Skip to content

[PR] fix: route codex context-only prewarms to the probe relay via the upstream proxy - #4595

Open
z0d0dOr wants to merge 23 commits into
ranxianglei:masterfrom
z0d0dOr:2026-10-24_codex-context-prewarm-probe
Open

z0d0dOr wants to merge 23 commits into
ranxianglei:masterfrom
z0d0dOr:2026-10-24_codex-context-prewarm-probe

Conversation

@z0d0dOr

@z0d0dOr z0d0dOr commented Oct 24, 2026

Copy link
Copy Markdown
Contributor

(ai: anthropic/claude-opus-5-5)

Fixes #4594

What

  • src/responses-ws.ts: a generate:false request that replays no conversation turn (empty input, or only developer/system/user messages) is a prewarm probe. It is forwarded verbatim on the upstream transport and never run as an ACP turn. Before, only empty input matched, so codex's startup / idle-thread warmup (PrewarmInput::Base: its context items plus generate:false) became a zero-message turn. That turn deactivated every fold (and persisted it), re-injected the whole codexForgedSummaries list, and added ACP tools.
  • The probe relay now passes the relay's upstream proxy (proxyDispatcher(egressProxyFor(url))). ws-bridge exposes the resolved proxy (egressProxyFor) instead of the boolean isEgressProxied; the tunnel guard derives its boolean from the same call, so both share one decision.
  • History prewarms (codex prewarm_with_history, which replay real turns) are unchanged and stay on the pipeline.

Why

Details in #4594. In the field, a machine restart made every resumed codex thread send a two-item prewarm. Seven sessions were saved with 0 active blocks. On a long session the re-injected forged summaries (~1.2M tokens) turned the prewarm into a 413, which made codex force a pre-turn compaction. The gate (0 active blocks) then forwarded that compaction verbatim (17.5 MB), and the upstream closed the WebSocket with 1009. This is also the trigger shape in #4036; #4040 (gate judged on the request's own view) stays complementary.

Tests

  • tests/codex-responses-ws.test.ts (12/12):
    • Context-only generate:false prewarm on a session with an active fold and a forged summary: upstream receives the frame verbatim (no forged summary, no ACP tools), stats.requests is unchanged, and the fold stays active. Red before the fix: the prewarm carried bili's developer message with the forged summary.
    • Negative: a history-replaying generate:false prewarm still rides the pipeline (ACP tools injected, counted as a request). Green before and after.
    • The probe egresses through a configured upstream proxy (CONNECT observed). Red before the fix: 0 CONNECTs.
    • The fake upstream now answers any generate:false frame with an empty terminal, as the real upstream does.
  • Related suites green locally (139/139): responses-ws-http-fallback (including "nonempty generate:false warmup … the continuation merges the developer message"), responses-prewarm-generate, responses-websocket, ws-bridge, ws-upgrade, tunnel-guard, upstream-proxy-routing, skipped-proxy-hint, native-ws-intercept, issue3701-ws-byte-budget.
  • npm run typecheck and npm run build are clean.
  • Real-client A/B with codex 0.164.0 (exec, then exec resume --last) against a fake WebSocket upstream:
master this PR
startup / resume prewarm processTurn: 0 msgs + degenerate-turn, 36.7→52.8 KiB, tools 9→14 relayed verbatim (tools 9), no pipeline turn
real turns 70.9 / 71.6 KiB out identical

Scope

No config, persistence, kernel or message-shape change, and no new env var. Not addressed here: codexForgedSummaries is append-only and is re-injected whole whenever the echo is missing; the WebSocket close-1009 oversize error is not relayed to the client (it waits for its idle timeout) and reports 0.0 MiB for non-string bodies.

🤖 Generated with Claude Code

@ranxianglei ranxianglei changed the title fix: route codex context-only prewarms to the probe relay via the upstream proxy [PR] fix: route codex context-only prewarms to the probe relay via the upstream proxy Oct 24, 2026
@ranxianglei
ranxianglei force-pushed the 2026-10-24_codex-context-prewarm-probe branch from cdda1b3 to e425e78 Compare October 24, 2026 18:07
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

查重: 不重复。检索 open+closed issue/PR(prewarm、generate:false、probe relay、fold 失活、forged summaries 等关键词)无同机制条目;#4036/#4040 是 #4594 已引用的前案(同症状家族——本 PR 修的是根机制,#4040「gate 用请求自身视图判定」保持互补)。

标签: bug / codex / scope/wiring / p1 —— 重启后所有 codex 会话折叠状态被清零(73→0 等七例)+ 长会话成本爆炸,属 owner 修订 p1 类①(状态丢失);#4594 与本 PR 均已打上。

审核进行中,初步结论:

  1. 方向正确。根因确为「只带上下文的 generate:false 帧被当成 pipeline turn」:零对话 turn 会 deactivate 全部 fold 并持久化、重注入整份 forged summaries、再给一个不产出的请求加 ACP 工具。修在决策点(gate 谓词),不是传输层遮症状。
  2. gate 谓词审过:isPrewarmProbe 接受空 input 或仅 developer/system/user item;空数组过 .every,旧的空 input 探针行为被包含;含 assistant/tool 的历史回放帧仍走 pipeline(负向测试钉住)。两个 codec(opencode-v2 插件车道与 codex 前缀车道)共用此 session 类,gate 放宽同时保护 OC V2 车道免于同样病理;relay 路径不碰会话状态(无 processTurn、fold 不变),Support Codex Responses WS clients with HTTP/SSE upstreams, preserving local delta checkpoints #3342 的 checkpoint commit 保留。
  3. 代理接线审过:proxyDispatcher(undefined) 返回 undefined → 直连行为不变;配置了 proxy 时复用带超时/keepalive/CA 策略的缓存 ProxyAgent,与其它出站路径同一套策略;ws-bridge 隧道守卫的布尔值改从同一个 resolveProxyDecision 调用推导,单一决策源,语义等价。
  4. base 过期(cd92b0b,落后 master 16 提交):漂移面只有 src/server.ts(+47/−6,preflight 区),与本 PR hunk(:709)无重叠。已在本地 rebase 到当前 master: typecheck 干净,定向套件 151/151 绿(codex-responses-ws + responses-ws-http-fallback + responses-prewarm-generate + responses-websocket + ws-bridge + ws-upgrade + tunnel-guard + upstream-proxy-routing + skipped-proxy-hint + native-ws-intercept + issue3701-ws-byte-budget)。

下一步:推送 rebase 后的分支刷新 PR head,看新 head 的 CI(master 测试分片此前因 #4555 红、修复在 #4569,需确认当前 master 已含),然后按八问给最终结论。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

PR #4595 修复完成(新 head e425e78):3 红(config-surface-guard/lock-sync/docs-sync)= 旧基线 cd92b0b 漂移(master 4d34abf 已修),rebase 到 146b34f 解决;第 4 红 unused-exports 的 ClientBuildCheck 为本侧 born-red,按 #45598042523 同款一词 de-export 独立 commit 修掉。本地 4 门禁 + typecheck + 定向测试全绿,等 CI 重触发。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

查重:不重复——4 个红全部是继承性/旧基线问题,无一来自本 PR diff,对照 CI 总控 #4599 逐一定位:PR 基线 cd92b0b 早于 master 的门禁修复提交 4d34abf(config-surface guard 脚本 bug + kernel lockfile 同步 + 派生文档再生),config-surface-guard / lock-sync / docs-sync 三个红正是该提交修掉的那组漂移;第 4 个 unused-exports 是 master 自身的 born-red(#4545 train 新增了一个只在文件内部使用的类型导出、de-export 清扫漏了它),与 master head 运行的错误行逐字节相同。

标签:不变(既有 PR 的 CI 失败修复,类别/范围/严重程度沿用原 PR 判定)。

修复动作(auto-fix lane attempt 1/2,直接修完未停):

  1. rebase 到 master tip 146b34f(无冲突;唯一双方触碰的 src/server.ts,本 PR hunk 在 installWebSocketBridge 调用处、master 改动在 preflight/auto-fold 区,互不重叠)→ config-surface-guard、lock-sync(root + kernel 两个 lockfile)、docs-sync 本地复跑全绿。
  2. 剩一个 unused-exports 红按几行级小修纪律在本 PR 顺手修掉:独立标注 commit 应用与 [PR] fix(pi): inline mode now intercepts codex Responses WebSocket traffic (#4526) #45598042523 完全相同的一词修复(type ClientBuildCheck 去掉 export 关键字)。已 grep 验证零外部使用点(5 处引用全在 client-build-note.ts 内部),纯编译期类型可见性变化,行为中性。宁可跨 PR 重复修也不让本 PR 继续阻塞等 [PR] fix(pi): inline mode now intercepts codex Responses WebSocket traffic (#4526) #4559 合入。
  3. 验证:4 个门禁脚本本地全绿 + npm run typecheck clean + 定向测试 tests/codex-responses-ws.test.ts 12/12 过(全量回归交给 CI)。
  4. 推送:新 head e425e78dd,force-with-lease(带显式期望值);推送前已验证远端分支 tip 就是 PR head cdda1b3 且 patch-id 与 rebase 后 commit 逐字节一致——作者侧无任何工作丢失。

