Skip to content

fix(acp): persist embedded text resources for transcript replay - #12585

Open
XIQIXIQIXIQI wants to merge 72 commits into
mainfrom
codex/issue-12538-embedded-resource-replay
Open

XIQIXIQIXIQI wants to merge 72 commits into
mainfrom
codex/issue-12538-embedded-resource-replay

Conversation

@XIQIXIQIXIQI

@XIQIXIQIXIQI XIQIXIQIXIQI commented Sep 24, 2026 •

Copy link
Copy Markdown
Collaborator

What this PR does

Persists bounded, original ACP embedded text resource blocks with their owning user prompt and replays them as typed user chunks. The daemon UI SDK now exposes these blocks separately from resource_link and native attachments, including offline transcript projection. Native attachment expansions are identified by their input positions, so a separate direct resource remains replayable even when it shares the same URI.

Why it's needed

Qwen Code already accepts embedded resources as model context, but refresh or restart previously restored only the prompt text. Clients therefore lost the original context selection while the assistant reply remained visible. This closes the persistence/replay gap tracked in #12538; it is separate from the resource_link fix in #12088.

Reviewer Test Plan

How to verify

Send an ACP prompt with text and a text resource, then a resource-only prompt and a later text-only prompt. Inspect the saved user records and load the transcript: each resource should appear as a typed user chunk on its original prompt with its URI, MIME type, text, metadata, prompt ID, and source-record identity; it must not migrate to the later turn or be fetched by URI. Also check rewind/active-branch replay, native attachments over 256 KiB, a direct resource sharing a native attachment URI, and retention-only handling of an oversized direct text resource (the prompt is admitted, the resource reaches the model live, and the durable record drops it with embeddedResourcesTruncated set). Focused bridge, CLI, persistence, and SDK tests cover these cases.

Evidence (Before & After)

Before: the new recording/replay regressions failed against the pre-fix source because typed resources were absent after replay. After: the focused tests pass, including resource-only, same-URI coexistence, and the oversized-resource retention-only drop. No running-daemon or browser recording is claimed here.

Tested on

OS Status
🍏 macOS ✅ build, bundle, typecheck, targeted lint and tests
🪟 Windows ⚠️ not tested locally
🐧 Linux ⚠️ not tested locally

Environment (optional)

Node 22.23.1; package-level test scripts and bundled dist/cli.js sanity check. A live ACP daemon session was not run.

Risk & Scope

  • Main risk or tradeoff: directly supplied embedded text resources add bounded user content to durable transcripts; the aggregate serialized replay budget is 256 KiB per prompt. Oversized direct text resources are admitted and delivered to the model live; only their durable retention is dropped, marked with embeddedResourcesTruncated — the prompt always runs. Native attachments use their existing separate storage path; current daemon requests exclude them by provenance. Legacy metadata without positions excludes only unambiguous URI matches; ambiguous blocks are retained and may hit the inline limit.
  • Not validated / out of scope: running-daemon JSONL and session/load readback after restart, browser refresh acceptance, persistence of inline blob resources, and recovery of resources absent from historical records. Downstream clients must render the new typed SDK field to display a card.
  • Breaking changes / migration notes: no record migration is required; old records remain readable. The new text-resource replay limit bounds durable retention only; it does not reject prompts or change their parameters.

Design: English · 简体中文.

Linked Issues

Fixes #12538.

中文说明

本 PR 做了什么

将有大小上限的原始 ACP 嵌入式文本 resource 块与所属用户 Prompt 一同持久化,并在回放时恢复为带类型的用户消息块。Daemon UI SDK 将其与 resource_link、原生附件区分,并支持离线 transcript 投影。原生附件展开结果按输入位置识别,因此即使另一个直接资源与它使用相同 URI,也仍可独立回放。

为什么需要

Qwen Code 已能把嵌入式资源用作模型上下文,但刷新或重启后此前只恢复 Prompt 文字。客户端因此丢失原始上下文引用,尽管助手回复仍可见。本 PR 修复 #12538 的持久化与回放缺口;它不同于 #12088 的 resource_link 修复。

审查验证计划

如何验证

依次发送包含文字和文本资源的 ACP Prompt、纯资源 Prompt、以及后续纯文字 Prompt。检查保存的用户记录与加载后的 transcript:每个资源都应作为带类型的用户消息块保留在原轮次,包含 URI、MIME 类型、文本、元数据、Prompt ID 和来源记录身份;不得迁移到后续轮次,也不得根据 URI 抓取内容。另需检查 rewind/有效分支回放、超过 256 KiB 的原生附件、与原生附件同 URI 的直接资源,以及超限直接文本资源的仅保留层处理(Prompt 被放行、资源实时送达模型,持久记录丢弃该资源并置 embeddedResourcesTruncated)。定向桥接、CLI、持久化和 SDK 测试覆盖这些场景。

前后证据

修复前:新增的录制与回放回归测试在旧源码上失败,类型化资源在回放后缺失。修复后:定向测试通过,包括纯资源、同 URI 并存和超限资源的仅保留层丢弃。本 PR 未声称完成运行中 Daemon 或浏览器录屏。

已测试系统

系统 状态
🍏 macOS ✅ 构建、打包、类型检查、定向 lint 与测试
🪟 Windows ⚠️ 未在本地测试
🐧 Linux ⚠️ 未在本地测试

环境(可选)

Node 22.23.1;包级测试脚本与打包后的 dist/cli.js 基础检查。未运行真实 ACP Daemon 会话。

风险与范围

  • 主要风险或取舍:直接提供的嵌入式文本资源会作为有界的用户内容进入持久 transcript;每个 Prompt 的序列化资源回放预算为 256 KiB。超限直接文本资源会被放行并实时送达模型;仅其持久保留被丢弃,并以 embeddedResourcesTruncated 标记——Prompt 始终执行。原生附件沿用独立存储路径;新 Daemon 请求按来源排除。缺少位置字段的旧元数据只排除 URI 唯一对应的块;歧义块会保留,也可能触及内联限额。
  • 未验证/范围外:运行中 Daemon 的 JSONL 与重启后 session/load 回读、浏览器刷新验收、内联 blob 资源持久化,以及从未保存资源的旧记录恢复。下游客户端需要渲染新的类型化 SDK 字段才能显示卡片。
  • 破坏性变更/迁移:无需记录迁移,旧记录继续可读。新的文本资源回放上限只约束持久保留,不拒绝 Prompt,也不改变其参数。

设计文档:English · 简体中文。

关联 Issue

Fixes #12538。

@github-actions github-actions Bot added the review/self-reported The linked issue was opened by the PR author (self-reported) label Sep 24, 2026
@XIQIXIQIXIQI

Copy link
Copy Markdown
Collaborator Author

Verification report

Status: VERIFIED_FIXED by focused test-script fallback, not by a live ACP daemon E2E run.

  • Baseline: the new recording/replay regression cases failed before the fix because typed embedded resources were absent from durable replay.
  • Final source: npm run build, npm run bundle, npm run typecheck, and targeted ESLint passed on macOS / Node 22.23.1.
  • Focused tests: ACP bridge replay/admission 134/134, CLI Session 10/10, core recording 1/1, daemon SDK 4/4 passed. Coverage includes resource-only prompts, same-URI direct/native coexistence, legacy metadata ambiguity, 257 native references, and rejection of oversized direct text before peer echo.
  • Bundled CLI sanity: node dist/cli.js --version returned 0.24.4 (exit 0).

Remaining runtime acceptance: launch a real ACP daemon, submit a typed resource, read back the persisted JSONL and paginated transcript/session/load after restart, then verify the downstream page after refresh. None of those runtime, deployment, or browser checks is claimed by this report. Existing records that never stored a resource cannot be reconstructed retroactively.

@yiliang114 yiliang114 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Blocking on one regression that CI caught and I traced to this diff:

Test (ubuntu-latest) is red on this head, and the failure belongs to this PR: acp-bridge > bridge.test.ts > sendPrompt > does not leak a prompt slot for a malformed null block now fails with AssertionError: expected TypeError: Cannot read properties of null… to not be an instance of TypeError. That test is pre-existing (this PR doesn't touch it), and on main the malformed null block in a prompt reaches the typed-error path. The new resolvedPrompt.flatMap((block, index) => block.type === 'resource' && …) reads .type on every element — a null element (which the malformed-block case legitimately produces) throws a bare TypeError before any typed error handling runs. Same for the new dispatchBlocks.filter(isSessionAttachmentReference) if that predicate doesn't null-guard.

The fix is small: null-guard the flatMap predicate (block != null && block.type === 'resource'), or run the snapshot computation after the existing malformed-input handling. Whichever, the red test must go green again.

The rest of the change (persisting embedded text resources for replay, the provenance-by-position tracking, the SDK normalizer types) reads fine — holding approval only on this regression.

@yiliang114

Copy link
Copy Markdown
Collaborator

Rechecked the existing feedback at ca51099bed336648048085353ee120d2dc6b92f6.

The earlier null-block P1 is fixed: the resource-index predicate now guards null, and the attachment-reference predicate also handles null safely. That thread can be resolved; the executed repository CI gates on this head are green.

R1-1 still needs a scope decision. The 256 KiB replay budget currently rejects the entire prompt, including from Session.prompt() before recording begins. That is a new input-admission limit, beyond retaining bounded history. My preference is to preserve prompt execution and make any retention loss explicit; if rejecting the prompt is the intended product contract, it needs an explicit maintainer decision before this merges.

This is a source recheck of existing feedback, not a fresh full review or live-daemon acceptance run. The PR already discloses that restart/session/load and browser-refresh acceptance remain unverified.

@wenshao

wenshao commented Sep 27, 2026

Copy link
Copy Markdown
Collaborator

@qwen-code /triage

@wenshao

wenshao commented Sep 27, 2026

Copy link
Copy Markdown
Collaborator

@qwen-code /takeover from 2

@qwen-code-dev-bot qwen-code-dev-bot added the autofix/takeover Summon the autofix loop to manage this PR (remove to release; needs triage+) label Sep 27, 2026
@qwen-code-dev-bot

Copy link
Copy Markdown
Collaborator

🤝 Takeover engaged: the autofix loop now manages this PR — it will address new review feedback and resolve base conflicts until the label is removed or the round cap is reached. This window's round counter starts at 2 (the rounds this PR spent in review before takeover), so the Critical-only brake engages after 3 more change-producing round(s) instead of a full fresh 5. Remove the autofix/takeover label (or comment @qwen-code /takeover stop) to release.

中文说明

🤝 已接管:autofix 循环现在管理此 PR —— 将持续处理新的评审反馈与 base 冲突,直到移除标签或达到轮次上限。本窗口轮次计数从 2 起算(即本 PR 托管前已进行的评审轮数),因此再经过 3 个产生改动的轮次即进入 Critical-only,而非重新计满 5 轮。移除 autofix/takeover 标签(或评论 @qwen-code /takeover stop)即可释放。

@qwen-code-dev-bot

qwen-code-dev-bot commented Sep 27, 2026 •

Copy link
Copy Markdown
Collaborator

✅ AutoFix round 8 finished — view run. See this round's report below.

中文说明

✅ AutoFix 第 8 轮已完成 —— 查看运行。本轮报告见下方。

qwen-code-autofix and others added 2 commits September 27, 2026 10:56
…edded-resource-replay

# Conflicts:
#	packages/acp-bridge/package.json
The 256 KiB bound was enforced as an admission limit that rejected the
whole prompt (daemon pre-echo and child pre-recording), including on the
direct-ACP path with no journal at all. Bound retention instead: the
prompt always runs, blocks beyond the byte budget (256 KiB) or count cap
(256, matching the live echo bound) are dropped from the durable record,
and the user record marks the loss with embeddedResourcesTruncated.

- capture guard skips empty-URI resources, mirroring the replay validator
- daemon no longer charges the raw request body; the child accounts once,
  on the schema-normalized blocks it persists (single derivation)
- parse daemon attachment references once per prompt under a shared bound
  so 257+ references stay recorded instead of vanishing from the record
- linearize readDaemonNativeResourceIndexes (Set/Map instead of
  includes/some per index)
- a resource-only record's empty displayText no longer flips the session
  title projection off (channel '' markers are preserved)

Updates both design docs and inverts the three throw-pinning tests to
assert admission plus bounded retention.

Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com>
@qwen-code-dev-bot

Copy link
Copy Markdown
Collaborator

🤖 Addressed the latest review feedback (round 3/100). What changed, and what I pushed back on: · 已处理最新评审反馈(第 3/100 轮)。改动内容与我反驳保留之处如下:

Round summary — PR #12585

Merge

--conflict true: merged origin/main (e0b8bea) into the branch. The only content conflict was packages/acp-bridge/package.json — the PR added the ./embeddedResourceReplay export while main added ./sessionStartupConfig and bumped the version to 0.24.6. Resolved by keeping both export entries and taking main's version. All other overlapping files auto-merged; the merge result was verified by a full build, typecheck, and the focused suites listed below.

