两个监听器的前缀归一及相称回归，不增加状态、provider 或第二个解析器。

最小方案正是从已拥有相同值的模块导入，而不是新建protocol模块、listener或配置。消息流仍先识别ready，再识别exited，最后忽略一般diagnostic；team-plan回调仍先忽略diagnostic，再解析JSON/schema及交给原handler。单消费者exited字串及reason regex留在其原逻辑块合理，无需把相邻内容都抽走。这个直接关系用文字足够，不额外绘图。

具体改动 / Changes

没有针对本次常量整理的独立书面规格；依据明确的维护结果和原消费者行为验证，不把作者新增census当作规范。不可变比较基线为 b6db7f7。完整3文件 +64/-6：两生产文件 +5/-6，新增架构测试59行。既有event_collector_runtime本身未改，导入依赖和实际消费者均已检查。

关键代码讲解

• EVENT_READY_PREFIX / EVENT_DIAGNOSTIC_PREFIX 是原提供方framing定义；字符串值未变。它们只描述stdout行，不能授予消息执行、批准或发送权限。
• stream_lark_goal_topic_profile 保留profile选择、ready先于diagnostic的顺序、事件分发与最终close/wait。子进程存在不能当作就绪；signal或未知退出仍由原逻辑报source_disconnected。
• start_team_plan_review_callback_stream 保留绑定chat的过滤、schema检查和handler异常隔离；只是消费共同diagnostic前缀，没有新decoder或持久状态。

两生产文件在移除明确的常量赋值/import差别并归一私有别名后，其余AST完全相同。新增测试做有界literal普查和对象来源断言；它只能覆盖所枚举的字面写法，不能独自证明所有等价拼写或真实运行。

对主干的风险 / Risks

最强反例是ready被更宽diagnostic前缀吞掉，或诊断/回调错误阻断后续消息。独立19场景在不可变base与当前head完整相同：17条真实子进程/stdout链、2条无配置不启动控制，涵盖ready有无、真实未匹配业务事件、四类退出、非法JSON/schema、handler首次拒绝后的合法后继及线程/子进程回收。保留完整健康变化、返回值、输出、callback正文和argv，无归一化。provider内容与handler是合成边界，没有连接真实Lark或调用模型。

在导入消费者前故意更换共同ready值，四个仅凭ready就绪的独立oracle失败，而当前head通过；不能只凭对象相等或census绿灯断言重构等价。实际callback首次handler拒绝后第二条仍被处理，close后线程结束且子进程已reap。

PASS：十个相关/架构模块479项，零skip；不可变base六个旧相邻模块236项。Ruff、配置17文件mypy、完整TS typecheck、diff/DCO/双语差分通过。标准premerge 6项direct与10项selected全部通过，无失败/跳过/人工hold；精确CQR有效。未查询或等待CI。

作者已披露长回传测试历史一次失败。本轮当前完整模块通过，另对同一具名case在base/head各运行3次均通过，原测试与处理逻辑未变；这不能改写旧失败或认证长期flake已修复。没有用放宽断言、删除有效覆盖或只保留最后一次成功来取得资格。

语义与 CI 对齐

复用既有Lark provider语法，不新增Core状态词汇、权限或默认listener。机器判断及原异常恢复仍是原逻辑，不把硬行为称为可选建议。差分语义advisory无支持的新carrier，真实stream对照及mutation才是本次行为证据；更广的身份/确认/持久作用域由原owner与现有测试保持。

我的整体评价 / Verdict

APPROVE，当前维护结果完整且相称。long_horizon 与 user_experience 均为 preserved：真实管道的错误恢复、后续callback和回收不变，原用户不增加配置、复制或确认步骤。移除重复知识有明确消费者，不靠PR或测试数量证明长期收益。

这两个值是既有提供方 stdout framing 的知识，复用现有 event_collector_runtime 即可；路由、确认与重试继续由原 owner 决定。 本 PR 完成既有两个监听器的前缀归一及相称回归，不增加状态、provider 或第二个解析器。

有界整理已应用，单消费者exited块保持原位，不需要额外框架或人为后继。已检查 Lark 消息与回调的现有入口；仅改变内部常量来源，没有卡片、页面或布局变化，视觉评审不适用。 真实 Lark 网络与个人账户、正式安装、原生 Windows 和长期运行未在本轮验证。 本评审不认证上述外部采用，不授予合并、外发或安装权限。

English verdict: APPROVE — 351412e. All19 real-pipe scenarios match the immutable base; a ready-prefix mutation fails four independent cases. 479 Python tests, static checks, standard premerge and exact-scope CQR pass. The disclosed historical timing failure and live-provider qualification remain separate.