八问:

  1. 是什么问题:PR 上 4 个 CI 门禁红,全是环境继承问题(基线太旧 3 个 + master 自身漏网的小导出 1 个),不是代码功能缺陷。
  2. 什么原因导致的:PR 从 cd92b0b 开出,之后 master 落了 4d34abf(修门禁脚本 bug、同步 kernel 0.0.163 lockfile、再生派生配置文档)和 [PR] dsh lane: note in-session when running an unverified @deepseek-ai/dsh client build (soft, exact-match) #4545 train(引入内部专用类型但漏去导出);旧基线树缺的正是这组修复,所以同一组门禁在旧树上必红。
  3. 影响范围与严重程度:仅影响本 PR 可合并性(checks 全红无法合入);无用户面影响、无运行时行为变化。
  4. 怎么修复的:策略 = 把基线对齐到 master tip(rebase)+ 对唯一残留的 born-red 应用 [PR] fix(pi): inline mode now intercepts codex Responses WebSocket traffic (#4526) #4559 里已被审过的一词 de-export 作为独立 commit;未动任何业务逻辑。
  5. 是否有回归问题:无。逐项审过:本 PR 自身 diff(responses-ws probe relay / ws-bridge egressProxyFor / tunnel-guard)未被我的操作改动,rebase 干净合入且 hunk 无重叠;de-export 只改类型可见性、运行时零影响;三个门禁转绿靠的是 master 自己的变更(4d34abf,披露见该提交)。高危模式(语义漂移/fail-fast 变吞错/阈值时序输出格式变化)逐项过:均不涉及。
  6. 是否破坏缓存:不涉及(未触碰任何缓存/重试/游标/去重键逻辑)。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及(无 env/config/默认值新增)。

行为变更披露(按规则 4):无——rebase 带入的是 master 自身已披露的变更;de-export 为纯类型层,运行时输出零差异。

判定:修复完成,新 head e425e78dd 已推送、等 CI 重触发。若后续 test 车道再红且签名与本 PR 无关,按 #4599 总控先判 master 红绿再定动不动手,不开新 issue。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

PR 审核(独立于 issue 侧结论重新做过,非复读)。

① 原 issue(方向关):#4594 要修的是「codex context-only prewarm 被当成 ACP turn 执行 + probe 出站绕过上游代理」。作者修的是生产侧根因(probe 认定范围 + egress 路由),不是症状拦截;消费侧(gate/恢复)已由 #4040/#4085 覆盖,本 PR 与之互补。方向无大问题。

② 回归审查(触碰路径的既有行为逐项清点):

  1. probe gate(src/responses-ws.ts):旧行为「空 input + generate:false → 逐字 relay」完整保留(空数组 .every() = true);新纳入的是「仅 developer/system/user 消息项 + generate:false」——这类帧旧行为是走管道(零对话 turn → fold wipeout + 摘要重注 + 工具膨胀),即 bug 本身,改为 relay 是有意变更(已披露)。其他客户端影响面核查:WS lane 只有两个 codec(opencode 插件车道 / codex 前缀车道);opencode 车道的非空 warmup 形状由 tests/responses-ws-http-fallback.test.ts:283 钉住——该测试未改、rebase 后本地仍绿(relay 与 pipeline 两条路径 commit 同一原始 body,Support Codex Responses WS clients with HTTP/SSE upstreams, preserving local delta checkpoints #3342 checkpoint 契约不变,continuation 合并语义不变)。
  2. tunnel-guard(src/ws-bridge.ts):入参类型 boolean→string|undefined,守卫真值由 egressProxyFor?.(url) !== undefined 推导,与旧 resolveProxyDecision(...).proxy !== undefined 逐位相同——无语义变化。唯一生产调用点(src/server.ts:712)已同步更新;tests/ws-bridge.test.ts 不传该可选参数,不受影响。全仓无 isEgressProxied 残留引用。
  3. probe egress:恒直连 → 配置了上游代理时经代理(修复目标)。机制核查:per-call dispatcher 通道本就存在且两条出站路径都吃(WS connect src/responses-ws.ts:281、HTTP fallback :327);proxyDispatcher(undefined) 直连;缓存键(proxyUrl+timeout+keepalive+CA identity,[Bug][tls] Proxy upstream TLS broken by security-software HTTPS scanning (Kaspersky MITM): client sees only a generic 502 | 安全软件HTTPS扫描致代理上游TLS失败,客户端只见泛502 #3776)与 pipeline 完全一致——probe 与 turn 共享同一 ProxyAgent,无 per-probe agent 泄漏。连接键不含 dispatcher,同会话内 resolver 确定性一致,不存在直连/代理 socket 混用。
  4. history prewarm:谓词要求每个 item 都是 dev/system/user 消息项(type 缺省或 "message"),assistant/function_call 等任一出现即留管道——负向测试钉住(tools 注入、stats.requests 递增)。
  5. stats.requests:context-only prewarm 不再计数(旧:计入)。有意变更(与空 input probe 历来不计一致)。
  6. x-bili-hop / x-bili-tunnel stamp(codex WS 转发:每个 turn 都新开一条上游 WebSocket 连接(probe 与 turn 的连接键不一致) #3073):relay 路径的 stamp 逻辑未动,probe 与 turn 握手标记一致性保持。

高危域点名:自动更新/自更新路径未触碰;dsh 相关未触碰;版本钉死/lockfile 未触碰;systemd/env 模板未触碰;网络/代理路径有触碰(probe egress)——已按上述第 3 条核实。

③ 代码审查:isPrewarmProbe 命名达意、单一职责;注释块更新为准确描述(引用 #4594,与代码事实相符);无死代码、无新依赖、风格与仓库一致。一个设计取舍认可:item.type === undefined 的宽容放行——Responses wire 上 item 规范应带 type,无 type 的 dev/system/user 对象最坏情况只是把一个 no-generation 请求多 relay 了一次(generate:false 下管道本来也无事可做),不放行反而会把合法 prewarm 变回 bug。可接受。

④ 测试有效性:三个新测试都有牙且是双向门禁(#3617)——正向(context-only prewarm:逐字 deepEqual、sentinel 缺席、tools 为空、stats 不变、fold 仍 active;修复前红)、负向(history prewarm 留管道;修复前后绿)、代理 CONNECT 计数(修复前 0 CONNECT 红)。断言测的是行为而非 mock 形状。fixture 改动(fake upstream 对任何 generate:false 帧回空 terminal)对齐真实上游行为,不影响既有测试语义。

⑤ diff 整洁度:4 文件 +124/−20,每行都与目的相关,无跑题文件、无生成物/锁文件连带变更。干净。

⑤b 代码整洁硬标准:命名/死代码/重复/错误处理/魔法值/注释/风格/依赖/函数职责逐项过,全部达标;外部提交从严标准同样满足。无需重构。

结构化八问:

  1. 是什么问题:codex 每次新建/恢复线程发的上下文预热(developer/user 项 + generate:false)被 bili 当成普通 turn 执行——wipe 掉全部 fold、整表重注伪造摘要、注入无用工具;且预热探测出站不走已配置的上游代理,代理后面必然失败。
  2. 什么原因导致的:probe 认定只写了「空 input」一种形状(bug: Responses degenerate retry overlooks custom tool calls and drops completed-attempt usage #1862 时代 codex 只发空预热),codex 新版预热带上下文项后掉进管道;probe relay 建 transport 时没带 dispatcher,而 ws-bridge 对外只暴露布尔,relay 拿不到可用于出站的代理决策。
  3. 影响范围与严重程度:所有走 codex WS lane 且带历史的会话,每次 resume 必中(fold wipeout + 摘要重注 + 字节膨胀);伪造摘要表够大时长会话触发 413 → 强制 pre-turn compaction → 超大转发 → WS 1009,resume 首 turn outright 失败(约 15 分钟);上游在代理后面的用户 probe 恒失败。p1(会话不可用类)。
  4. 怎么修复的:把 probe 认定从「空 input」扩到「不复现任何对话 turn」(generate:false 且全部项为 developer/system/user 消息),这类帧逐字转发、永不进管道;ws-bridge 把布尔升级为「解析出的代理 URL」,tunnel-guard 与 probe relay 共用同一个决策,probe 出站带上与 turn 相同的 ProxyAgent。
  5. 是否有回归问题:审过既有行为清单(见②节六项)——空 input probe 行为逐字保留、tunnel-guard 真值不变、checkpoint 契约不变、history prewarm 留管道、stamp 一致性保持;无意回归未发现。三处有意行为变更均已披露。
  6. 是否破坏缓存:不破坏,方向相反——修复消除的正是 fold wipeout 导致的缓存接缝崩坏;prefix cache key、重试、游标、去重策略均未触碰;probe 多走的代理 hop 不改变帧内容。
  7. 是否改了设计方向:没有。扩展既有 probe-relay 车道覆盖它本就该覆盖的帧类;ws-bridge 参数类型变化是内部接线统一(一个决策两处消费),非架构转向。
  8. 是否引入了新的配置:无新 env/config/默认值。