Findings addressed in code

  • [rc:4089460670] P1 null-guard (blocking CI failure) — Already fixed at the head commit (block != null && … in the daemon predicate; isSessionAttachmentReference null-guards). Re-verified after the merge: the full acp-bridge suite (2574 tests, including bridge.test.ts > does not leak a prompt slot for a malformed null block) passes. Resolving.
  • [rc:4094627427] R1-1 Critical — budget is now a retention bound, not an admission gate. snapshotReplayableEmbeddedResources no longer throws; it retains what fits and reports truncated. The prompt always runs; over-budget blocks still reach the model; the user record marks the loss with embeddedResourcesTruncated: true (payload type extended in core). This matches the maintainer's stated preference ("preserve prompt execution and make any retention loss explicit"). The mid-turn settle-then-promote door can no longer fail on the budget because nothing throws. Both design docs (EN + zh-CN) updated to the retention semantics, and the three throw-pinning tests inverted to assert admission plus bounded retention, per the finding's acceptance criterion.
  • [rc:4094627506] R1-22 — daemon no longer charges the raw request body. The daemon-side snapshot call (whose result was only ever used for the throw) is removed; the daemon still computes and forwards the position-based attachmentResourceIndexes, but the only byte accounting now happens in the child on the zod-normalized blocks it actually persists — one number, on the durable shape.
  • [rc:4094627544] R1-2 — single derivation. The duplicate snapshot/parse chain in Session.prompt() is removed; the decision is derived once in #executePromptInner, where the retained value is actually used for recording. No failed-turn record can be journaled for an over-budget prompt because the admission throw no longer exists.
  • [rc:4094627466] R1-17 — capture guard mirrors the replay validator. Resources without a non-empty URI are skipped (neither retained nor charged), byte-identical in effect to the replay filter at transcript-replay.ts:1014-1022. New unit case: [{uri:'', 200 KiB}, {uri:'context://ok', 100 KiB}] returns only the valid block with truncated: false.
  • [rc:4094627491] R1-6 — count cap. MAX_RECORDED_EMBEDDED_RESOURCES = 256 (matching MAX_ECHO_CONTENT_BLOCKS and the attachment-reference cap), applied as a retention bound per the finding's note for the R1-1 rework: blocks past the cap are dropped with truncated: true, never rejected.
  • [rc:4094627459] R1-5 — capture-side guard coverage. New packages/acp-bridge/src/embedded-resource-replay.test.ts pins the blob/text selection (blob-only returns [] without throwing) and the malformed shapes (null, {type:'resource'}, resource:null, resource:5, non-string text) against both snapshotReplayableEmbeddedResources and readDaemonNativeResourceIndexes (legacy and position branches), asserting valid siblings are retained and nothing throws. One simplification fell out of probing: the typeof block.resource !== 'object' condition is redundant now that the URI-typeof guard runs first (property access on primitives is safe; only null/undefined can throw and !block.resource covers those) — removing it kept every guard test green, so it is removed rather than left unpinned.
  • [rc:4094627553] R1-12 — linear validator. readDaemonNativeResourceIndexes now uses a Set for seen indexes and a precomputed Set of attachment:/// URIs instead of indexes.includes / references.some(...) per index. The 20 000-entry case runs in ~50 ms (was ~15 s, over the vitest timeout).
  • [rc:4094627548] R1-11 — one parse, one bound. attachmentReferences is parsed once per prompt with Math.max(MAX_DAEMON_ATTACHMENT_REFERENCES, params.prompt.length) and shared by the exemption and the record, so the 257-reference shape admitted by the PR's own test now journals all 257 references instead of writing an empty user record. The default 256 bound is unchanged for the other two read sites.
  • [rc:4094627560] R1-20 — auto-titling preserved for resource-only sessions. The finding's two literal suggestions conflict with the pinned protection at sessionTitle.test.ts:330-352 (['',''] must stay in projection mode so hidden channel context never reaches the title model): the track-site displayText || undefined would silently strip the channel marker, and the consumer-site check would turn that pinned test red. Instead the fix is provenance-aware: a record's displayText: '' is tracked as undefined only when the payload carries embeddedResources (the shape this PR introduced), at both the record and restore call sites. Channel '' markers are preserved (pinned by a new test).

Deferred to the next round (batch budget, ~8 findings/round)

R1-24 (residual acceptance case — note: the updated same-URI bridge test now asserts forwarded indexes [0], which already catches the identity→URI-uniqueness mutation post-R1-1), R1-23, R1-8, R1-26, R1-19, R1-28, R1-7, R1-4, R1-16, R1-29, R1-3, R1-21. Each thread has a reply with the scoped plan; nothing was silently dropped.

Mutation probes (each run locally, then reverted)

  • Count cap removed → bounds the retained resource count RED ✓
  • Empty-URI guard removed → neither retains nor charges a resource with an empty URI RED ✓
  • Byte budget neutralized → drops only the blocks that would exceed the byte budget RED ✓
  • !block.resource removed → malformed-shape case RED with TypeError: Cannot read properties of null ✓
  • text-typeof guard removed → malformed-shape case RED ✓
  • legacy-branch !block.resource removed → legacy-path case RED with TypeError ✓
  • Set/Map reverted to includes/some (quadratic) → validates large position metadata in linear time RED via vitest timeout ✓
  • Recording parse reverted to the 256 default cap → extended 257-reference Session test RED (recordUserMessage payload lost) ✓
  • Truncation marker removed at either record site → does not retain text resources… / advisor witness RED ✓
  • Provenance-blind title tracking → both new autoTitle cases RED; the channel-marker guard case stays green ✓

Test-contract notes

The two inverted bridge cases and the one inverted Session case are declared in test-weakening.json: the rejection assertions were replaced by admission assertions because the rejection itself was the defect (Critical R1-1, maintainer CHANGES_REQUESTED). On the pre-round head these changed tests fail (the code threw), satisfying the verification gate's red-before requirement.

Diff growth this round

Source: +27 net lines (embedded-resource-replay.ts +23, session-control-plane.ts −5, Session.ts −3, chatRecordingService.ts +12). Test: +399 net lines (including the new 195-line unit file). Docs: ±0 net. Within the 400/400 window budgets.

Verification

  • npm run build (COREPACK_HOME=/tmp/corepack-home — the runner's default corepack cache path is EACCES) — passed
  • npm run typecheck — passed
  • npm run lint — passed
  • vitest packages/acp-bridge (full) — 2574 passed (54 files)
  • vitest packages/cli Session.test.ts (full) — 1072 passed
  • vitest packages/core chatRecordingService.test.ts chatRecordingService.autoTitle.test.ts sessionTitle.test.ts — 207 passed
  • vitest packages/sdk-typescript daemon-embedded-resources.test.ts daemon-resource-links.test.ts (merge overlap check) — 18 passed
  • Mutation probes — 10 applied, all red as expected, all reverted (list above)
  • Integration tests: not run — the changed behavior is exercised directly by the package suites above (daemon admission, child recording, replay guard, title tracking), not only through the bundled CLI.
中文说明

本轮总结 — PR #12585

合并

--conflict true:将 origin/main(e0b8bea9e0)合并入本分支。唯一的内容冲突是 packages/acp-bridge/package.json——本 PR 新增了 ./embeddedResourceReplay 导出,而 main 新增了 ./sessionStartupConfig 并把版本升到 0.24.6。解决方式:两个导出都保留,版本号取 main 的。其余重叠文件均自动合并;合并结果通过了下文列出的完整构建、类型检查与聚焦测试套件验证。

本轮在代码中处理的发现

  • [rc:4089460670] P1 空值守卫(阻塞 CI 的失败)——在头提交中已修复(Daemon 谓词中的 block != null && …;isSessionAttachmentReference 有空值守卫)。合并后重新验证:acp-bridge 全量套件(2574 个测试,包含 bridge.test.ts > does not leak a prompt slot for a malformed null block)通过。关闭该线程。
  • [rc:4094627427] R1-1 Critical——预算改为保留上限,不再是准入门槛。 snapshotReplayableEmbeddedResources 不再抛错;保留可容纳的块并报告 truncated。Prompt 始终执行;超预算的块仍送达模型;用户记录以 embeddedResourcesTruncated: true 显式标记损失(core 中的 payload 类型已扩展)。这与维护者明确表述的偏好一致("保留 Prompt 执行、让保留损失显式可见")。mid-turn 的结算后提升路径不会再因预算失败,因为不再有抛错。两份设计文档(英文+中文)已更新为保留语义,三个钉住抛错的测试按发现的验收标准反转为断言准入加有界保留。
  • [rc:4094627506] R1-22——Daemon 不再按原始请求体计费。 Daemon 侧的快照调用(其结果只被用于抛错)已移除;Daemon 仍计算并转发基于位置的 attachmentResourceIndexes,但唯一的字节计费现在发生在子进程一侧,作用于其实际持久化的 zod 规范化块——同一个数字、作用于持久化形状。
  • [rc:4094627544] R1-2——单次推导。 Session.prompt() 中重复的快照/解析链已移除;判定只在 #executePromptInner 中推导一次,即保留值实际用于记录的位置。由于准入抛错不复存在,超预算 Prompt 不可能再被写入失败 turn 记录。
  • [rc:4094627466] R1-17——捕获守卫镜像回放校验器。 没有非空 URI 的资源被跳过(既不保留也不计费),效果与 transcript-replay.ts:1014-1022 的回放过滤器逐字节一致。新增单元用例:[{uri:'', 200 KiB}, {uri:'context://ok', 100 KiB}] 只返回有效块且 truncated: false。
  • [rc:4094627491] R1-6——数量上限。 MAX_RECORDED_EMBEDDED_RESOURCES = 256(与 MAX_ECHO_CONTENT_BLOCKS 及附件引用上限一致),按发现中对 R1-1 改造的说明以保留上限方式应用:超过上限的块被丢弃并置 truncated: true,绝不拒绝。
  • [rc:4094627459] R1-5——捕获侧守卫覆盖。 新增 packages/acp-bridge/src/embedded-resource-replay.test.ts,钉住 blob/text 选择(纯 blob 返回 [] 且不抛错)与畸形形状(null、{type:'resource'}、resource:null、resource:5、非字符串 text),同时覆盖 snapshotReplayableEmbeddedResources 与 readDaemonNativeResourceIndexes(传统与位置两个分支),断言有效兄弟块仍被保留且不抛错。探针过程中得出一个简化:typeof block.resource !== 'object' 条件在 URI 类型检查先行的情况下是冗余的(对原始类型的属性访问是安全的;只有 null/undefined 会抛错,而 !block.resource 已覆盖)——删除它后所有守卫测试保持绿色,因此将其移除而不是留下无钉住的代码。
  • [rc:4094627553] R1-12——线性校验器。 readDaemonNativeResourceIndexes 现在用 Set 记录已见索引、用一次构建的 attachment:/// URI Set 替代逐索引的 indexes.includes / references.some(...)。20 000 条目的用例约 50 ms 完成(原先约 15 s,超过 vitest 超时)。
  • [rc:4094627548] R1-11——一次解析、一个上限。 attachmentReferences 每个 Prompt 只解析一次,上限为 Math.max(MAX_DAEMON_ATTACHMENT_REFERENCES, params.prompt.length),由豁免与记录共享,因此本 PR 自己的测试所承认的 257 引用形态现在会完整写入 257 条引用,而不是写出一条空用户记录。另外两个读取点的默认 256 上限保持不变。
  • [rc:4094627560] R1-20——纯资源会话保留自动标题。 该发现的两个字面建议都与 sessionTitle.test.ts:330-352 钉住的保护冲突(['',''] 必须保持投影模式,隐藏的渠道上下文绝不能送达标题模型):埋点处的 displayText || undefined 会静默移除渠道标记,消费端的检查会让那个被钉住的测试变红。因此改为按来源区分:仅当 payload 携带 embeddedResources(本 PR 引入的形状)时,记录的 displayText: '' 才按 undefined 追踪,记录与恢复两个调用点都生效。渠道 '' 标记被保留(由新测试钉住)。

顺延到下一轮(批次预算,每轮约 8 个发现)

R1-24(剩余验收用例——注意:更新后的同 URI 桥接测试现在断言转发索引 [0],在 R1-1 之后已能捕获身份规则→URI 唯一性规则的变异)、R1-23、R1-8、R1-26、R1-19、R1-28、R1-7、R1-4、R1-16、R1-29、R1-3、R1-21。每个线程都有回复说明范围计划,没有任何发现被静默丢弃。

变异探针(均在本地执行后还原)