验证记录:rebase 到 master tip(146b34f27)无冲突(本地已做,未推分支——BEHIND 不构成阻塞,等 #4606 落地后 rebase 一次即可);rebase 头上定向 5 套件(codex-responses-ws / responses-ws-http-fallback / ws-bridge / tunnel-guard / upstream-proxy-routing)60/60 绿,typecheck 干净。CI 现状红为继承性失败(base cd92b0b 早于 4d34abf 的门禁修复),已在 #4599 登记。

判定:可合——待 #4606 合入、本分支 rebase 至 tip、CI 全绿后由 owner 合并。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

最终审核结论(八问)

  1. 是什么问题:codex 启动/空闲预热帧(只带上下文、不要求生成)被 bili 当成正常对话 turn 处理——把会话已有的折叠状态全部清零并落盘,同时往一个不产出的请求里塞整份历史摘要和工具;另外探针出站不走上游代理,配了代理的机器上直接握手失败。
  2. 什么原因导致的:两个缺陷同一机制家族(探针识别 + 探针出站):① gate 谓词只认「空 input」为探针,只带 developer/user 上下文的帧漏进 processTurn,零对话 turn 触发 fold 失活持久化 + forged summaries 全量重注入;② ws-bridge 只向 session 层暴露布尔值,探针 relay 拿不到解析出的代理 URL,只能直连。
  3. 影响范围与严重程度:p1(已打标)。触发面 = 所有 codex WebSocket 车道会话的启动/空闲预热,机器重启后每个恢复的线程必发此帧;长会话(大量历史摘要)升级为:预热 413 → codex 强制 pre-turn compaction → 17.5 MB 逐字转发 → 上游 1009 断连,turn 卡 ~15 分钟失败。现场七例会话折叠清零(73→0 等)。
  4. 怎么修复的:修决策点不遮症状——探针谓词放宽为「generate:false 且 input 仅含 developer/system/user item(空数组自然包含)」,此类帧逐字转发、永不入 pipeline;ws-bridge 改暴露解析后的代理 URL(egressProxyFor),隧道守卫布尔从同一个 resolveProxyDecision 调用推导,单一决策源。回放真实 turn 的历史预热(含 assistant/tool item)仍走 pipeline,不变。
  5. 是否有回归问题:审过的既有行为逐项: (a) 旧的空 input 探针行为被包含,不变;(b) 历史回放预热仍走 pipeline——负向测试钉住(ACP 工具注入、计一次请求);(c) 直连场景 proxyDispatcher(undefined) 返回 undefined,行为不变;(d) 隧道守卫布尔语义等价(同一决策结果推导);(e) Support Codex Responses WS clients with HTTP/SSE upstreams, preserving local delta checkpoints #3342 的 checkpoint commit 在 relay 路径保留;(f) 两个 codec(opencode-v2 插件车道 / codex 前缀车道)共用 session 类,gate 放宽对 OC V2 车道是保护而非回归(它此前同样会踩 fold 清零病理);(g) relay 路径不碰会话状态,无新 fold/锚点副作用。高危模式扫描:无语义漂移、无 fail-fast 变吞错、无阈值/时序/输出格式漂移。未发现回归。
  6. 是否破坏缓存:不破坏,反而消除旧路径的成本损伤(fold 失活逼原文全量重发、forged summaries 注入膨胀)。作者 A/B 实测真实 turn 字节级一致。
  7. 是否改了设计方向:没有。「探针 = 传输层噪音、逐字转发」原则不变,只是识别范围从「空 input」扩到「不含对话 turn」。
  8. 是否引入新的配置:没有。无新 env/config/默认值。

验证证据:rebase 到当前 master(146b34f)后本地 typecheck 干净;定向套件 151/151 绿(codex-responses-ws 12/12 含三个新增正反用例 + responses-ws-http-fallback / responses-prewarm-generate / responses-websocket / ws-bridge / ws-upgrade / tunnel-guard / upstream-proxy-routing / skipped-proxy-hint / native-ws-intercept / issue3701-ws-byte-budget)。CI:静态门禁全绿;11 个 test 分片红 = master 继承(LIVE_FILES tripwire,一行测试未跑,与 #4599 盘点第 2 条逐字一致,已在 #4599 留痕)。PR 作者在审核期间自行 rebase 并追加了 ClientBuildCheck de-export 小修(head e425e78,与我 rebase 的内容 diff 完全一致+该一行),我此前的推送已被其取代,无冲突。

遗留(作者已声明不在本 PR):codexForgedSummaries append-only/echo 缺失时全量重注入、WS 1009 超大错误未回传客户端(等空闲超时)且非字符串 body 报 0.0 MiB——属同链条相邻缺陷,来源链限一层规则下不另立项,留在 PR 描述供 owner 决定后续安排。

判定:代码可合。CI 红灯全部继承自 master(#4599 已接管,#4606 修复在途)——#4606 合入后 rebase 本 PR 即全绿,可合并。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

补充一行状态(我 18:16 的审核结论针对 cdda1b3,期间 CI worker 已推新 head,此处对齐):

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

CI round 2 — test shards: root cause found, fixed in-branch, new head pushed (d25f4f1)

After the rebase (head e425e78), CI came back with all four originally reported gates green (config-surface-guard / lock-sync / docs-sync / unused-exports), plus every other gate job (docs-coverage, lifecycle-ledger-guard, test-import-graph, kernel-guard, pi-subagents-guard, lane-guard, version-guard). But all 11 test-shard jobs failed at startup — no single test ran — with one identical signature:

process-spawning test files missing from LIVE_FILES in scripts/run-ci-tests.mjs:
tests/issue3890-parent-gone-confirm-drain.test.ts

Diagnosis. #4590 (tier5-prepare-train, merged to master as 146b34f — the rebase target) added tests/issue3890-parent-gone-confirm-drain.test.ts without registering it in scripts/live-files.mjs. The SPAWN_RE tripwire at the top of scripts/run-ci-tests.mjs exists precisely to catch that and aborts the whole shard before any test executes. Master's own last run had its test lanes plan-skipped, so this born-red never surfaced on master; it hits every PR head rebased past 146b34f. It does not come from this PR's diff (which touches only src/responses-ws.ts, src/server.ts ×1 line, src/ws-bridge.ts, tests/codex-responses-ws.test.ts).

Fix (commit d25f4f1, fast-forward append — nothing overwritten). Appended the suite to LIVE_FILES, which is the tripwire's own prescribed remedy. Per the #1248 discipline I reviewed its concurrency assumptions first: the unit half only imports the pure tickWatchedPid; the live half spawns ONE fixture child that binds listen(0) and reports READY <port> on stdout, with isolated XDG dirs per run — no fixed port, so its zone slot band stays inert and --test-concurrency 4 is safe. Append-only placement keeps every existing slot/band un-renumbered. Verified locally by replicating the exact tripwire scan (existence + SPAWN_RE over all tests/*.test.ts): clean.

八问 (this round)

  1. 是什么问题: 每个测试分片在启动时直接中止——新增的 [Bug] Windows: Proxy falsely detects parent-gone and self-terminates on every forwarded request (same live PID triggered | [Bug] Windows:代理每转发一次请求就误判 parent-gone 自杀(同一存活 PID 在 2.5h 内触发 12 次) #3890 子进程测试套件没登记进 LIVE_FILES,测试一条都没跑。
  2. 什么原因导致的: [PR] merge: tier-5 train — dsh prepare/streaming domain (7 PRs: bili_final terminal marker, auto-fold storm gate, wire-only guard) #4590 合入 master 时加了测试文件却漏改 scripts/live-files.mjs(tripwire 就是为抓这种漏登记而设);master 自己的测试车道被 plan-skip,所以红只暴露在 rebase 过 146b34f 的 PR 头上。
  3. 影响范围与严重程度: 所有 rebase 到该 master 之后的 PR 测试车道整体不可用(CI 验证能力,非产品行为);终端用户不受影响。
  4. 怎么修复的: 按表内「append-only」规则把套件追加进共享 LIVE_FILES 表;登记前完成并发假设审查(listen(0)+READY 端口发现,无固定端口)。
  5. 是否有回归问题: 无——仅注册表加一行,既有 slot/端口带不变;副作用仅是新文件按设计参与分片重排(无实测权重时回退行数加权),不丢任何文件。
  6. 是否破坏缓存: 不涉及。
  7. 是否改了设计方向: 不涉及。
  8. 是否引入新配置: 不涉及。

判定: 本轮继承性红全部在分支内收敛完毕(rebase 消掉 gate 漂移 + ClientBuildCheck de-export + LIVE_FILES 登记),PR 头 d25f4f1、base 已是 master 最新 146b34f、mergeable=True。等 CI 结果;若分片再暴露别的失败,由下一轮 auto-fix 处理。已按规在 #4599 留一行结论,未开新 issue。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

收工判定 — attempt 1/2 完成,全部继承性红已在分支内收敛并推送

PR 头现指向 d25f4f15a(base = master 最新 146b34f27,mergeable=True),CI 正在重跑。本轮修掉的都不是本 PR diff 引入的,而是分支追上 master 后暴露的三处继承性红:

  1. config-surface-guard / lock-sync / docs-sync:stale-base 漂移,master 侧 4d34abfaf 已修 → rebase 到 146b34f27 即消。
  2. unused-exports(ClientBuildCheck):[PR] dsh lane: note in-session when running an unverified @deepseek-ai/dsh client build (soft, exact-match) #4545 de-export 漏网的 master-born-red → 按 [PR] fix(pi): inline mode now intercepts codex Responses WebSocket traffic (#4526) #4559 同款一字 de-export 修掉。
  3. 11 个测试分片启动即中止:[PR] merge: tier-5 train — dsh prepare/streaming domain (7 PRs: bili_final terminal marker, auto-fold storm gate, wire-only guard) #4590 新增 tests/issue3890-parent-gone-confirm-drain.test.ts 却漏登记进 scripts/live-files.mjs,SPAWN_RE tripwire 在任何测试运行前就 abort(master 自身测试车道被 plan-skip 故从未暴露)→ 按「append-only」规则补登记,登记前完成 tests/claude-native.test.ts linux live-E2E subtests fail/hang in shared multi-agent sandboxes #1248 并发审查(fixture 用 listen(0)+READY 端口,无固定端口,--test-concurrency 4 安全)。

本地已验证:四个 gate 脚本全绿、tripwire 扫描干净(49 条、零未登记 spawner)、typecheck 干净、定向 tests/codex-responses-ws.test.ts 12/12。全程无需设计决策,均为几行级小修。已按规在 #4599 留结论、未开新 issue。若 d25f4f15a 的 CI 再暴露别的失败,交由下一轮 auto-fix 处理。

ranxianglei pushed a commit that referenced this pull request Oct 24, 2026
…ot this PR's

master 146b34f fails the unused-exports gate on this exact line
(INTERNAL_ONLY src/agent/client-build-note.ts: ClientBuildCheck): the
type-only export #4545 introduced was missed by the 4d34abf de-export
sweep. This PR touches none of it — every PR rebased past that master
inherits the red (#4599 storm log). Same one-word fix as #4606 commit 1
and the in-branch precedents (#4607/#4595/#4540/#4618/#4622): drop the
leading export on a symbol referenced only within its own file. Drop
this commit when rebasing after #4606 merges.
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

attempt 2/2 完成 — 10 个测试分片红已修,推送 952d80b57(在 d25f4f15a 上快进追加,未覆盖任何提交),CI 重跑中。

查重: 同家族但机制不同——本轮 5 个失败测试全部是 PR base(cd92b0b7d)上的 born-red(rebase 启用测试车道后才暴露,原 run 的 test 车道被 plan-skip 从未跑过),与本 PR diff 无关。与 #4606 的关系:generation-boundary 一行是有意逐字节复制 #4606 commit 389b81565(即 #4559 e82ccd7d6 同法),#4606 合入后 rebase 时丢弃即可;error-dump 家族不在 #4606 内,是本分支的首个落地修复(master 侧仍由 #4599 sentinel 清单跟踪)。issue 检索无匹配存量条目。

八问

  1. 是什么问题:CI 10 个测试分片全红,实际失败测试只有 5 个(ubuntu/windows×node22/24 各分片跑同一批文件,签名相同)。
  2. 什么原因导致的:两个独立机制,均非本 PR 引入——
  3. 影响范围与严重程度:Switching model mid-conversation restarts the ref ledger, but the rendered transcript still shows the old refs — the model keeps aiming compress at dead ranges until the circuit breaker fires | 切模型后消息编号整体换账本,但渲染历史仍显示旧编号,模型对着死区间反复压缩直至熔断 #4305 项纯测试面,无运行时影响。error-dump 项影响诊断产物(err-*.json / summary err dump):文件超出 BILI_DUMP_4XX_MAX_BYTES 至多一个 banner 行——诊断卫生缺陷,不涉可用性/缓存;叠加效应是 born-red 卡住所有 rebase 到 tip 的 PR 合并。
  4. 怎么修复的:gen-boundary 一行按正确位置补参(与 [PR] fix(ci): clear the two master-red gates in one branch — unused-exports (#4602) + the #3890 LIVE_FILES tripwire (#4604) #4606 同款);error-dump 两个写函数各自把 banner 字节预留进预算,使最终文件(含 banner)≤ 上限,并把四处过时断言更新为钉住一致契约(banner 存在、尺寸含 banner、经 stripDumpBanner 解析)。
  5. 是否有回归问题:审过既有行为——截断 marker 语义不变、pretty-print 门不变、summary 双端预算分配的短边相等性保持;issue4482-dump-banner(banner 功能契约)、issue1993-preflight-400、reasoning-echo-400、本 PR 自己的 codex-responses-ws(12/12)本地全绿;无 fail-fast 变吞错、无阈值/时序漂移(除已披露项)。
  6. 是否破坏缓存:不涉及(未触碰任何缓存/重试/游标/去重键)。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新配置:不涉及(无新 env/config/默认值)。

行为变更披露(旧→新+原因):dump 文件的字节预算——旧:banner 在上限之外,最终文件可超 BILI_DUMP_4XX_MAX_BYTES 至多 ~89B;新:最终文件(含 banner)严格 ≤ 上限,落在 (cap−banner, cap] 的 body 现在会被 [truncated:] marker 截断而非溢出。原因:让 #4482(无条件 banner)与 #4297(硬字节上限)两个已发布契约自洽,溢出即 shipped bug。

收工判定:本轮均为几行级小修,无需设计决策,已全部落地并推送;若 952d80b57 的 CI 再暴露这些签名之外的失败,超出 attempt 2/2 范围,将停手留结论。

z0d0dOr and others added 3 commits October 25, 2026 11:59
…tream proxy (ranxianglei#4594)

A generate:false request that replays no conversation turn (empty input, or
only developer/system/user messages) is a prewarm probe: forward it verbatim
instead of running it as a zero-message ACP turn, which deactivated every fold
block and re-injected the whole forged-summary list. History prewarms keep
riding the pipeline. The probe relay now egresses through the same upstream
proxy the relay resolves; ws-bridge exposes the resolved proxy instead of a
boolean so the tunnel guard and the probe share one decision.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The PR branch (codex context-only prewarm probe relay) had diverged from
master. Merging master in produced a single conflict in src/responses-ws.ts:
the PR added isPrewarmProbe() while master added the x.ai continuation-rejection
helpers (XAI_RESPONSE_MISSING / continuationRejected / untypedErrorEnvelope) at
the same location. Both are independent and both are used, so the resolution
keeps both blocks.
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Merge conflict resolved and pushed. Merged current master (e1f31ad) into the PR branch 2026-10-24_codex-context-prewarm-probe, resolved the single content conflict, committed, and pushed to the fork head ebb80ef63b.

  • Conflict: one file — src/responses-ws.ts. The PR added isPrewarmProbe() while master added the x.ai continuation-rejection helpers (XAI_RESPONSE_MISSING / continuationRejected / untypedErrorEnvelope) at the same spot. Both are independent and both are referenced in the merged file, so the resolution keeps both blocks (verified each symbol is used; no markers left; structure balanced).
  • Method: merge (not rebase) — preserves the branch's existing test-fix commits and avoids a force-push. Before pushing I re-fetched the fork head, confirmed it was still ae4b193 (an ancestor of my head), so it was a clean fast-forward push.
  • Diff integrity: effective PR diff vs master is unchanged — exactly the 7 intended files (5 src/ + 2 tests/), no kernel/ or src/update.ts pollution from the merge.
  • Verification: mergeable: True now (was False/dirty). I did not run a local typecheck (no node_modules in this worktree; installing would churn the lockfile) — the full typecheck/test verdict comes from CI on the pushed head ebb80ef63b per local-testing discipline. The mirror check cache is still on the stale sha, so I cannot yet confirm the new head's check results; they will run on ebb80ef63b.