  • 移除数量上限 → bounds the retained resource count 变红 ✓
  • 移除空 URI 守卫 → neither retains nor charges a resource with an empty URI 变红 ✓
  • 字节预算失效 → drops only the blocks that would exceed the byte budget 变红 ✓
  • 移除 !block.resource → 畸形形状用例以 TypeError: Cannot read properties of null 变红 ✓
  • 移除 text 类型守卫 → 畸形形状用例变红 ✓
  • 移除传统分支 !block.resource → 传统路径用例以 TypeError 变红 ✓
  • Set/Map 还原为 includes/some(二次方)→ validates large position metadata in linear time 因 vitest 超时变红 ✓
  • 记录解析还原为默认 256 上限 → 扩展后的 257 引用 Session 测试变红(recordUserMessage payload 丢失)✓
  • 任一记录点移除截断标记 → does not retain text resources… / advisor 见证用例变红 ✓
  • 不按来源区分的标题追踪 → 两个新 autoTitle 用例变红;渠道标记守卫用例保持绿色 ✓

测试契约说明

两个反转的桥接用例与一个反转的 Session 用例已在 test-weakening.json 中声明:拒绝断言被准入断言取代,因为"拒绝"本身就是缺陷(Critical R1-1,维护者 CHANGES_REQUESTED)。在本轮前的分支头上这些修改过的测试会失败(当时代码抛错),满足验证门的"先红"要求。

本轮 Diff 增长

源码:净 +27 行(embedded-resource-replay.ts +23,session-control-plane.ts −5,Session.ts −3,chatRecordingService.ts +12)。测试:净 +399 行(含新增的 195 行单元测试文件)。文档:净 ±0。在 400/400 窗口预算内。

验证

  • npm run build(COREPACK_HOME=/tmp/corepack-home——runner 默认的 corepack 缓存路径报 EACCES)——通过
  • npm run typecheck ——通过
  • npm run lint ——通过
  • vitest packages/acp-bridge(全量)——2574 通过(54 个文件)
  • vitest packages/cli Session.test.ts(全量)——1072 通过
  • vitest packages/core chatRecordingService.test.ts chatRecordingService.autoTitle.test.ts sessionTitle.test.ts——207 通过
  • vitest packages/sdk-typescript daemon-embedded-resources.test.ts daemon-resource-links.test.ts(合并重叠检查)——18 通过
  • 变异探针——执行 10 项,全部按预期变红,全部已还原(清单见上)
  • 集成测试:未运行——本轮改动的行为由上述包级套件直接覆盖(Daemon 准入、子进程记录、回放守卫、标题追踪),并非只能经由打包后的 CLI 触达。

Base-conflict check · 基分支冲突检查: conflicted with main — resolved in this push. · 与 main 有冲突——已在本次推送中解决。

🧵 Resolved all 10 selected review thread(s). · 已关闭全部选中的 10 条评审线程。

Re-review when you have a moment. After round 100 this bot stops and leaves the PR for a human. · 有空请复审;第 100 轮后本 bot 停止并将 PR 交给人工。


🧠 Handled by Qwen Code · model/模型 kimi-k3 · CLI 0.24.6

@qwen-code-dev-bot

Copy link
Copy Markdown
Collaborator

🤖 Reviewed the latest feedback — no changes needed. Why, point by point: · 已审阅最新反馈——无需改动。逐点说明原因如下:

Autofix round — no action

What this round found

No actionable code feedback was present for PR #12585 in this round:

  • Reviews: none.
  • Inline comments: none.
  • Issue-level comments: one automated comment from qwen-code-ci-bot (ic:5857791211) reporting that the Qwen Code review workflow did not complete: "Qwen review exited with status 1. A transient error is retried automatically; if you are seeing this, retry with @qwen-code /review."
  • Failed checks: review-pr: FAILURE — the same workflow run the bot comment refers to. No other checks are failing, and there are no still-red checks from before the last evaluation.

Why no code change was made

The failed review-pr check is a transient failure of the repository's own review infrastructure, not a defect in this PR:

  • The review job checks out the trusted base code, not the PR head (.github/workflows/qwen-code-pr-review.yml, step comment: "SECURITY: checkout trusted base code; /review fetches PR diff context").
  • The review runs the released npm CLI (npm install -g '@qwen-code/qwen-code@latest'), not a build of this PR — so nothing in this PR's diff can cause the review agent's exit status 1.
  • The workflow itself classifies this exact outcome as transient ("A transient error is retried automatically") and the bot comment is its standard fallback notice.

Additionally, the review workflow (.github/) is CI machinery that this PR is not about, so modifying it is out of scope for an autofix round regardless. Re-running the check requires GitHub credentials, which this environment does not have; per the bot comment, the failure is retried automatically and a maintainer can also re-trigger it manually by commenting @qwen-code /review on the PR.

Verification

No files were changed and no commit was created, so no build/typecheck/lint/test runs were required. Evidence inspected this round:

  • feedback.md — confirmed zero review findings, zero inline comments, one bot issue-level comment, one failed check (review-pr).
  • .github/workflows/qwen-code-pr-review.yml — confirmed the review job runs the published CLI against base-branch code and labels this failure mode as transient with automatic retry (fallback comment template at line 2803 matches the posted bot comment verbatim).
  • git status — working tree clean; branch codex/issue-12538-embedded-resource-replay is up to date with its origin head.
中文说明

Autofix 本轮 —— 无需操作

本轮发现

本轮 PR #12585 没有任何可执行的代码反馈:

  • Review 意见: 无。
  • 行内评论: 无。
  • Issue 级评论: 仅有一条来自 qwen-code-ci-bot 的自动评论(ic:5857791211),报告 Qwen Code 评审工作流未能完成:"Qwen review exited with status 1. A transient error is retried automatically; if you are seeing this, retry with @qwen-code /review."
  • 失败的检查: review-pr: FAILURE —— 即上述机器人评论所指的同一个工作流运行。没有其他失败的检查,也没有上次评估之前遗留的持续失败检查。

为什么不做代码修改

review-pr 检查失败是仓库自身评审基础设施的一次瞬时故障,而不是本 PR 的缺陷:

  • 评审 job 检出的是可信的基线代码,而非 PR 头部(见 .github/workflows/qwen-code-pr-review.yml 中的步骤注释:"SECURITY: checkout trusted base code; /review fetches PR diff context")。
  • 评审运行的是已发布的 npm CLI(npm install -g '@qwen-code/qwen-code@latest'),而不是由本 PR 构建的产物 —— 因此本 PR 的改动不可能是评审代理以状态码 1 退出的原因。
  • 工作流本身就将这一结果归类为瞬时错误("A transient error is retried automatically"),该机器人评论正是其标准的兜底通知。

此外,评审工作流(.github/)属于 CI 基础设施,与本 PR 的主题无关,无论如何都不在 autofix 轮次的可修改范围内。重新运行该检查需要 GitHub 凭据,而当前环境没有凭据;根据机器人评论,该失败会自动重试,维护者也可以在 PR 下评论 @qwen-code /review 手动重新触发。

验证情况

本轮未修改任何文件、未创建提交,因此无需运行构建/类型检查/lint/测试。本轮核查的证据如下:

  • feedback.md —— 确认没有评审意见、没有行内评论,只有一条机器人的 Issue 级评论,以及一个失败的检查(review-pr)。
  • .github/workflows/qwen-code-pr-review.yml —— 确认评审 job 使用已发布的 CLI 在基线分支代码上运行,并将此类失败标记为会自动重试的瞬时错误(第 2803 行的兜底评论模板与已发布的机器人评论逐字一致)。
  • git status —— 工作区干净;分支 codex/issue-12538-embedded-resource-replay 与其 origin 头部保持一致。

Base-conflict check · 基分支冲突检查: no conflict with main. · 与 main 无冲突。


🧠 Handled by Qwen Code · model/模型 kimi-k3 · CLI 0.24.6

@qwen-code-dev-bot

Copy link
Copy Markdown
Collaborator

🔀 Base updated: red check(s) [review-pr] pass on current main — merged current main via update-branch; CI will re-run.

中文说明

🔀 已更新 base:红色检查 [review-pr] 在当前 main 上通过 —— 已通过 update-branch 合入当前 main,CI 将重新运行。

- core: exempt truncation-only records (displayText '' +
  embeddedResourcesTruncated, no retained resources) from the title
  projection marker, so an over-budget resource-only prompt no longer
  disables auto-titling for the whole session
- cli/acpAgent: strip qwen.daemon.attachmentReferences and
  qwen.daemon.attachmentResourceIndexes from untrusted prompt _meta and
  re-arm both only for a trusted parent, so a direct-ACP client cannot
  exempt its own blocks from the durable journal
- acp-bridge: share the attachment expansion URI format between the
  SessionAttachmentStore producer and both replay matchers via the
  dependency-free attachmentResourceUri() helper
- acp-bridge: project a placeholder chunk on replay when a truncation
  marker leaves a user turn with nothing else projectable, instead of
  emitting zero updates
- cli/Session: record attachmentReferences and resourceLinks on the
  deferred /advisor branch, matching the main recording branch
- acp-bridge tests: pin the forwarded prompt content (not just the
  provenance metadata) for the oversized direct-resource admissions
- docs: state the 256-block echo cap and the 256 KiB journal budget as
  separate bounds in both language versions
@qwen-code-dev-bot

Copy link
Copy Markdown
Collaborator

🤖 Addressed the latest review feedback (round 4/100). What changed, and what I pushed back on: · 已处理最新评审反馈(第 4/100 轮)。改动内容与我反驳保留之处如下:

Round summary

Addressed all 7 findings from the round-2 review (R2-1 through R2-7) in commit 434878e2. Each fix below names its reproduction/mutation-probe evidence; every new guard or branch has a test that goes red when the guard is removed.

Findings

R2-1 (rc:4119082230, Critical) — truncation-only record disabled auto-titling — FIXED

The title-projection exemption in chatRecordingService.trackUserDisplayTextForTitlePayload keyed only on embeddedResources.length, so the truncation-only payload this PR creates ({ displayText: '', hookContext: '', embeddedResourcesTruncated: true }) was tracked as '', flipping hasDisplayProjection on and returning empty_history for the whole session (including after restore). Widened the condition to also exempt embeddedResourcesTruncated === true, keeping the bare { displayText: '' } channel-marker behavior pinned by the existing test.

Evidence: added keeps truncation-only resource records eligible for auto-titling and restores truncation-only resource records without a title projection marker to chatRecordingService.autoTitle.test.ts. Both were written first and confirmed RED on the unfixed code (Tests 2 failed | 27 skipped, received [""] where [undefined] was expected), then green after the fix (29/29). The channel-marker sibling test (keeps an empty channel display projection as a title projection marker) still passes, so the constraint holds.

R2-2 (rc:4119082249, Critical) — untrusted ACP client could self-exempt blocks from the journal — FIXED

acpAgent.ts#prompt deleted ten daemon-only _meta keys but not qwen.daemon.attachmentReferences / qwen.daemon.attachmentResourceIndexes, so a direct-ACP caller could forge both sides of the only validation and exempt arbitrary attachment:/// blocks from the durable record while the content still reached the model. Both keys are now captured, deleted alongside the existing ten, and re-armed only when privateParentState === 'trusted' (the daemon re-injects its own computed values after its own strip), mirroring the CHANNEL_PROMPT_META_KEY handling. Deleting only the new key would be a no-op (the legacy URI-only branch still exempts), so both are stripped.

Evidence: extended strips forged invocation metadata from an untrusted ACP client with both forged keys (its exact _meta: { keep: true } match goes red without the deletes) and passes trusted invocation context out of band and strips reserved metadata with the daemon's own marking (goes red without the trusted re-arm). Mutation probes run this round: deletes removed → untrusted test FAILED; re-arm removed → trusted test FAILED; restored → both green.

R2-3 (rc:4119082254, Suggestion) — retention-cap rationale cited a nonexistent echo byte coupling — FIXED

daemon-embedded-resource-replay.md claimed the 256 KiB budget "matched the live echo bound", but the echo's only bound is the 256-block MAX_ECHO_CONTENT_BLOCKS and it publishes non-text blocks verbatim. Restated as two independent bounds in both daemon-embedded-resource-replay.md and daemon-embedded-resource-replay.zh-CN.md (same edit, per the design-doc sync requirement).

R2-4 (rc:4119082262, Suggestion) — oversized-admission tests never asserted the forwarded content — FIXED

Test 3 (admits a direct oversized resource sharing a native attachment URI) now pins the full forwarded prompt array (native expansion at index 0, direct block at index 1 with the 256 KiB text) inside the existing toMatchObject, not just the _meta provenance. Test 1 now asserts promptCalls[0] carries the 256 KiB text and that the echoed user_message_chunk content is the context://example/oversized resource. These are the assertion shapes the finding's own mutation witness measured going red under a URI-dedup/drop mutation in resolveContent; this round adds no production guard on this path (test hardening only).

R2-5 (rc:4119082264, Suggestion) — attachment-expansion URI re-derived by hand in the new matchers — FIXED

Added the dependency-free attachmentResourceUri() helper (packages/acp-bridge/src/attachment-resource-uri.ts) and used it at the producer (sessionAttachments.ts resolve) and both matcher branches in embedded-resource-replay.ts. The matcher tests in embedded-resource-replay.test.ts now build expected URIs through the helper, so a producer-side format change turns the matcher suite red (the exact produced string stays pinned by sessionAttachments.test.ts). Mutation probe: reverting the position-branch matcher to a disagreeing literal turned the two position-path tests RED; restored → green. The fourth hand-derived site named in the finding (packages/core/src/utils/transcript-records.ts:176, the @attachment:/// display-token strip) is left as-is deliberately: core cannot import acp-bridge (dependency direction), and giving core its own export entry for the helper would touch packages/core/package.json's exports field, which is outside this PR's footprint and this round's boundaries.

R2-6 (rc:4119082270, Suggestion) — marker-only record projected zero replay updates — FIXED

projectUserAttachmentReferences in transcript-replay.ts now emits a [Embedded resource too large to replay] user chunk when embeddedResourcesTruncated === true and nothing else is projectable (no references, links, or retained resources, and an empty/absent display text), so the over-budget resource-only turn stays visible on replay instead of vanishing. The gate is deliberately narrow: a truncated prompt with visible text projects only its text (pinned by the new negative case), and the recorded displayText: '' contract is untouched (load-bearing for the channel-marker and title logic). New cases in transcript-replay.test.ts; the Session recording test now pins the marker via expect.objectContaining instead of exact-pinning the blank displayText pairing, per the acceptance criterion (recorded in test-weakening.json: no assertion removed, the shape's title semantics and replay projection are pinned by the new core/bridge tests). Mutation probe: placeholder removed → placeholder test FAILED; restored → green.

R2-7 (rc:4119082276, Suggestion) — deferred /advisor branch omitted attachment provenance — FIXED

The deferred-/advisor recordUserMessage branch now spreads attachmentReferences and resourceLinks (locally recomputed, matching branch 1) and its payload condition matches branch 1's, so a custom advisor command with daemon attachments keeps its attachment cards on replay. New test records daemon attachment references for a deferred custom advisor command dispatches /advisor … with populated qwen.daemon.attachmentReferences and asserts the payload carries them. Mutation probe: spreads removed → new test FAILED; restored → green. The existing exact-match sibling tests for this branch (input annotations, no-payload arm) still pass unchanged.

Not in this round: the four [probe] items under qwen-review-deferred are explicitly recorded as not requested this round and were left untouched. The review's disclosed gap ("Integration Tests (CLI, No Sandbox) was skipped in CI") is a CI-selection note; the changed behavior is covered by the unit suites below.

Verification

  • npm run build (COREPACK_HOME=/tmp/corepack-home — the default corepack cache path is not writable in this sandbox) — passed
  • npm run typecheck — passed (exit 0)
  • npm run lint — passed (exit 0, no warnings)
  • npx prettier --check on all 15 touched files — passed
  • packages/core: vitest run src/services/chatRecordingService.autoTitle.test.ts — 29 passed (incl. 2 new); vitest run src/services/chatRecordingService.test.ts — 148 passed
  • packages/acp-bridge: full suite — 54 files, 2576 passed (incl. new placeholder cases, helper-pinned matcher cases, strengthened bridge admission assertions)
  • packages/cli: vitest run src/acp-integration/acpAgent.test.ts — 811 passed; vitest run src/acp-integration/session/Session.test.ts — 1073 passed (incl. new advisor attachment-references test)
  • Mutation probes (all restored to green afterwards): R2-1 pre-fix run red on the 2 new tests; R2-2 deletes removed → untrusted test red, re-arm removed → trusted test red; R2-5 disagreeing matcher literal → 2 position-path tests red; R2-6 placeholder removed → placeholder test red; R2-7 spreads removed → advisor test red
  • Settings schema: no settings source changed; the build's generate:settings-schema step produced no diff
  • Integration tests not run: all changed behavior is exercised by the unit suites above (no bundled-CLI-only path touched)
中文说明

本轮摘要

在提交 434878e2 中处理了第 2 轮评审的全部 7 条发现(R2-1 至 R2-7)。下面每条修复都附有其复现/变异探针证据;每个新增的守卫或分支都有对应的测试,在移除该守卫时会变红。

发现处理

R2-1(rc:4119082230,Critical)——仅截断标记的记录导致自动命名失效 —— 已修复

chatRecordingService.trackUserDisplayTextForTitlePayload 中的标题投影豁免只以 embeddedResources.length 为判据,因此本 PR 新造出的仅截断载荷({ displayText: '', hookContext: '', embeddedResourcesTruncated: true })会被记为 '',从而打开 hasDisplayProjection,使整个会话(包括恢复后)的自动命名返回 empty_history。已将条件加宽,同时豁免 embeddedResourcesTruncated === true;裸 { displayText: '' } 频道标记行为仍由既有测试钉住。

证据:在 chatRecordingService.autoTitle.test.ts 新增 keeps truncation-only resource records eligible for auto-titling 与 restores truncation-only resource records without a title projection marker。两个测试先写并在未修复代码上确认变红(Tests 2 failed | 27 skipped,期望 [undefined] 实际 [""]),修复后转绿(29/29)。频道标记的兄弟测试(keeps an empty channel display projection as a title projection marker)仍通过,约束成立。

R2-2(rc:4119082249,Critical)——不可信 ACP 客户端可自行豁免于持久化日志 —— 已修复

acpAgent.ts#prompt 原本删除十个 daemon 专用 _meta 键,但不包括 qwen.daemon.attachmentReferences / qwen.daemon.attachmentResourceIndexes,因此 direct-ACP 调用方可同时伪造唯一校验的两侧,把任意 attachment:/// 块豁免于持久化记录,而内容仍送达模型。现在两个键都会先被捕获、与既有十个键一并删除,并仅在 privateParentState === 'trusted' 时重新装配(daemon 在自身剥除后会重新注入其自行计算的值),与 CHANNEL_PROMPT_META_KEY 的处理一致。只删除新键是无效操作(旧 URI 分支仍会豁免),所以两个键都剥除。

证据:在 strips forged invocation metadata from an untrusted ACP client 中加入两个伪造键(其 _meta: { keep: true } 精确匹配在缺少删除时会变红),并在 passes trusted invocation context out of band and strips reserved metadata 中加入 daemon 自己的标记(缺少可信重装时会变红)。本轮变异探针:移除删除 → 不可信测试失败;移除重装 → 可信测试失败;恢复 → 两者转绿。

R2-3(rc:4119082254,Suggestion)——保留上限的理由引用了不存在的回显字节耦合 —— 已修复

daemon-embedded-resource-replay.md 声称 256 KiB 预算"与实时回显上限一致",但回显唯一的边界是 256 块的 MAX_ECHO_CONTENT_BLOCKS,且非文本块会被原样发布。已在 daemon-embedded-resource-replay.md 与 daemon-embedded-resource-replay.zh-CN.md 中把两个边界改为各自独立的表述(同一次修改,符合设计文档双语言同步要求)。

R2-4(rc:4119082262,Suggestion)——超限接纳测试从未断言被转发的内容 —— 已修复

测试 3(admits a direct oversized resource sharing a native attachment URI)现在在既有 toMatchObject 中钉住完整的转发 prompt 数组(索引 0 为原生展开块,索引 1 为携带 256 KiB 文本的直接块),而不再只钉 _meta 来源信息。测试 1 现在断言 promptCalls[0] 携带 256 KiB 文本,且回显的 user_message_chunk 内容确为 context://example/oversized 资源。这些断言形状与该发现自带变异实证所测得的变红形状一致(在 resolveContent 的 URI 去重/丢弃变异下变红);本轮在该路径上未新增任何生产代码守卫(仅加固测试)。

R2-5(rc:4119082264,Suggestion)——附件展开 URI 在新匹配器中被手工重新推导 —— 已修复

新增零依赖辅助函数 attachmentResourceUri()(packages/acp-bridge/src/attachment-resource-uri.ts),并用于生产端(sessionAttachments.ts 的 resolve)与 embedded-resource-replay.ts 的两个匹配分支。embedded-resource-replay.test.ts 的匹配器测试现在通过该辅助函数构造期望 URI,因此生产端格式一旦变化,匹配器测试套件就会变红(精确的产出字符串仍由 sessionAttachments.test.ts 钉住)。变异探针:把位置分支匹配器改回不一致的字面量后,两个位置路径测试变红;恢复后转绿。发现中点名的第四处(packages/core/src/utils/transcript-records.ts:176 的 @attachment:/// 显示标记剥除)有意保留原样:core 不能反向依赖 acp-bridge,而为 core 新增导出条目需要改动 packages/core/package.json 的 exports 字段,这超出本 PR 的足迹与本轮边界。

R2-6(rc:4119082270,Suggestion)——仅标记记录在回放时投影出零条更新 —— 已修复

当 embeddedResourcesTruncated === true 且没有其他可投影内容(无引用、无链接、无保留资源,且 displayText 为空或缺失)时,transcript-replay.ts 的 projectUserAttachmentReferences 现在会发出一条 [Embedded resource too large to replay] 用户块,使超预算的纯资源轮次在回放中保持可见,而不是消失。该门槛有意收窄:带有可见文本的截断 Prompt 只投影其文本(由新增反向用例钉住),记录中的 displayText: '' 契约保持不变(它对频道标记与标题逻辑是承重的)。新增用例位于 transcript-replay.test.ts;按验收标准,Session 记录测试改为用 expect.objectContaining 钉住标记,不再精确钉住空白 displayText 的组合(已记录在 test-weakening.json:没有删除任何断言,该形态的标题语义与回放投影由新增的 core/bridge 测试钉住)。变异探针:移除占位符 → 占位符测试失败;恢复后转绿。

R2-7(rc:4119082276,Suggestion)——延迟 /advisor 分支遗漏附件来源信息 —— 已修复

延迟 /advisor 的 recordUserMessage 分支现在展开 attachmentReferences 与 resourceLinks(本地重新计算,与分支 1 一致),其载荷条件也与分支 1 对齐,因此带 daemon 附件的自定义 advisor 命令在回放时保留附件卡片。新增测试 records daemon attachment references for a deferred custom advisor command 以填充的 qwen.daemon.attachmentReferences 派发 /advisor … 并断言载荷携带这些引用。变异探针:移除展开 → 新测试失败;恢复后转绿。该分支的既有精确匹配兄弟测试(输入标注、无载荷分支)仍原样通过。

本轮未处理:qwen-review-deferred 下的四条 [probe] 条目被明确记录为本轮不要求处理,未做改动。评审披露的缺口("Integration Tests (CLI, No Sandbox) 在 CI 中被跳过")属于 CI 选择说明;本轮改动的行为已由下列单元套件覆盖。

验证

  • npm run build(COREPACK_HOME=/tmp/corepack-home —— 本沙箱中默认 corepack 缓存路径不可写)—— 通过
  • npm run typecheck —— 通过(退出码 0)
  • npm run lint —— 通过(退出码 0,无警告)
  • 对全部 15 个改动文件执行 npx prettier --check —— 通过
  • packages/core:vitest run src/services/chatRecordingService.autoTitle.test.ts —— 29 通过(含 2 个新增);vitest run src/services/chatRecordingService.test.ts —— 148 通过
  • packages/acp-bridge:全量套件 —— 54 个文件,2576 通过(含新增占位符用例、helper 钉住的匹配器用例、加固后的 bridge 接纳断言)
  • packages/cli:vitest run src/acp-integration/acpAgent.test.ts —— 811 通过;vitest run src/acp-integration/session/Session.test.ts —— 1073 通过(含新增 advisor 附件引用测试)
  • 变异探针(事后均恢复为绿):R2-1 修复前运行 2 个新测试为红;R2-2 移除删除 → 不可信测试红,移除重装 → 可信测试红;R2-5 不一致的匹配器字面量 → 2 个位置路径测试红;R2-6 移除占位符 → 占位符测试红;R2-7 移除展开 → advisor 测试红
  • Settings schema:未改动任何 settings 源;构建中的 generate:settings-schema 步骤无差异
  • 集成测试未运行:上述单元套件已覆盖全部改动行为(未触及仅经打包 CLI 才能走到路径)

Base-conflict check · 基分支冲突检查: no conflict with main. · 与 main 无冲突。

🧵 Resolved all 7 selected review thread(s). · 已关闭全部选中的 7 条评审线程。

Re-review when you have a moment. After round 100 this bot stops and leaves the PR for a human. · 有空请复审;第 100 轮后本 bot 停止并将 PR 交给人工。


🧠 Handled by Qwen Code · model/模型 kimi-k3 · CLI 0.24.6

@qwen-code-dev-bot

Copy link
Copy Markdown
Collaborator

🔀 Base updated: red check(s) [Lint & Static (ubuntu-latest, Node 22.x)] pass on current main — merged current main via update-branch; CI will re-run.