八问

  1. 是什么问题:PR [PR] fix: route codex context-only prewarms to the probe relay via the upstream proxy #4595 与 master 基线产生合并冲突(mergeable_state: dirty),分支落后 master 1406 个提交,无法合并。
  2. 什么原因导致的:PR 分支(codex context-only prewarm 走 probe relay)长期未同步 master;期间 master 在 src/responses-ws.ts 同一位置新增了 x.ai continuation-rejection 相关函数,与 PR 新增的 isPrewarmProbe() 发生文本冲突。
  3. 影响范围与严重程度:仅阻塞该 PR 合并(此前 mergeable=False/dirty),不影响线上功能;属合并阻塞,非功能缺陷。
  4. 怎么修复的:把 master merge 进 PR 分支(选 merge 而非 rebase:保留分支上已有的测试修复提交,且免 force-push);唯一冲突 src/responses-ws.ts 两侧是相互独立的新增函数,均被引用,故同时保留。
  5. 是否有回归问题:冲突解决只保留两个独立函数块,未改动任何既有逻辑;其余文件由 git 三方 auto-merge 处理;有效 PR diff 仍为原 7 个文件,无 kernel/ / src/update.ts 污染。既有行为未被破坏。
  6. 是否破坏缓存:不涉及(纯合并接线,未触碰缓存/重试/去重键逻辑)。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:冲突已解除并推送,PR 现 mergeable=True(state blocked = 等待 CI/审批,非冲突)。等待 CI 在 ebb80ef63b 上跑绿后即可由 owner 合并。

Inherited CI red: the visibleNotices failures (ubuntu 22/24 + win-22-5,
storage-sqlite-default missing) are master-born-red after ranxianglei#6480 flipped the
storage default to pure-sqlite — the notice-registry suite preload pins
BILI_STORAGE_UNIFIED=legacy, silencing the notice. ci-fixes c6369fb pins
the pure-sqlite selection in the suite. Per CI-coordination policy (inherited
red -> merge origin/ci-fixes), bringing that convergence in rather than
re-fixing on the PR branch. The Windows EBUSY unlink index.db shards
(win-22-1 store-zstd, win-22-4 metadata-fold) are the separate ranxianglei#4347
file-lock flake, still pending ci-fixes convergence.
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Red CI diagnosed: all 5 failing shards are inherited (none are this PR's regression). Merged ci-fixes into the PR branch and pushed — new head 4a1119e8b8.

Re-verified live first: the mirror check cache was still stale (old sha, all queued), so I pulled the 5 job logs directly. Failure map:

Shard Failing tests Root cause Inherited?
ubuntu-22-1, ubuntu-24-1, win-22-5 visibleNotices ×2 (storage-sqlite-default missing) After #6480 flipped the storage default to pure-sqlite, the notice-registry suite's preload pins BILI_STORAGE_UNIFIED="legacy", silencing the notice → born-red on master ✅ yes — fixed on ci-fixes c6369fb6a1
win-22-1 store-zstd ×10 EBUSY: unlink …/index.db — Windows AV/indexer file-lock flake (#4347 family) ✅ yes — pending ci-fixes convergence
win-22-4 metadata-fold #5506 ×3 same EBUSY unlink index.db flake ✅ yes — pending ci-fixes convergence

This PR's diff only touches src/responses-ws.ts / inline-engine.ts / inline-ws.ts / server.ts / ws-bridge.ts + 2 test files — it does not touch notices, persist, or storage, so none of these are its regression.

Action taken (per CI-coordination policy: inherited red → merge origin/ci-fixes, don't re-fix master-level failures on the PR branch):

Effective PR diff vs master is clean: the original 7 files + the 5-line ci-fixes test pin, no kernel/ or src/update.ts pollution. mergeable: True (state blocked = awaiting CI/approvals, not a conflict).


八问

  1. 是什么问题:PR [PR] fix: route codex context-only prewarms to the probe relay via the upstream proxy #4595 的 5 个 test 分片红——ubuntu 22/24 与 win-22-5 的 visibleNotices ×2,win-22-1 的 store-zstd ×10,win-22-4 的 metadata-fold Scheduled full test failed: ee0dcc7c | 定时全量测试失败: ee0dcc7c #5506 ×3。
  2. 什么原因导致的:全部继承性失败,非本 PR 回归。① visibleNotices(storage-sqlite-default 缺失)= master 在 [PR] storage: pure-SQLite default — JSON envelopes stop being written unless dual-write is requested (#6479) #6480 把存储默认翻成 pure-sqlite 后,notice-registry 套件 preload 仍钉 BILI_STORAGE_UNIFIED=legacy,使该 notice 静默,born-red;② win-22-1/4 的 EBUSY unlink index.db = windows-22-1 lane randomly red: during test cleanup, unlink index.db hit an EBUSY external lock, couldn't survive tmp-rm | windows-22-1 车道随机红:测试清理期 unlink index.db 遇 EBUSY 外部锁,熬不过 tmp-rm 的 9.6 秒重试梯子(#2799/#2908 族复发) #4347 家族 Windows 文件锁 flake(OS AV/indexer 锁,清理时 unlink 失败),[PR] storage: pure-SQLite default — JSON envelopes stop being written unless dual-write is requested (#6479) #6480 让 index.db 成加载路径后更多测试在 Windows 上 create+unlink。本 PR diff 只动 responses-ws/inline/server/ws-bridge + 2 个测试,不触碰 notices/persist/storage。
  3. 影响范围与严重程度:仅阻塞该 PR 的 CI 信号,无用户侧功能影响;属继承红 + 环境 flake。
  4. 怎么修复的:按 CI 协调政策(继承红 → merge origin/ci-fixes,不在自己分支重复修 master 级失败),把 ci-fixes 合进 PR 分支,带入 c6369fb6a1(notice-registry 套件钉 pure-sqlite 选择,5 行测试改动)修掉 visibleNotices 面;Windows EBUSY 面尚未在 ci-fixes 落地,不在本分支修。
  5. 是否有回归问题:无。合入的 ci-fixes 改动只动 tests/notice-registry.test.ts(5 行),本 PR 原有 7 文件 diff 不变,无 kernel/ 或 src/update.ts 污染。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:visibleNotices 红已由 ci-fixes 合入修掉(新 head 4a1119e8b8,等 CI 重跑转绿);Windows EBUSY(win-22-1/4)为继承 #4347 flake、非本 PR 回归,若新 run 仍红属同一 flake,应换新 runner 重跑(哨兵在跟踪 ci-fixes 收敛)。

Conflict resolution: src/ws-bridge.ts — kept the PR's egressProxyFor routing (egressProxyFor?.(upstream) !== undefined, consistent with the param/interface/createSession in the merged tree) and added master's shutdown guard (if (closing || socket.destroyed) return;). No other conflicts. This is the final reviewed head (full CI pass).
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Conflict resolved and pushed — new head ed0492a8a9. PR is now mergeable: True (state blocked = awaiting CI/approvals, not a conflict).

What happened: the PR went dirty again because master advanced e1f31ad → a446a3c1c8 (494 commits) past the PR head 4a1119e8b8. I merged origin/master into the PR branch.

The one conflict — src/ws-bridge.ts, the egressProxied: call site. Both sides changed the same line:

  • PR (ours): renamed the param isEgressProxied → egressProxyFor (returns the proxy URL string, needed to actually route the prewarm through it) and set the call site to egressProxied: egressProxyFor?.(upstream) !== undefined.
  • master (theirs): kept isEgressProxied and added a shutdown guard if (closing || socket.destroyed) return; after the async checkTunnelDestination.

Resolution: kept the PR's egressProxyFor?.(upstream) !== undefined and added master's guard. This is the only coherent option — the non-conflicting regions (the function param, the WsBridgeContext interface field, and the createSession call) all auto-merged to the PR's egressProxyFor, so using master's isEgressProxied at the call site would be a compile error. I verified no stray isEgressProxied reference remains anywhere in src/.

The other three auto-merged files (inline-engine.ts, responses-ws.ts, server.ts) just carry master's own 494-commit work; the PR's core change (isPrewarmProbe, the egressProxyFor routing) is intact.

Push: verified the fork head was still 4a1119e8b8 (no concurrent agent), then pushed fast-forward 4a1119e8b8..ed0492a8a9 (no force-push). Per the CI-hold policy this is the final reviewed head, so the commit carries no [ework:ci-held] marker — it triggers this branch's one full CI pass.