中文说明

🔀 已更新 base:红色检查 [Lint & Static (ubuntu-latest, Node 22.x)] 在当前 main 上通过 —— 已通过 update-branch 合入当前 main,CI 将重新运行。

@qwen-code-dev-bot

Copy link
Copy Markdown
Collaborator

🤖 AutoFix ran out of time before finishing (timeout (7200000ms)) (attempt 5/100) — it will retry on the next scan.

What I found before stopping:
Qwen failed during address-review: timeout (7200000ms).

See the Qwen Autofix agent step logs for model/tool output.

中文说明

🤖 AutoFix 在完成前耗尽了时间(timeout (7200000ms))(第 5/100 次尝试)—— 将在下次扫描时重试。

Run log: https://github.com/QwenLM/qwen-code/actions/runs/36467708558


🧠 Handled by Qwen Code · model/模型 kimi-k3 · CLI 0.24.6

@wenshao
wenshao dismissed stale reviews from qwen-code-review-bot, qwen-code-review-bot, qwen-code-review-bot, and qwen-code-review-bot October 31, 2026 04:03

Stale automated-review block: every finding is fixed on head ed5fb43 or tracked in #13282; maintainer verification round 6 passed 60/60 assertions and the PR-body mismatch (its last blocker) is corrected. Takeover cleanup 2026-10-31.

@wenshao

wenshao commented Oct 31, 2026

Copy link
Copy Markdown
Collaborator

Takeover complete — merge-ready pending approvals

Maintainer takeover (2026-10-31), head ed5fb433903 unchanged. Actions:

  1. PR body corrected — the four English and four Chinese phrases claiming pre-echo rejection / invalid-parameters now describe the shipped semantics: admission + live delivery to the model + retention-only durable drop marked embeddedResourcesTruncated (the prompt always runs). This was the sole blocker in verification round 6 (60/60 assertions, ic:6491633555). The design docs already carried the correct contract; no code change was needed.
  2. All 12 unresolved threads closed — 3 fixed or addressed by a documented alternative (legacy same-URI exception recorded in both design docs; whole-object byte bound + skeleton cascade replacing the unconditional whitelist; clone-isolation test landed), 9 Suggestion-level items tracked in Deferred review findings from PR #12585: fix(acp): persist embedded text resources for transcript replay #13282 (5 newly registered in the census comment, 4 already queued). Each thread carries its evidence or queue link.
  3. 19 stale CHANGES_REQUESTED reviews dismissed (18 automated-review + 1 whose regression the author confirmed fixed on 2026-09-27); a full REST census confirms 0 remaining. The review-decision cache refreshes on the next review event.
  4. Autofix loop released (autofix/takeover label removed).

State: CI green (30 pass / 35 skipping / 0 fail), trial merge into current main (674d751bcd4) conflict-free (git merge-tree exit 0).

Remaining for merge: the main ruleset requires 2 approving reviews; there are currently none valid. No further code change is planned — the remaining Suggestions live in #13282.

中文摘要

维护者接管(2026-10-31),head ed5fb433903 未变。已修正 PR 正文 8 处「回显前拒绝/参数错误」表述为实际语义(放行 + 实时送达模型 + 仅保留层丢弃并置 embeddedResourcesTruncated,Prompt 始终执行);12 个未决线程全部关闭(3 个已修复或以文档化替代方案解决,9 个 Suggestion 级登记到 #13282);19 条过期 CHANGES_REQUESTED 评审已 dismiss(REST 全量复核余 0);autofix 循环已释放。CI 全绿,与当前 main 试合并无冲突。合并尚需 ruleset 要求的 2 个批准评审(当前无有效批准)。

wenshao
wenshao previously approved these changes Oct 31, 2026

@wenshao wenshao left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Approving at head ed5fb433903 after the maintainer takeover (2026-10-31).

Basis:

  • Verification round 6 (ic:6491633555): 60/60 scripted assertions on real processes — a real qwen --acp child, a real JSONL journal, a real SIGKILL + session/load restart. Base replays 0 typed resource chunks (resource-only turns vanish); head replays 2, byte-identical, attributed to their owning prompts, zero URI re-fetch. Round 5's two Criticals are closed by measurement (M10/M11 mutation witnesses killed).
  • Its sole blocker is corrected: the PR body no longer describes a pre-echo rejection path the code does not have. It now states the shipped semantics — admission + live delivery to the model + retention-only durable drop marked embeddedResourcesTruncated — matching both design docs and the PR's own test (bridge-embedded-resource-replay.test.ts:12, "admits an oversized direct text resource; the bound limits retention only").
  • Takeover census: all 12 unresolved threads closed with evidence (3 fixed or addressed by a documented alternative, 9 Suggestions tracked in #13282); 19 stale CHANGES_REQUESTED reviews dismissed (full REST census: 0 remaining); CI green (30 pass / 35 skipping / 0 fail); trial merge into current main conflict-free.

No code change since round 6's verified head. The remaining Suggestions live in #13282, not in this PR. This is approval 1 of the 2 the main ruleset requires.

中文摘要

在 head ed5fb433903 批准。依据:round-6 真实进程验证 60/60(base 回放 0 个类型化资源块、head 回放 2 个且字节一致、归属原 prompt、零 URI 抓取;round-5 两条 Critical 经变异见证关闭);其唯一阻断项(正文描述代码不存在的回显前拒绝路径)已修正为实际语义,与设计文档和自带测试一致;接管核对关闭全部 12 个未决线程(9 个 Suggestion 登记 #13282)、dismiss 19 条过期评审(REST 全量复核余 0)、CI 全绿、与当前 main 试合并无冲突。自 round-6 验证 head 以来零代码改动。本批准为 ruleset 要求的 2 个批准之一。

@wenshao

wenshao commented Oct 31, 2026 •

Copy link
Copy Markdown
Collaborator

Verdict: merge-ready — 214/214 scripted assertions passed. Round 7 (follow-up): the head has not moved since round 6 (ed5fb433903f372db0c486f236a758af0358c51b), the PR body's rejection wording is now corrected and the shipped behaviour matches it by measurement, and the change was re-verified on a fresh trial-merge build against today's main (f87fd6a037) — every cell that runs against the merge candidate is green. No new blocking finding; the two residual Suggestions are the same ones round 6 recorded and they are tracked in #13282.

Verified head ed5fb433903f372db0c486f236a758af0358c51b · base 9a41afe0d46317b9432281c62e305220c74b2dd8 (= merge-base = base tip) · trial merge conflict-free at today's main@f87fd6a03719c14b2ec4ff93b70e70bf90f527cf (git merge-tree exit 0, tree c55a11c9dcb43e8c07db0d74137bd5c7eac28ba0). The measured merged build was constructed at main@107e1cfab1 (merged commit 45ea3a08); main's 40-file advance to the final tip touches only packages/managed-agent-server/**, sdk-java.yml, design docs and scripts/tests/hosted-process-ci.test.js — nothing this PR touches or the gates below import.

中文摘要(点击展开)

结论:merge-ready —— 214/214 条脚本断言通过。head 自 round 6 起未变(ed5fb4339),PR 正文的「回显前拒绝」表述已修正且与实测行为一致,并在针对当前 main(f87fd6a037)的试合并树上重新构建、重新测量,全部通过(实测树构建于 main@107e1cfab1,其后 40 个文件的差异只涉及 managed-agent-server Java、CI 与设计文档,未触及本 PR 或门禁涉及的文件)。没有新的阻断项;仅剩的两条 Suggestion 与 round 6 相同,已登记于 #13282。

  • 中心主张(真实进程 A/B):真实 dist/cli.js --acp + 真实 JSONL + 真实 SIGKILL + session/load。base 回放 0 个带类型 resource chunk(纯资源轮次整轮消失),head 回放 2 个、字节一致、归属原 prompt,零 URI 抓取,两臂用户文字块均为 2(活性对照)。三臂(base/head/merged)各 13/13,见 Central claim and A/B 表与 01-central-ab-base-vs-head.png。
  • round-6 唯一阻断项已关闭:PR 正文 4 处英文 + 4 处中文表述改为「放行 + 实时送达模型 + 仅保留层丢弃并置 embeddedResourcesTruncated」;big 与 big-only 两格实测与正文一致(见 Edge shapes 表)。
  • 新覆盖(round 6 未覆盖的 Reviewer Test Plan 项):① 原生附件位置/旧元数据歧义(u1,23 格);② 不可信 ACP 调用方伪造 qwen.daemon.attachmentReferences(h3,base 伪造引用卡 1 张 → head 0 张、改为持久化的真实资源);③ SDK 离线投影(s3,用真实回放帧驱动 shipped dist/daemon/ui/index.js,含 session.rewound);④ 试合并树三臂(merged 全部 13/13 + 门禁 190/172/34)。
  • 变异矩阵:M1(provenance 门禁)→ h3 6/10 转红;M2/M3(原生位置/歧义守卫)→ u1 各 2 格转红且 PR 自带测试各 1 红;M4(256 KiB→1 KiB 正对照)→ 套件 4 红。所有变异均在 dist 中留有标记,恢复后 head bundle 哈希回到 c5c2a0d372da850ba6e5。
  • 仍未覆盖:R21-1 残留 RSS/时间测量、M12 变异、真实 qwen serve socket 与浏览器刷新、Windows/macOS 对照、仓库级 lint/typecheck/preflight、Session.test.ts / acpAgent.test.ts 复跑。详见 Not covered。

Delta classification since round 6

The PR head is the same commit round 6 verified (ed5fb433903f372db0c486f236a758af0358c51b), and the base/merge-base are unchanged, so every source file this feature touches is byte-identical to that round. What changed around it:

  • main advanced 61b7fdb1 → 107e1cfab1 → f87fd6a037 (round 6's trial-merge target → the tree measured this round → today's final tip). Three of those commits touch files this PR also changes: ca6ca0d374 (branch points through the session-load replay — session-transcript-reader.ts, history-replay-page.ts, acpAgent.ts), cc2a80e619 (fresh-prompt stop point), 4dd7ac7d41 (subagent identity in replay frames). 16 files overlap in total.
  • The PR body changed (the round-6 blocker): four English and four Chinese phrases moved from "rejected before echo / invalid parameters" to "admitted and delivered live; only durable retention is dropped".

Because the merge candidate is what lands, this round treats the trial-merge tree as a third arm, not only head vs base.

Previous-finding status (follow-up protocol; re-measured at ed5fb4339)

# finding (round) severity status at ed5fb4339 evidence
C-1 = L-2 PR body describes a rejection path the code does not have suggestion FIXED body now says admitted + live + retention-only drop; big/big-only cells match the new wording
C-2 partial truncation drops user content with no client-visible trace suggestion STANDS, narrowed big (text + oversized resource): placeholderChunks=0; new big-only cell (no text): replay emits exactly 1 [Embedded resource too large to replay]
R2-#4 intra-turn block order not preserved on replay note STANDS, re-measured submitted [resource, text] replays text,resource on head and merged
R21-1 cold restore retains per-turn displayText and cloned apiHistory note not re-measured no code change since round 6; no RSS measurement this round (see Not covered)
M12 incremental accumulator has no committed test note not re-run unchanged; still a coverage observation, not a defect
L-3 base lets any ACP caller inject forged attachmentReferences security, closed by PR closed, measured h3: base replays a forged reference card (1) and journals nothing; head strips the keys, journals the real resource, replay emits 0 forged cards
C-4 / R2-#3 / M0–M3 count cap, URI encoding, replay liveness, byte budget suggestion re-run u1 G/F/H + h1 liveness cells all green

Central claim and A/B on real processes

Claim. An ACP session/prompt carrying embedded text resource blocks is persisted with its owning user prompt and replayed as typed user chunks after a real restart.

Harness harness/h1-restart-e2e.mjs spawns node <tree>/dist/cli.js --acp in an isolated HOME/QWEN_HOME, talks to a loopback OpenAI-compatible server (the only fake — the far side of the wire), sends text + 1 resource, a resource-only prompt, then text-only, reads the real JSONL, SIGKILLs, and loads the session in a second process. Base expectations are encoded, so a base cell that loses the resources is a passing control.

observable (real process, real JSONL, real restart) base 9a41afe0 head ed5fb4339 trial merge 45ea3a08 (main@107e1cfab1)
durable user records 3 3 3
user records carrying embeddedResources 0 (0,0,0) 2 (1,1,0) 2 (1,1,0)
journaled resource blocks 0 2 2
replayed user_message_chunk with content.type === 'resource' 0 2 (byte-identical text) 2 (byte-identical text)
resource chunk attribution (source-record → turn index) — 0, 1 0, 1
resource-only turn visible after restart no — the turn vanishes yes yes
replayed user text chunks (liveness) 2 2 2
model calls carrying the resource text live 6 6 6
URI re-fetch over HTTP 0 0 0
new model call during session/load 0 0 0
scripted cells 13/13 (all expected-red) 13/13 13/13

Witness 01-central-ab-base-vs-head.png (live output of both arms, re-run this round on freshly rebuilt bundles: base fddc789c0339ae9ee721, head c5c2a0d372da850ba6e5).

Edge shapes, re-measured on real processes

Harness harness/h2-edge-e2e.mjs, one fresh session per shape so the 400 KiB payload cannot disturb the others.

cell base head trial merge
[resource, text] submission resource absent; replay text replay text,resource (order still not preserved) text,resource
same-URI duplicate resources journal 0, replay 0 (turn invisible) journal 2, replay 2, both context://dup journal 2, replay 2
400 KiB direct resource + text: prompt admitted (end_turn) yes yes yes
model received it live (max wire body) 481,696 B 481,696 B 481,858 B
durable record absent payload dropped + embeddedResourcesTruncated: true + displayText kept same
replay of that turn text text (no placeholder — round-6 C-2) text
400 KiB resource-only prompt: admitted, live bytes 481,728 B 481,728 B 481,890 B
its replay turns: 0 chunks (turn vanishes) exactly 1 [Embedded resource too large to replay] placeholder same as head

Native-attachment provenance and retention bounds (u1)

harness/u1-native-positions.mjs drives the shipped packages/acp-bridge/dist/embedded-resource-replay.js (no mocks) with the shapes the acceptance plan names; 23 cells, green on head and on the trial merge.

cell head / merged
position metadata [1] + one direct and one native block sharing attachment:///att-1 native index excluded, direct block retained (DIRECT-RESOURCE)
legacy metadata (no positions), unique URI match excluded
legacy metadata, two blocks share the URI (ambiguous) both retained, no silent drop
out-of-range / duplicate / non-resource / non-array indexes, empty reference list all fail closed ([])
daemon-owned _meta keys stripped; client _meta kept
byte bound bisect JSON bytes 262144 retained; 262145 dropped with truncated=true
block cap 256 retained of 257, truncated=true
attachment URI format attachment:///a%2Fb%20c stable

Untrusted provenance claim (h3) — the L-3 gate

A plain ACP caller sends a prompt whose resource carries _meta claiming daemon native provenance (qwen.daemon.attachmentReferences + qwen.daemon.attachmentResourceIndexes: [1]). The daemon re-injects those keys only for its trusted children.

observable base 9a41afe0 head ed5fb4339 trial merge
durable resource retained 0 1 (attachment:///att-1, text exact) 1
replayed embedded resource chunks 0 1 1
replayed forged reference cards (content.attachmentId) 1 0 0
cells 10/10 10/10 10/10

SDK offline projection from the real replay wire (s3)

The captured session/update frames of each arm's real session/load are wrapped in daemon event frames and fed to the shipped dist/daemon/ui/index.js (normalizeDaemonEvent → reduceDaemonTranscriptEvents). Nothing is a fixture.

observable base head trial merge
normalized UI events 9 11 11
user blocks 2 (resource-only turn absent) 3 3
turns carrying embeddedResources 0 2 2
turn 0 / turn 1 resource URI — context://one / context://two same
resource text sha (turn 0 / turn 1) — da5c08597e… / 8b5cf5d0… same
invented attachmentId none none none
session.rewound to turn index 1 1 user block, 0 resources 1 user block, turn-0 resource kept same
cells 13/13 13/13 13/13

Mutation matrix at head

Baseline first (M0): the PR's embedded-resource-replay suite is 15/15 green and u1 is 23/23. Every mutant is a single-literal edit with a marker constant, and each marker was found in the shipped artifact before the probe ran (dist/chunks/acpAgent-*.js for M1, dist/embedded-resource-replay.js for M2–M4). Every restore is sha256-identical and the head bundle returns to c5c2a0d372da850ba6e5.

# mutant probe PR suite
M1 provenance gate trusted → always on (acpAgent.ts) h3 6/10 red: no durable resource, 1 forged card not run
M2 native-position exemption disabled u1 2/23 red (A.retainedCount 1→2, B.retained 0→1) 1 failed / 14 passed
M3 legacy same-URI ambiguity guard removed u1 2/23 red (C.legacyIndexes []→[0,1], C.retained 2→0) 1 failed / 14 passed
M4 256 KiB byte bound → 1 KiB (positive control) u1 1/23 red (G.cap.count 256→11: the byte arm binds first) 4 failed / 11 passed

Kills attributed: M2 → snapshotReplayableEmbeddedResources > exempts daemon-native positions from retention; M3 → readDaemonNativeResourceIndexes > keeps ambiguous same-URI blocks on the legacy path. M4 is the runner liveness control: the same command that reports green on the clean tree reports four reds with the bound mutated, so the green rows are measurements rather than a dead runner. No mutant row is presented as per-commit coverage: the PR head is unchanged since round 6, so git log --no-merges <round-6 head>..<head> --not <main tip> is empty and the matrix is per-guard.

Findings

No Critical or blocking finding. Two Suggestions stand from round 6, both tracked in #13282:

1. Suggestion — intra-turn block order is still not preserved. Submitting [resource, text] replays text,resource, because projectUserAttachmentReferences emits references/resources after projectMessageParts. Reproduce: node harness/h2-edge-e2e.mjs --tree <head> --arm head --expect head --shape order → replay.sequence="text,resource".

2. Suggestion — a mixed turn that lost its oversized resource still replays silently. When the turn has visible text and a dropped resource, replay emits the text alone (placeholderChunks=0). The truncation marker exists only for the turn with nothing else to show (new big-only cell: placeholderChunks=1). Both cells pass because the shipped behaviour is what the design documents; the author's plan records the loss in the durable record via embeddedResourcesTruncated, which no client is obliged to render.

Not covered

  • R21-1 residual retention: per-turn displayText and cloned apiHistory were not RSS-measured. No code changed since round 6; the fix remains structural.
  • M12 (accumulator vs ChatRecord[]) not mutated; unchanged since round 6.
  • C-3 shape matrix A..G, inline blob resources (not persisted by design), and legacy records that never stored resources.
  • Real qwen serve socket and browser refresh: the SDK probe consumes the real replay frames, but no web client rendered them and no running daemon served them.
  • Session.test.ts / acpAgent.test.ts were not re-run (round 5 exceeded a 10-minute cap on them). The suites that did run are listed below.
  • Windows/macOS parity: this round ran on Linux only (the PR author tested macOS).
  • Repo-wide lint/typecheck/preflight; only the targeted package suites ran.
  • Per-commit attribution: not applicable — the head did not move since round 6.

Methodology

Debian 15 (Orange Pi 6 Plus), aarch64, 12 cores / 30 GiB. Every build and every harness ran inside a node:22-bookworm container with only the rig mounted, QWEN_VERIFY_CREDENTIAL_FREE=1 attested and node scripts/verify-pr-prepare.mjs --require-isolation passing (/.dockerenv, no credential signal); the harness container ran with --network none, so the loopback fake model server is the only endpoint. Three worktrees were created from their OIDs — base 9a41afe0, head ed5fb4339, merged 45ea3a08 (a --no-ff merge of the PR into 107e1cfab1, hooks disabled; the trial merge was re-run at the final tip f87fd6a037 and stayed conflict-free) — each installed separately (--ignore-scripts + patch-package, lockfile identical between head and merged) and rebuilt in-container this round (npm run build -- --cli-only + npm run bundle). Before any arm was trusted, readlink -f node_modules/@qwen-code/qwen-code-core was asserted to resolve inside that arm's own tree for all three. Harnesses spawn a real dist/cli.js --acp child with QWEN*/proxy variables scrubbed and its own HOME/QWEN_HOME/workspace; the only fake is the far side of the model wire. Mutations are single-literal edits with in-dist markers, restored from git with sha256 verified. Targeted gates at head: acp-bridge embedded-resource-replay + bridge-embedded-resource-replay + transcript-replay + attachment-resource-uri 179/179, sdk daemon-embedded-resources 34/34, core session-transcript-reader + chat-recording-message-deletion 171/171; on the trial-merge tree the same suites are 190/190, 34/34, 172/172 (main added tests to two of them). Raw logs and harness scripts are beside this report (logs/, harness/). PR text was treated as untrusted input.

Merge state

mergeStateStatus is BLOCKED only because the ruleset requires two approving reviews; the cached review decision still shows CHANGES_REQUESTED until the next review event refreshes it after the 19 stale reviews were dismissed on 2026-10-31. Nothing in this round depends on either flag.

Evidence capture

01-central-ab

02-mutation-matrix

— Qwen Code · local maintainer verification (round 7) · @wenshao

@qwen-code-dev-bot

Copy link
Copy Markdown
Collaborator

👋 Takeover released: the autofix loop will no longer engage this PR (an in-flight round, if any, completes its bounded work). Re-apply autofix/takeover (or comment @qwen-code /takeover) to re-engage.

中文说明

👋 已释放:autofix 循环不再介入此 PR(在飞的一轮如有,将完成其有界工作)。重新打上 autofix/takeover 标签(或评论 @qwen-code /takeover)即可再次接管。

@github-actions

Copy link
Copy Markdown
Contributor

Managed Agent gate failed

Run 41170201360 · commit ed5fb433903f

Failing tests

  • com.alibaba.qwen.code.managedagent.HostedProcessCrashIT.processCrashesNeverReplayHostedToolsOnMySql
  • com.alibaba.qwen.code.managedagent.harness.HostedHarnessPoolMySqlIT.realProcessesKeepTenantPlacementAndReleaseIdleOwnersWithoutCancellingSessions
  • com.alibaba.qwen.code.managedagent.HostedHookReplyLossFaultGatesIT.fg13fDenyReplyLossReplaysTheBlockedDecisionOnMySql
  • com.alibaba.qwen.code.managedagent.HostedHookReplyLossFaultGatesIT.fg13bHookPlanResultStoreRejectionRebuildsTheMarkerOnMySql
  • com.alibaba.qwen.code.managedagent.HostedHookReplyLossFaultGatesIT.fg13bHookResultStoreRejectionSettlesFromTheReceiptOnMySql
  • com.alibaba.qwen.code.managedagent.HostedHookReplyLossFaultGatesIT.fg13aStatusReplyLossStaysUnknownUntilTheNextPollOnMySql
  • com.alibaba.qwen.code.managedagent.HostedHookReplyLossFaultGatesIT.fg13aHookResultStoreAckLossSettlesWithoutASecondCallOnMySql
  • com.alibaba.qwen.code.managedagent.HostedHookReplyLossFaultGatesIT.fg13fFailClosedReplyLossReplaysTheBlockedFailureOnMySql
  • com.alibaba.qwen.code.managedagent.HostedHookFaultGatesIT.fg13cHarnessKillDuringCommandSettlesFromTheOriginalReceiptOnMySql
  • com.alibaba.qwen.code.managedagent.HostedHookCancelFaultGatesIT.fg13dHeldCancelEscalatesToCgroupKillAndSettlesCancelledOnMySql
  • com.alibaba.qwen.code.managedagent.HostedHookCancelFaultGatesIT.fg13dCancelWithDroppedReplySettlesThroughStatusOnMySql
  • com.alibaba.qwen.code.managedagent.HostedHookCancelFaultGatesIT.fg13dCancelWithStoppedWorkerStaysUnknownAndKeepsTheHoldOnMySql
  • com.alibaba.qwen.code.managedagent.HostedEventStreamLivenessIT.aKilledHarnessBehindAHalfOpenStreamEndsTheReadWithinTheBound
  • com.alibaba.qwen.code.managedagent.HostedEventStreamLivenessIT.aStalledStreamReattachesWithoutReplayingTheTurn
  • com.alibaba.qwen.code.managedagent.HostedCommittedEventLineReplayIT.readerAcceptsTheCommittedEventShapes
  • com.alibaba.qwen.code.managedagent.HostedPublicWorkspaceIT.publicShellReceiptFaultsNeverRepeatEffects(String)[1]
  • com.alibaba.qwen.code.managedagent.HostedPublicWorkspaceIT.publicShellReceiptFaultsNeverRepeatEffects(String)[2]
  • com.alibaba.qwen.code.managedagent.HostedPublicWorkspaceIT.publicShellV2DeletionTriggersCollectionWithoutRetirementSeams
  • com.alibaba.qwen.code.managedagent.HostedPublicWorkspaceIT.ownerAnswersHostedApprovalsThroughBothSurfaces
  • com.alibaba.qwen.code.managedagent.HostedPublicWorkspaceIT.storedDefinitionApprovalSelectionRoundTripsThroughRealWorkspaceToolsOnMySql
  • …and 5 more

Resolve the session-title conflicts so both sides survive: this PR's authoritative projectionMode flag, per-turn projection state and embedded-resource persistence, plus main's Mod-prompt exclusion and session-title sidecar.