八问

  1. 是什么问题:PR [PR] fix: route codex context-only prewarms to the probe relay via the upstream proxy #4595 与 master 再次冲突(merge state: dirty)——master 从 e1f31ad 前进到 a446a3c1c8(494 个提交),PR head 4a1119e8b8 落后。
  2. 什么原因导致的:master 在 PR 上次同步(e1f31ad)之后又合入 494 个提交,其中对 src/ws-bridge.ts 的改动与 PR 改动落在同一行(egressProxied 调用点)产生语义冲突:PR 把参数 isEgressProxied 改名成 egressProxyFor(返回代理 URL 字符串,用于实际经代理路由),master 保留 isEgressProxied 并新增一行关闭期守卫 if (closing || socket.destroyed) return;。
  3. 影响范围与严重程度:仅阻塞该 PR 的合并(冲突态),无用户侧功能影响;属常规同步冲突。
  4. 怎么修复的:把 origin/master 合进 PR 分支;唯一冲突在 src/ws-bridge.ts 的 egressProxied 调用点——保留 PR 的 egressProxyFor?.(upstream) !== undefined(与合并树里已自动合并的参数/接口/createSession 一致,用 isEgressProxied 会编译报错),并加上 master 的关闭期守卫。其余 3 个自动合并文件带入的是 master 自身 494 提交的工作,PR 核心改动(isPrewarmProbe、egressProxyFor 路由)完整保留。
  5. 是否有回归问题:无。冲突解决只动了 src/ws-bridge.ts 一行调用点 + 加一行守卫,PR 原有 7 文件 diff 不变;master 带入的改动是 master 自身已合并的工作,非本 PR 引入。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:冲突已解决,PR head 现为 ed0492a8a9(mergeable: True,state: blocked = 等 CI/审批,非冲突)。这是最终 reviewed head(无 [ework:ci-held] 标记),已触发一轮完整 CI。

…param shape

The master-added tests/ws-bridge-pending-close.test.ts was written against the old isEgressProxied?: (url) => boolean param. The PR renamed that param to egressProxyFor?: (url) => string | undefined, so the test's () => proxied mock no longer typechecks (TS2345 on static-gates). Pass a proxy URL string when proxied, undefined otherwise — preserves the exact egressProxied boolean the tunnel-guard derives (egressProxyFor?.(upstream) !== undefined). Final reviewed head (full CI pass).
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Re-verified live first: the two static-gates failures (ubuntu-24, windows-22) were genuinely red on head ed0492a8a9 — not already fixed. Diagnosed and fixed; new head 13e6b51984 pushed (fast-forward, no force, no [ework:ci-held] marker → one full CI pass).

Root cause — a merge interaction, not a runtime bug. static-gates runs typecheck, and it failed with:

tests/ws-bridge-pending-close.test.ts(24,87): error TS2345: Argument of type '() => boolean' is not assignable to parameter of type '(destinationUrl: string) => string | undefined'.

Master's 494 commits added tests/ws-bridge-pending-close.test.ts (testing its new closing/destroyed shutdown guard), written against the old param shape isEgressProxied?: (url) => boolean — its mock passed () => proxied (a boolean). This PR renamed that param to egressProxyFor?: (url) => string | undefined (returns the proxy URL string, needed to actually route through it). After the merge, the test's boolean mock no longer typechecks against the renamed param.

The fix (one line, tests/ws-bridge-pending-close.test.ts:24):

- const handler = installWebSocketBridge(server, async () => {}, () => {}, [codec], () => proxied);
+ const handler = installWebSocketBridge(server, async () => {}, () => {}, [codec], () => (proxied ? "http://127.0.0.1:3128" : undefined));

This preserves the test's exact semantics: when proxied is true the function returns a proxy URL string (→ egressProxied = true), when false it returns undefined (→ egressProxied = false) — the same boolean the tunnel-guard derives via egressProxyFor?.(upstream) !== undefined. The proxy URL value itself is never consumed (only its defined-ness matters), so the sentinel is inert.

Verified before pushing: no isEgressProxied reference remains anywhere in src/ or tests/; the other installWebSocketBridge caller (tests/ws-bridge.test.ts:45) passes only 4 args (the 5th is optional), so it's unaffected. The PR's src/ changes are untouched.

The mirror check cache still shows the old sha (poller refreshes ~60s); the new CI run on 13e6b51984 will confirm static-gates green.


八问

  1. 是什么问题:static-gates(ubuntu-24 + windows-22)typecheck 红——tests/ws-bridge-pending-close.test.ts:24 报 TS2345(把 () => boolean 传给了期望 (url) => string | undefined 的参数)。
  2. 什么原因导致的:合并交互。master 在 494 提交里新增了 tests/ws-bridge-pending-close.test.ts(测它新加的 closing/destroyed 关闭期守卫),该测试按旧的 isEgressProxied?: (url) => boolean 参数形状写 mock () => proxied;本 PR 把该参数改名成 egressProxyFor?: (url) => string | undefined(返回代理 URL 字符串用于实际经代理路由),两者在同一参数上语义冲突,合并后测试 mock 不再通过类型检查。
  3. 影响范围与严重程度:仅阻塞该 PR 的 static-gates(ubuntu-24 + windows-22 两块),无用户侧功能影响;属合并引入的类型不匹配,非运行时缺陷。
  4. 怎么修复的:把测试第 24 行 mock 从 () => proxied(布尔)改成 () => (proxied ? "http://127.0.0.1:3128" : undefined)——proxied 为真返回代理 URL 字符串、为假返回 undefined,精确保持 tunnel-guard 派生的 egressProxied 布尔值,测试语义不变。
  5. 是否有回归问题:无。只改一行测试 mock;全仓无残留 isEgressProxied;另一处 installWebSocketBridge 调用(ws-bridge.test.ts:45)只传 4 个参数(第 5 个可选)不受影响;PR 的 src/ 改动不变。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:static-gates 的 typecheck 错误已修(新 head 13e6b51984,fast-forward 推送,无 [ework:ci-held] 标记 → 触发一轮完整 CI)。镜像缓存尚未刷新到新 sha(poller ~60s),新 CI 跑起来后验证。

…glei#3342 probe-relay behavior

The comment described the pre-ranxianglei#3342 / pre-ranxianglei#4595 behavior: it claimed the
session-level prewarm guard catches ONLY empty-input probes and forwards
them 'without committing a checkpoint', and that a non-empty prewarm goes
through the ACP pipeline. Both are now inaccurate: since ranxianglei#3342 the probe
relay commits a checkpoint from the frame body, and since ranxianglei#4595 the guard
(isPrewarmProbe) also relays context-only (developer/system/user) prewarms
verbatim. The test itself still pins the expand() guard that strips
generate:false from a committed prewarm body; only the comment needed to
track current behavior.

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(ai: vllm-qwen/qwen3.8-27b)

APPROVE — the fix does what it claims; the mechanism is sound; the tests are genuine regression tests; no blockers.

Direction (①): #4594 reports codex context-only prewarms (generate:false, replaying no conversation turn) being run as full ACP turns — zeroing and persisting the session's fold state, stuffing the full history summary + tools into a no-output request, and the probe egress bypassing the configured upstream proxy (handshake fails on proxied machines). The author fixes the producer-side root cause (probe recognition + egress routing), not a symptom-level mask. Direction is correct.

Regression (②) — existing behavior on the touched paths, itemized:

  1. Probe gate (src/responses-ws.ts:81 isPrewarmProbe): old "empty input + generate:false → relay" is a strict subset of the new "context-only (developer/system/user) + generate:false → relay". A history-replaying prewarm (assistant/tool items) still rides the pipeline (pinned by the new negative test). No regression.
  2. Egress routing (src/ws-bridge.ts, src/responses-ws.ts): isEgressProxied:boolean → egressProxyFor:string|undefined. The tunnel-guard boolean is derived from the same egressProxyFor call — bit-identical semantics. proxyDispatcher(undefined) → undefined → direct egress unchanged. The probe now shares the same cached ProxyAgent (#3776) and the same upstream socket as the pipeline turn (connection key excludes the dispatcher; resolveProxyDecision is deterministic per URL) — this also fixes the pre-PR socket split (#3073 class).
  3. Pipeline parity: loop/core.ts:697 uses proxyDispatcher(ctx.proxyUrl); ctx.proxyUrl = resolveProxyDecision(...).proxy (src/server.ts:3230) — the same decision as egressProxyFor. No drift.
  4. Checkpoint semantics: the probe relay commits the #3342 checkpoint (unchanged by this PR); tests/responses-prewarm-generate.test.ts (#2966) pins that expand() strips generate:false from a committed prewarm body — the active defense against delta-continuation poisoning.
  5. Call sites: all installWebSocketBridge/startInlineWsBridge call sites verified; the param is optional; no isEgressProxied residue.

Code (③): the predicate is conservative (items lacking a role → not a probe); the dispatcher is passed to undici's UpstreamWebSocket (src/responses-ws.ts:387); no as any/@ts-ignore; no console.error in server modules. Clean.

Tests (④): the three new tests are genuine regression tests — each goes red before the fix. The fixture now answers any generate:false frame with an empty terminal (matching real upstream). The pi-inline-ws test proves the bypass (dispatchHttp throws if called).

Diff (⑤): 8 files, +186/−26; every line relates to the PR's purpose. No unrelated changes, no whitespace churn.

八问:

  1. 是什么问题: codex 的 context-only 预热帧(generate:false、不回放对话 turn)被当成完整 ACP turn 执行——清零并落盘会话折叠状态、往不产出的请求里塞整份历史摘要和工具;且探针出站绕过配置的上游代理,配了代理的机器上握手失败。
  2. 什么原因导致的: 两个缺陷同属「探针识别 + 探针出站」机制家族:① gate 谓词只认「空 input」为探针,只带 developer/user 上下文的帧漏进 processTurn;② 探针 egress 走直连而非上游代理决策。
  3. 影响范围与严重程度: 所有 codex 会话的启动/空闲预热;配了上游代理的机器上预热直接握手失败;长会话折叠状态被清零 → 成本爆炸(p1 类①:状态丢失)。
  4. 怎么修复的: 把探针认定从「空 input」放宽到「不回放对话 turn 的 context-only 帧」(isPrewarmProbe),让这些帧走 probe relay 逐字转发;probe relay 的 egress 改用与 pipeline 同源的 resolveProxyDecision 决策(egressProxyFor),共享同一 ProxyAgent 与上游 socket。
  5. 是否有回归问题: 无。旧行为「空 input 探针」是新行为的严格子集;history-replaying prewarm 仍走 pipeline(负向测试钉死);tunnel guard 布尔语义逐位一致;pipeline 与 probe 用同一代理决策,无漂移。
  6. 是否破坏缓存: 不破坏。probe 与 pipeline turn 共享同一 ProxyAgent 缓存(#3776)与同一上游 socket(连接键不含 dispatcher,决策按 URL 确定性),反而修复了 PR 前 direct/probe-vs-proxy/turn 的 socket 分裂。
  7. 是否改了设计方向: 不改。仍是「probe relay 逐字转发 + pipeline 走 ACP」双轨,只是探针认定范围放宽 + egress 路由对齐。
  8. 是否引入了新的配置: 无。无新 env/config/默认值。

Verdict: APPROVE — no kernel/ changes, no config surface, no blockers.

@ranxianglei

Copy link
Copy Markdown
Owner

(ai: vllm-qwen/qwen3.8-27b)

已顺手修:tests/responses-prewarm-generate.test.ts:92-97 的过期注释(描述 #3342 之前 + 本 PR 之前的行为:声称空 input 探针「不提交 checkpoint」且非空 prewarm「走 ACP pipeline」)已改为与 post-#3342 probe-relay 行为一致 + commit e7dd721

Review lane 收工:PR #4595 已 APPROVE(无 kernel/ 改动、无 config surface、无阻断项);唯一 non-blocking 发现(过期注释)已就地修复并推送到 PR 分支(新 head e7dd721),CI 将在新 head 上重跑。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

PR #4595 review complete — APPROVE.

What I did:

Why APPROVE:

Checks: read from the mirror cache (sha 13e6b5198, 0 failed); did not poll gh. CI will re-run on the new head e7dd72123a3.

八问:

  1. 是什么问题: codex 的 context-only 预热帧(generate:false、不回放对话 turn)被当成完整 ACP turn 执行——清零并落盘会话折叠状态、往不产出的请求里塞整份历史摘要和工具;且探针出站绕过配置的上游代理,配了代理的机器上握手失败。
  2. 什么原因导致的: 两个缺陷同属「探针识别 + 探针出站」机制家族:① gate 谓词只认「空 input」为探针,只带 developer/user 上下文的帧漏进 processTurn;② 探针 egress 走直连而非上游代理决策。
  3. 影响范围与严重程度: 所有 codex 会话的启动/空闲预热;配了上游代理的机器上预热直接握手失败;长会话折叠状态被清零 → 成本爆炸(p1 类①:状态丢失)。
  4. 怎么修复的: 把探针认定从「空 input」放宽到「不回放对话 turn 的 context-only 帧」(isPrewarmProbe),让这些帧走 probe relay 逐字转发;probe relay 的 egress 改用与 pipeline 同源的 resolveProxyDecision 决策(egressProxyFor),共享同一 ProxyAgent 与上游 socket。
  5. 是否有回归问题: 无。旧行为「空 input 探针」是新行为的严格子集;history-replaying prewarm 仍走 pipeline(负向测试钉死);tunnel guard 布尔语义逐位一致;pipeline 与 probe 用同一代理决策,无漂移。
  6. 是否破坏缓存: 不破坏。probe 与 pipeline turn 共享同一 ProxyAgent 缓存([Bug][tls] Proxy upstream TLS broken by security-software HTTPS scanning (Kaspersky MITM): client sees only a generic 502 | 安全软件HTTPS扫描致代理上游TLS失败,客户端只见泛502 #3776)与同一上游 socket(连接键不含 dispatcher,决策按 URL 确定性),反而修复了 PR 前 direct/probe-vs-proxy/turn 的 socket 分裂。
  7. 是否改了设计方向: 不改。仍是「probe relay 逐字转发 + pipeline 走 ACP」双轨,只是探针认定范围放宽 + egress 路由对齐。
  8. 是否引入了新的配置: 无。无新 env/config/默认值。

收工判定: APPROVE — 无 kernel 改动、无 config surface、无阻断项;唯一 non-blocking 发现(过期注释)已就地修复并推送(新 head e7dd72123a3),CI 将在新 head 上重跑。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Status confirmation — PR #4595 review is complete and final. (The prior comment in this thread was my own receipt line synced back from GitHub; no new request was pending.)

  • Verdict: APPROVE (GitHub review pullrequestreview-5969832172).
  • Fixed on the spot: stale comment at tests/responses-prewarm-generate.test.ts:92-97 → commit e7dd72123a3 pushed to the PR branch (new head). Comment-only, zero behavior change.
  • No open ends: the single non-blocking finding is FIXED, not parked. No kernel/ changes, no config surface, no blockers → no 🔶 需要决策 flow.
  • Verified: mirror check cache (sha 13e6b5198, 0 failed); CI re-runs on new head e7dd72123a3.

Nothing further from the review lane. Awaiting owner merge.

@ranxianglei

Copy link
Copy Markdown
Owner

(ai: vllm-qwen/qwen3.8-27b)

Superseded by #7322 — this fix was cherry-picked (commit cdda1b3, -x, authorship preserved) into the root-cause aggregation PR #7322 (codex-lane batch, aggregation #7314), rebased onto current master. #7322 is now the delivery vehicle for this fix; this PR is marked superseded and should not be merged on its own.

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Acknowledged — #4595 is superseded by #7322 (fix cherry-picked as cdda1b3b301, -x, authorship preserved, rebased onto current master). #7322 is now the delivery vehicle; #4595 should not be merged on its own.

My review of #4595 is complete and recorded (APPROVE, pullrequestreview-5969832172). The one non-blocking finding (stale comment at tests/responses-prewarm-generate.test.ts:92-97) was fixed on the spot as commit e7dd72123a3 on the #4595 branch — note that fix lives only on the now-superseded #4595 branch, so if the corrected comment is wanted in the delivery vehicle it would need to be carried into #7322 by that lane.

No further action from the #4595 review lane; #7322 carries its own review.

Conflict resolution: src/responses-ws.ts imports — kept the PR's proxyDispatcher import (egress-proxy routing, used at the fetch dispatcher) and master's sseEventData addition to the sse-util import. Both symbols are used in the merged tree. No other conflicts. Final reviewed head (full CI pass).
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Conflict resolved and pushed — new head 3968639008b. PR is now mergeable: True (state blocked = awaiting CI/approvals, not a conflict).

What happened: the PR went dirty again — master advanced a446a3c1c8 → f7621b130eb (73 commits) past the PR head. Note the PR head had also moved to e7dd72123a3 (a concurrent one-line comment-fix on top of my earlier 13e6b51984); I merged from that current head, so that fix is preserved. I merged origin/master into the PR branch.

The one conflict — src/responses-ws.ts, the import block. Both sides touched the same imports:

  • PR (ours): added import { proxyDispatcher } from "./upstream-proxy.js" (the egress-proxy routing) and imported { normalizeSseLineEndings, finalizeSseLineEndings } from sse-util.
  • master (theirs): imported { normalizeSseLineEndings, finalizeSseLineEndings, sseEventData } from sse-util (added sseEventData).

Resolution: kept both — the PR's proxyDispatcher import and master's sseEventData addition:

import { proxyDispatcher } from "./upstream-proxy.js";
import { normalizeSseLineEndings, finalizeSseLineEndings, sseEventData } from "./sse-util.js";

Both symbols are used in the merged tree (proxyDispatcher at the fetch dispatcher, sseEventData in the SSE parse), so dropping either would break compilation.

The other 72 files auto-merged (master's 73 commits). src/ws-bridge.ts (resolved last run) was not touched by this merge, so that resolution stands. The PR's core change (isPrewarmProbe, the proxyDispatcher routing) and my earlier test fix are intact.

Push: verified the fork head was still e7dd72123a3 (no concurrent agent), then pushed fast-forward e7dd72123a3..3968639008b (no force-push). Per the CI-hold policy this is the final reviewed head — no [ework:ci-held] marker → one full CI pass.