- tryGenerateSessionTitle now takes (userDisplayTexts, modPromptKeys, projectionMode); the recorder and /rename pass all three.
- trackTitleText folds the per-turn projection bookkeeping into main's record-level tracker: a Mod's prompt contributes null, is never retained as rescue text, and cannot establish projection mode on its own. The cold-restore accumulator mirrors the rule; both are pinned by new tests.
- rebuildTurnBoundaries re-derives the Mod keys with the boundaries, so resume and message deletion no longer track title text twice.
- Retarget the call-shape assertions on both sides to the merged signature, and main's two Mods-context payload assertions to the displayProjection marker. No assertion was dropped.

Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com>
@wenshao

wenshao commented Nov 7, 2026

Copy link
Copy Markdown
Collaborator

Takeover round — base merged, conflict resolved

Head ed5fb433903 → ce481fe9f9 (a merge of main 752e399b3b6).

Conflict

The branch was CONFLICTING against main. git merge-tree named exactly 8 conflicting files, all of them files this PR already changes: renameCommand.ts / renameCommand.test.ts, core/client.ts, chatRecordingService.ts / chatRecordingService.autoTitle.test.ts, sessionTitle.ts / sessionTitle.test.ts, and sdk-typescript/daemon/ui/transcript.ts.

All 8 were resolved so that both sides survive — this PR's embedded-resource persistence/replay and authoritative projection flag, and main's Mod-prompt exclusion and title sidecar. No conflict was resolved by taking one side wholesale:

  • tryGenerateSessionTitle now takes (config, signal, userDisplayTexts, modPromptKeys, projectionMode). Main's Mod-key filter stays in the non-projection branch (filterToDialog(fullHistory, modPromptKeys)), and the PR's projectionMode branch still never reads raw history. Main's hasDisplayProjection heuristic is superseded by the PR's authoritative flag — which is what that flag was introduced for.
  • client.ts keeps main's displayText fallback chain and hookContext: additionalContext, plus the PR's displayProjection: true.
  • One merge rule was needed to keep both behaviours true: a Mod-submitted prompt contributes null to the title ring and cannot establish projection mode on its own. Without it, main's "a Mod-only session still titles from the conversation" is swallowed by the PR's projection path. The rule is mirrored in the cold-restore accumulator (session-transcript-reader.ts) so a restart derives the same state, and rebuildTurnBoundaries now re-derives title texts and Mod keys together, so the restore and message-deletion paths no longer track twice.
  • Two regression tests pin that rule, each mutation-witnessed (dropping the !isModPrompt guard turns it red; restoring it turns it green). Existing assertions were retargeted to the merged signature, not deleted — the autoTitle suite is +478/−0 against main.

The two red checks

SDK Java (full) / Hosted process fault gates / MySQL 8.4 / Java 21 and its summary job Managed Agent gate (run 41170201360) were not this PR's defect:

  • The diff touches 46 files, all TypeScript or docs — zero Java files.
  • The log's 20 errors are all Could not find field 'attachments' on target object [...SdkHarnessBoundary], plus NoSuchBeanDefinition ... HarnessPoolConnector. SdkHarnessBoundary.java does not exist anywhere in the pre-merge PR head tree.
  • That run was triggered by pull_request_review on 2026-10-31 and built the merge ref against main's intermediate Java state. Current main has restructured packages/managed-agent-server into a multi-module Maven layout.
  • After this merge, packages/managed-agent-server and packages/sdk-java are byte-identical to current main (git diff origin/main HEAD -- … is empty).

No Java code and no CI configuration were changed. The Java lane was not run locally and no green Java lane is claimed — the new head's CI re-evaluates it.

Verification (local, real results)

node scripts/setup-worktree.js → 0; npm run build → 0; npm run typecheck → 0; eslint on the 11 changed files → 0; prettier --check on the staged set → clean; the pre-commit lint-staged hook passed.

Vitest, one file at a time, all green — core: sessionTitle 42, autoTitle 38, chatRecordingService 198, title-sidecar 19, session-title-sidecar 4, sessionService.title-sidecar 13, customTitle 24, message-deletion 16, session-transcript-reader 157, writer-lease 138 (+10 skip), corruption 76, sessionService 268, bun-pty 32 (+1 skip), client 624; acp-bridge: 4 / 3 / 15 / 98 / 173; sdk-typescript: 34 / 44 / 23 / 11; cli: renameCommand 25, box-metrics 13, scheduled-tasks 138, hosted-hook 129, hosted-harness 708, acpAgent 939, Session 1446.

Three non-assertion reds were diagnosed and excluded, not suppressed: client.test.ts default-heap OOM (re-run at 12 GiB / 1 worker); acpAgent.test.ts inheriting a leaked QWEN_CODE_SERVE=1 / QWEN_SERVE_MCP_BUDGET_MODE=off from the environment (unset → 939 green); and a coverage/.tmp write race (re-run alone → 1446 green). sessionTitle.test.ts was independently re-run after the commit: 42/42.

Merge fidelity was checked in both directions: the commit's diff against the merged main is exactly this PR's 46 files, no conflict markers remain, and the worktree is clean.

Base drift during the round

main advanced 5 commits while this round ran (to 5faa620af9b). git merge-tree HEAD origin/main is clean against it, and those 5 commits touch none of this PR's 46 files.

Approvals

main requires 2 approving reviews and dismisses stale reviews on push, so this push dismisses the 2026-10-31 approval. Two fresh approvals are needed. This watch does not approve, change review state, merge, or enable auto-merge.

Carried items

The 102 review / issue_comment events read this round are historical round reports, base-update notices and maintainer verification reports, already dispositioned by the 2026-10-31 takeover; all 80 review threads are resolved and no outstanding request was found in them. The 3 Criticals deferred in round 13 (fails-closed / new-surface, explicitly non-blocking) and the 2 Suggestions carried from round 6 remain tracked in #13282 and were not reopened.

中文说明

接管轮次 —— 已合入 base,冲突已解

Head ed5fb433903 → ce481fe9f9(合并 main 752e399b3b6)。

冲突

分支相对 main 为 CONFLICTING。git merge-tree 给出恰好 8 个冲突文件,全部是本 PR 已改动的文件:renameCommand.ts / renameCommand.test.ts、core/client.ts、chatRecordingService.ts / chatRecordingService.autoTitle.test.ts、sessionTitle.ts / sessionTitle.test.ts、sdk-typescript/daemon/ui/transcript.ts。

8 处全部按「两侧语义都存活」解决,未整块取任一侧:

  • tryGenerateSessionTitle 签名合并为 (config, signal, userDisplayTexts, modPromptKeys, projectionMode)。main 的 Mod 键过滤保留在非投影分支(filterToDialog(fullHistory, modPromptKeys)),PR 的 projectionMode 分支仍绝不读原始 history。main 的 hasDisplayProjection 启发式由 PR 的权威标志取代——该标志正是为此引入。
  • client.ts 保留 main 的 displayText 回退链与 hookContext: additionalContext,叠加 PR 的 displayProjection: true。
  • 为让两侧行为同时成立,确立一条合并规则:Mod 提交的 prompt 向标题 ring 贡献 null,且不能单独把会话切入投影模式。 否则 main 的「仅 Mod 轮次时仍从对话取标题」会被 PR 的投影路径吃掉。该规则在冷启动累加器(session-transcript-reader.ts)中镜像,使重启后推导出同一状态;rebuildTurnBoundaries 现在同时重推标题文本与 Mod 键,restore 与消息删除路径不再二次记账。
  • 新增 2 条回归用例钉住该规则,各做一次变异见证(去掉 !isModPrompt 守卫则红,还原则绿)。既有断言只改靶不删除——autoTitle 套件相对 main 为 +478/−0。

两条红检查

SDK Java (full) / Hosted process fault gates / MySQL 8.4 / Java 21 及其汇总 job Managed Agent gate(run 41170201360)不是本 PR 的缺陷:

  • diff 共 46 个文件,全为 TypeScript 或文档,零 Java 文件。
  • 日志中 20 个 error 全是 Could not find field 'attachments' on target object [...SdkHarnessBoundary],外加 NoSuchBeanDefinition ... HarnessPoolConnector。而 SdkHarnessBoundary.java 在合并前的 PR head 树里根本不存在。
  • 该 run 由 2026-10-31 的 pull_request_review 触发,编的是当时 main 中间态 Java 代码的 merge ref。当前 main 已把 packages/managed-agent-server 重构为多模块 Maven 布局。
  • 本次合并后,packages/managed-agent-server 与 packages/sdk-java 与当前 main 逐字节相同(git diff origin/main HEAD -- … 为空)。

未改任何 Java 代码与 CI 配置。未在本地运行 Java lane,也不声称其变绿——由新 head 的 CI 重新评判。

验证(本地真实结果)

node scripts/setup-worktree.js → 0;npm run build → 0;npm run typecheck → 0;11 个改动文件 eslint → 0;暂存集 prettier --check 全合规;pre-commit lint-staged 钩子通过。

Vitest 逐文件运行,全绿 —— core:sessionTitle 42、autoTitle 38、chatRecordingService 198、title-sidecar 19、session-title-sidecar 4、sessionService.title-sidecar 13、customTitle 24、message-deletion 16、session-transcript-reader 157、writer-lease 138(+10 skip)、corruption 76、sessionService 268、bun-pty 32(+1 skip)、client 624;acp-bridge:4 / 3 / 15 / 98 / 173;sdk-typescript:34 / 44 / 23 / 11;cli:renameCommand 25、box-metrics 13、scheduled-tasks 138、hosted-hook 129、hosted-harness 708、acpAgent 939、Session 1446。

三次非断言性红已定位并排除,未压制:client.test.ts 默认堆 OOM(改 12 GiB / 单 worker 重跑);acpAgent.test.ts 继承了环境泄漏的 QWEN_CODE_SERVE=1 / QWEN_SERVE_MCP_BUDGET_MODE=off(unset 后 939 全绿);coverage/.tmp 竞写(单独重跑 1446 全绿)。提交后独立重跑 sessionTitle.test.ts:42/42。

合并保真度双向核对:提交相对所合并 main 的差异恰好是本 PR 的 46 个文件,无残留冲突标记,工作树干净。

本轮期间的 base 漂移

本轮运行期间 main 前进了 5 个提交(至 5faa620af9b)。git merge-tree HEAD origin/main 对其为干净,且这 5 个提交与本 PR 的 46 个文件零重叠。

批准

main 要求 2 个批准且推送即作废旧批准,故本次推送作废了 2026-10-31 那条批准。需要 2 个新批准。 本值守不 approve、不改评审状态、不合并、不开自动合并。

遗留项

本轮读到的 102 条 review / issue_comment 事件均为历史轮次汇报、base 更新通知与维护者验证报告,已由 2026-10-31 的接管处置;80 条评审线程全部 resolved,其中未发现遗留请求。第 13 轮延后的 3 条 Critical(fails-closed / new-surface,明确非阻断)与第 6 轮遗留的 2 条 Suggestion 仍记录在 #13282,本轮未重开。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

@wenshao

wenshao commented Nov 7, 2026

Copy link
Copy Markdown
Collaborator

Watch round — web-shell E2E Smoke (shard 4/4) needs a human decision

No code was changed and nothing was pushed this round. Head remains ce481fe9f9.

What happened

The first run of this lane on ce481fe9f9 failed 3 of 4 shards. It was re-run once, which cleared most of it:

Shard First run After re-run
1/4 5 failed, 21 flaky, 96 passed success
2/4 25 failed, 16 flaky, 71 passed still running
3/4 success success
4/4 12 failed, 11 flaky, 60 passed 2 failed, 9 flaky, 72 passed

The re-run budget for this check on this head is now spent: the helper reports it "was re-run on this head already; treat the failure as real", so this round records it as needing a human rather than spending another re-run.

Why this still looks environmental, not a defect in this PR

Offered as evidence for whoever decides, not as a claim that the check is green:

  • This PR changes no web-shell code. Its diff against the merged main is exactly 46 files, none under packages/web-shell, and neither failing spec (web-shell.composer.mobile.spec.ts) is touched.
  • The remaining two failures are page-not-mounted, not assertion mismatches. Both are expect(locator).toBeVisible() with Error: element(s) not found — textarea[data-web-shell-composer-editor] at composer.mobile.spec.ts:1111, and [data-web-shell-root]:not([data-web-shell-gate]) at :1202. The root element itself never rendered, after both retries.
  • The first run's dominant signature was backend unavailability: TimeoutError: page.waitForFunction: Timeout 20000ms exceeded inside the E2E helper's waitForConnection(sessionId, { timeout = 20_000 }), plus sustained connect ECONNREFUSED 127.0.0.1:4170 on /workspace/models and /workspace/hooks. That signature is gone after the re-run (0 occurrences), and the failure count fell from 12 to 2.
  • An unrelated PR shows the same signatures at the same time. feat/core-pr-view (run 42359558892, shard 4/4) has 26 waitForConnection hits, 233 ECONNREFUSED, 47 data-web-shell-root / composer-editor locator failures with element(s) not found, and the same two composer.mobile.spec.ts:863 / :1292 cases entering retries. Two further unrelated PRs failed this lane in the same window with different shard subsets (codex/managed-session-diagnostics-20261106: 2/4; fix/14481-a3-rewritten-hook: 1/4 + 2/4).