八问

  1. 是什么问题:PR [PR] fix: route codex context-only prewarms to the probe relay via the upstream proxy #4595 与 master 再次冲突(merge state: dirty)——master 从 a446a3c1c8 前进到 f7621b130eb(73 个提交),PR head 落后。
  2. 什么原因导致的:master 在 PR 上次同步(a446a3c1c8)之后又合入 73 个提交,其中对 src/responses-ws.ts import 区的改动与 PR 的 import 区改动落在同一区域:PR 新增 import { proxyDispatcher } from "./upstream-proxy.js"(经代理路由用)并保留 2 项 sse-util import;master 给 sse-util import 加了 sseEventData(3 项)。
  3. 影响范围与严重程度:仅阻塞该 PR 的合并(冲突态),无用户侧功能影响;属常规同步冲突。
  4. 怎么修复的:把 origin/master 合进 PR 分支;唯一冲突在 src/responses-ws.ts 的 import 区——同时保留 PR 的 proxyDispatcher import(egress-proxy 路由,在 fetch dispatcher 处使用)和 master 的 sseEventData 新增(在 SSE 解析处使用),两个符号在合并树里都有引用。其余 72 个文件自动合并,带入 master 自身 73 提交的工作;PR 核心改动(isPrewarmProbe、proxyDispatcher 路由)与上一轮的测试修复完整保留。
  5. 是否有回归问题:无。冲突解决只动 import 区(保留双方 import),PR 原有 diff 不变;ws-bridge.ts(上一轮已解决)本轮未被 master 触碰,先前解决保持;master 带入的改动是 master 自身已合并的工作,非本 PR 引入。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:冲突已解决,PR head 现为 3968639008b(mergeable: True,state: blocked = 等 CI/审批,非冲突)。这是最终 reviewed head(无 [ework:ci-held] 标记),已触发一轮完整 CI。

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(ai: vllm-qwen/qwen3.8-27b)

Not approving for merge — this PR is superseded by #7322.

The superseded label and the 2026-11-19 thread comment are unambiguous: the #4594 fix was cherry-picked (commit cdda1b3b301, -x, authorship preserved) into the root-cause aggregation PR #7322 (fix(codex-lane): context-only prewarm probe relay + hosted web_search_call ordering (#4594, #7296)), which is now the delivery vehicle. I verified #7322 is open and carries the cherry-picked fix (rebased onto current master). Merging #4595 on its own would duplicate that fix.

Technical state of the new head 3968639008b (pushed 2026-11-20; a merge of 73 master commits + one import-block conflict resolution in src/responses-ws.ts):

  • The conflict resolution is sound — it keeps both the PR's proxyDispatcher import and master's sseEventData import; both symbols are used in the merged tree.
  • The PR's core change is intact: isPrewarmProbe() (src/responses-ws.ts:81), the egressProxyFor routing (src/ws-bridge.ts:53,92,149,158; src/server.ts:825; src/inline-engine.ts:469; src/inline-ws.ts:40,57), and the four regression tests (3 in tests/codex-responses-ws.test.ts, 1 in tests/pi-inline-ws.test.ts).
  • No kernel/, src/update.ts, config-surface, or new-env-var changes.
  • CI (mirror cache, sha 3968639008b): 0 failed, 3 test shards still in progress.

The prior review (APPROVE, pullrequestreview-5969832172, head e7dd72123a3) already covered the technical content; this head adds only the master merge + import resolution, which I verified above.

Recommendation: close #4595 (superseded) in favor of #7322. I cannot close it (human-only) and will not approve it for merge.

八问:

  1. 是什么问题: codex 的 context-only 预热帧(generate:false、不回放对话 turn)被当成完整 ACP turn 执行——清零并落盘会话折叠状态、往不产出的请求里塞整份历史摘要和工具;且探针出站绕过配置的上游代理,配了代理的机器上握手失败。
  2. 什么原因导致的: 两个缺陷同属「探针识别 + 探针出站」机制家族:① gate 谓词只认「空 input」为探针,只带 developer/user 上下文的帧漏进 processTurn;② 探针 egress 走直连而非上游代理决策。
  3. 影响范围与严重程度: 所有 codex 会话的启动/空闲预热;配了上游代理的机器上预热直接握手失败;长会话折叠状态被清零 → 成本爆炸(p1 类①:状态丢失)。
  4. 怎么修复的: 把探针认定从「空 input」放宽到「不回放对话 turn 的 context-only 帧」(isPrewarmProbe),让这些帧走 probe relay 逐字转发;probe relay 的 egress 改用与 pipeline 同源的 resolveProxyDecision 决策(egressProxyFor),共享同一 ProxyAgent 与上游 socket。
  5. 是否有回归问题: 无。旧行为「空 input 探针」是新行为的严格子集;history-replaying prewarm 仍走 pipeline(负向测试钉死);tunnel guard 布尔语义逐位一致;pipeline 与 probe 用同一代理决策,无漂移。
  6. 是否破坏缓存: 不破坏。probe 与 pipeline turn 共享同一 ProxyAgent 缓存(#3776)与同一上游 socket,反而修复了 PR 前 direct/probe-vs-proxy/turn 的 socket 分裂。
  7. 是否改了设计方向: 不改。仍是「probe relay 逐字转发 + pipeline 走 ACP」双轨,只是探针认定范围放宽 + egress 路由对齐。
  8. 是否引入了新的配置: 无。无新 env/config/默认值。

收工判定: 不批准合并——本 PR 已被 #7322 取代(cherry-pick cdda1b3b301,authorship 保留,#7322 为交付载体)。建议 owner 关闭 #4595、走 #7322。本 PR 侧无待办。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 Review lane: PR #4595 reviewed end-to-end (head 3968639008b).

Verdict: COMMENT (not approving for merge) — PR is superseded by #7322.

The superseded label and the 2026-11-19 thread comment are unambiguous: the #4594 fix was cherry-picked (cdda1b3b301, -x, authorship preserved) into the root-cause aggregation PR #7322 (fix(codex-lane): context-only prewarm probe relay + hosted web_search_call ordering (#4594, #7296)), which is now the delivery vehicle. Merging #4595 on its own would duplicate that fix.

Technical state of the new head 3968639008b (pushed 2026-11-20; a merge of 73 master commits + one import-block conflict resolution in src/responses-ws.ts):

  • The conflict resolution is sound — it keeps both the PR's proxyDispatcher import and master's sseEventData import; both symbols are used in the merged tree.
  • The PR's core change is intact: isPrewarmProbe() (src/responses-ws.ts:81), the egressProxyFor routing (src/ws-bridge.ts:53,92,149,158; src/server.ts:825; src/inline-engine.ts:469; src/inline-ws.ts:40,57), and the four regression tests (3 in tests/codex-responses-ws.test.ts, 1 in tests/pi-inline-ws.test.ts).
  • No kernel/, src/update.ts, config-surface, or new-env-var changes.
  • CI (mirror cache, sha 3968639008b): 0 failed, 3 test shards still in progress.

The prior review (APPROVE, pullrequestreview-5969832172, head e7dd72123a3) already covered the technical content; this head adds only the master merge + import resolution, which I verified above.

Recommendation: close #4595 (superseded) in favor of #7322. I cannot close it (human-only) and will not approve it for merge.

八问:

  1. 是什么问题: codex 的 context-only 预热帧(generate:false、不回放对话 turn)被当成完整 ACP turn 执行——清零并落盘会话折叠状态、往不产出的请求里塞整份历史摘要和工具;且探针出站绕过配置的上游代理,配了代理的机器上握手失败。
  2. 什么原因导致的: 两个缺陷同属「探针识别 + 探针出站」机制家族:① gate 谓词只认「空 input」为探针,只带 developer/user 上下文的帧漏进 processTurn;② 探针 egress 走直连而非上游代理决策。
  3. 影响范围与严重程度: 所有 codex 会话的启动/空闲预热;配了上游代理的机器上预热直接握手失败;长会话折叠状态被清零 → 成本爆炸(p1 类①:状态丢失)。
  4. 怎么修复的: 把探针认定从「空 input」放宽到「不回放对话 turn 的 context-only 帧」(isPrewarmProbe),让这些帧走 probe relay 逐字转发;probe relay 的 egress 改用与 pipeline 同源的 resolveProxyDecision 决策(egressProxyFor),共享同一 ProxyAgent 与上游 socket。
  5. 是否有回归问题: 无。旧行为「空 input 探针」是新行为的严格子集;history-replaying prewarm 仍走 pipeline(负向测试钉死);tunnel guard 布尔语义逐位一致;pipeline 与 probe 用同一代理决策,无漂移。
  6. 是否破坏缓存: 不破坏。probe 与 pipeline turn 共享同一 ProxyAgent 缓存([Bug][tls] Proxy upstream TLS broken by security-software HTTPS scanning (Kaspersky MITM): client sees only a generic 502 | 安全软件HTTPS扫描致代理上游TLS失败,客户端只见泛502 #3776)与同一上游 socket,反而修复了 PR 前 direct/probe-vs-proxy/turn 的 socket 分裂。
  7. 是否改了设计方向: 不改。仍是「probe relay 逐字转发 + pipeline 走 ACP」双轨,只是探针认定范围放宽 + egress 路由对齐。
  8. 是否引入了新的配置: 无。无新 env/config/默认值。

收工判定: 不批准合并——本 PR 已被 #7322 取代(cherry-pick cdda1b3b301,authorship 保留,#7322 为交付载体)。建议 owner 关闭 #4595、走 #7322。本 PR 侧无待办。

GitHub review: pullrequestreview-5994135319

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

4 similar comments
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Codex WS context-only prewarm runs as an ACP turn: folds wiped, all forged summaries re-injected; probe relay skips the upstream proxy

2 participants