What a person has to decide

Whether to accept this lane as environmental on ce481fe9f9 and re-run it once run 42359021212 finishes (shard 2/4 is still in flight), or to treat the two remaining mobile-chromium cases as real and investigate them. This watch cannot spend another re-run on this head, and it does not approve, change review state, merge, or enable auto-merge.

For context, main itself is green on this workflow in the same window (runs at 09:42–09:45 UTC and earlier all success), so this is not a broken base; it is concentrated on PR runners.

中文说明

值守轮次 —— web-shell E2E Smoke (shard 4/4) 需要人工决定

本轮未改任何代码、未推送,head 仍为 ce481fe9f9。

经过

该 lane 在 ce481fe9f9 上首次运行 4 个 shard 中 3 个红。重跑一次后大部分转绿:

Shard 首次 重跑后
1/4 5 failed、21 flaky、96 passed success
2/4 25 failed、16 flaky、71 passed 仍在跑
3/4 success success
4/4 12 failed、11 flaky、60 passed 2 failed、9 flaky、72 passed

该 check 在本 head 的重跑预算已用完:helper 报「was re-run on this head already; treat the failure as real」,故本轮按流程记为需要人工,而不是再花一次重跑。

为何仍倾向环境性、而非本 PR 缺陷

以下是给决定者的证据,不是「该检查已绿」的主张:

  • 本 PR 不改任何 web-shell 代码:相对所合并 main 的差异恰好 46 个文件,无一在 packages/web-shell 下,失败的 web-shell.composer.mobile.spec.ts 也未被触及。
  • 剩余两条失败是页面未挂载,而非断言不符:均为 expect(locator).toBeVisible() 且 Error: element(s) not found——composer.mobile.spec.ts:1111 的 textarea[data-web-shell-composer-editor],以及 :1202 的 [data-web-shell-root]:not([data-web-shell-gate])。两次重试后根元素始终未渲染。
  • 首次运行的主导签名是后端不可用:E2E helper waitForConnection(sessionId, { timeout = 20_000 }) 内的 TimeoutError: page.waitForFunction: Timeout 20000ms exceeded,加上 /workspace/models、/workspace/hooks 持续 connect ECONNREFUSED 127.0.0.1:4170。重跑后该签名归零,失败数由 12 降至 2。
  • 无关 PR 在同一时间出现相同签名:feat/core-pr-view(run 42359558892,shard 4/4)有 26 次 waitForConnection、233 次 ECONNREFUSED、47 处 data-web-shell-root / composer-editor 定位失败伴 element(s) not found,且同两条 composer.mobile.spec.ts:863 / :1292 同样进入重试。另两个无关 PR 在同一时间窗以不同 shard 组合失败(codex/managed-session-diagnostics-20261106 红 2/4;fix/14481-a3-rewritten-hook 红 1/4+2/4)。

需要人工决定的内容

是接受该 lane 在 ce481fe9f9 上属环境性、待 run 42359021212 收尾后(shard 2/4 仍在飞)再重跑一次,还是把剩余两条 mobile-chromium 用例当作真实缺陷去排查。本值守无法在本 head 再花一次重跑,也不 approve、不改评审状态、不合并、不开自动合并。

补充参照:同一时间窗内 main 自身在该 workflow 上为绿(09:42–09:45 UTC 及更早的 run 均 success),故不是 base 损坏,问题集中在 PR runner。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

@wenshao

wenshao commented Nov 7, 2026

Copy link
Copy Markdown
Collaborator

Watch round — the whole web-shell E2E Smoke lane now needs a human

Update to the previous note. No code changed, nothing pushed; head remains ce481fe9f9. Run 42359021212 has now completed with failure.

Final state after the one re-run

Job First run After re-run
shard 1/4 5 failed, 21 flaky, 96 passed success
shard 2/4 25 failed, 16 flaky, 71 passed 21 failed, 26 flaky, 65 passed
shard 3/4 success success
shard 4/4 12 failed, 11 flaky, 60 passed 2 failed, 9 flaky, 72 passed
summary job failure failure (test "$SMOKE_RESULT" = success)

Shard 2/4 and the summary job have now also spent their re-run on this head — the helper reports "was re-run on this head already; treat the failure as real" — so both are recorded as needing a human, joining shard 4/4 from the previous round. The summary job is only the aggregate gate over the shards, not an independent failure.

What the re-run changed, and what it did not

The re-run halved the lane (shard 1/4 went green, shard 4/4 fell from 12 failures to 2) but did not clear shard 2/4. Its backend-unavailability signature recurred: 7 waitForConnection hits and 3 TimeoutError: page.waitForFunction: Timeout 20000ms exceeded, with 8 ECONNREFUSED. Its 21 failures again span unrelated feature areas — web-shell.files.spec.ts (6), loading-phrases, managed-progress, and others — and it ran 32.6 minutes, the longest shard.

The attribution evidence from the previous note still holds and was re-checked this round: this PR's diff against the merged main is exactly 46 files with none under packages/web-shell, and the unrelated PR feat/core-pr-view shows the same signatures in the same window. main itself is green on this workflow in that window.

What a person has to decide

The re-run limit that stopped this round is this watch's own budget, not GitHub's — a person can re-run the lane freely. The decision is whether to:

  1. treat the lane as a runner/daemon-environment problem, fix or wait out the environment, and re-dispatch it; or
  2. treat the surviving failures as real and investigate them, starting with shard 2/4, which did not improve under re-run.

This watch will not spend more re-runs on ce481fe9f9, and it does not approve, change review state, merge, or enable auto-merge. Separately, the PR still needs 2 approving reviews — the merge push dismissed the 2026-10-31 approval under main's stale-review rule.

中文说明

值守轮次 —— 整条 web-shell E2E Smoke lane 均需人工处置

本条是上一条说明的增量更新。未改代码、未推送,head 仍为 ce481fe9f9。Run 42359021212 现已收尾,结论 failure。

一次重跑后的最终状态

Job 首次 重跑后
shard 1/4 5 failed、21 flaky、96 passed success
shard 2/4 25 failed、16 flaky、71 passed 21 failed、26 flaky、65 passed
shard 3/4 success success
shard 4/4 12 failed、11 flaky、60 passed 2 failed、9 flaky、72 passed
汇总 job failure failure(test "$SMOKE_RESULT" = success)

shard 2/4 与汇总 job 在本 head 的重跑预算同样已用完——helper 报「was re-run on this head already; treat the failure as real」——故两者均记为需要人工,与上一轮的 shard 4/4 并列。汇总 job 只是各 shard 的聚合闸,不是独立失败。

重跑改善了什么、没改善什么

重跑使该 lane 失败面减半(shard 1/4 转绿,shard 4/4 由 12 条降至 2 条),但未能消除 shard 2/4 的失败。其后端不可用签名复现:7 次 waitForConnection、3 次 TimeoutError: page.waitForFunction: Timeout 20000ms exceeded,伴 8 次 ECONNREFUSED。21 条失败再次横跨互不相关功能区——web-shell.files.spec.ts(6 条)、loading-phrases、managed-progress 等——且耗时 32.6 分钟,为最长 shard。

上一条说明中的归因证据本轮已复核并仍成立:本 PR 相对所合并 main 的差异恰好 46 个文件,无一在 packages/web-shell 下;无关 PR feat/core-pr-view 在同一时间窗呈现相同签名;main 自身在该窗口内为绿。

需要人工决定的内容

本轮停止所依据的重跑上限是本值守自己的预算,不是 GitHub 的限制——人工可自由重跑该 lane。需要决定的是:

  1. 将其视为 runner/daemon 环境问题,修复或等待环境恢复后重新派发;或
  2. 将存活失败视为真实缺陷排查,从重跑后未改善的 shard 2/4 入手。

本值守不会在 ce481fe9f9 上再花重跑,也不 approve、不改评审状态、不合并、不开自动合并。另外,本 PR 仍需要 2 个批准——按 main 的陈旧评审规则,合并推送已作废 2026-10-31 的那条批准。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

@qwen-code-review-bot qwen-code-review-bot left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Partially reviewed — gaps disclosed.

Not reviewed: reverse audit — stopped after audit round 4 without two consecutive dry rounds: rounds 1-4 each reported new findings, so the loop was closed unconverged rather than dry.

Not explored to full depth (tool budget reached): "agent 1c": did not read the web-shell chat-message renderer to confirm whether an empty content: '' user message is filtered before paint (the blank-turn part of the sce…; "agent 1c": did not run the affected suites ( packages/core , packages/acp-bridge , packages/sdk-typescript ) to confirm the added tests pass at this head..

Deferred under the convergence posture (round 24, not a blocker) — recorded, not requested in this round:

  • docs/design/daemon-embedded-resource-replay.md:23 — [review] The design doc states the journal's block cap *matches* MAX_ECHO_CONTENT_BLOCKS , but the implementation re-declares the number as an independent literal ( export const MAX_RECOR…
  • packages/acp-bridge/src/bridge-embedded-resource-replay.test.ts:108 — [review] The only test of the 256 KiB daemon-native exemption accepts *any* string as the expanded resource text, so the very claim in its name ("without applying the inl…
  • packages/acp-bridge/src/embedded-resource-replay.test.ts:171 — [review] The clone is pinned only by reference identity ( not.toBe at :39), so a shallow clone that shares the nested _meta object with the caller's block passes both this te…
  • packages/acp-bridge/src/embedded-resource-replay.test.ts:183 — [review] The only test of the nativeResourceIndexes exemption uses a 6-byte native block, so the ordering that makes the exemption correct (the nativeIndexes.has(index) skip…
  • packages/acp-bridge/src/embedded-resource-replay.test.ts:193 — [review] The only test for readDaemonAttachmentReferences pins the count cap and the happy path; none of its ~9 per-field rejection predicates is exercised. Failure scenario: …
  • packages/acp-bridge/src/transcript-replay.ts:1517 — [review] The new truncation-placeholder gate re-derives "has visible text" from the raw payload['displayText'] , while every other place in projectUserRecord decides visibility from st…
  • packages/cli/src/acp-integration/session/Session.test.ts:17564 — [review] The "exceed the budget" fixture sits exactly on MAX_RECORDED_EMBEDDED_RESOURCES_BYTES (262,144 code units), so it drives only the serialized-size branch; the siblin…
  • packages/cli/src/acp-integration/session/Session.ts:12463 — [review] Mid-turn drained user messages never journal their embedded text resources, so the incident this PR fixes still happens for a steering message that carries a selection. Fa…
  • packages/cli/src/ui/commands/renameCommand.ts:191 — [review] The new projectionMode === false arm in /rename has no test, and no test distinguishes its two arms. Failure scenario: Deleting the : [] arm keeps renameCommand.test.ts gr…
  • packages/cli/src/ui/commands/renameCommand.ts:191 — [review] The gate's false branch — the one the new comment justifies ("fall back to the full history instead") — is not pinned by any test: every /rename title-path test either uses a re…
  • packages/core/src/services/chat-recording-message-deletion.test.ts:176 — [review] The rewind regression test rewinds to turn 0, discarding both recorded turns, so it exercises only the empty-window case and cannot distinguish the correct tr…
  • packages/core/src/services/chatRecordingService.autoTitle.test.ts:846 — [review] The only test for the new Mod-prompt title exemption records a *framed* Mod prompt ( asUser absent); the asUser arm — the one trackTitleText 's own modPro…
  • packages/core/src/services/chatRecordingService.test.ts:3242 — [review] No test pins the restore side of the retained-title-text mechanism: restoreProjectedState assigns lastProjectionTextForTitle (chatRecordingService.ts:1832) but ever…
  • packages/core/src/services/chatRecordingService.test.ts:3367 — [review] The new realigns title state when a rebuilt boundary list shrinks test pins only the display-text ring reset in rebuildTurnBoundaries , not the per-turn turnTitleSt…
  • packages/core/src/services/session-transcript-reader.test.ts:1876 — [review] The cold-path emission of userDisplayTextsForTitle — new behaviour in this diff — has no assertion that can fail; the rewind test's not.toContain is satisfied …
  • packages/core/src/services/session-transcript-reader.ts:288 — [review] SessionRuntimeResumeState.recording declares only turnTitleStates of the four title-state fields the reader emits; displayProjectionForTitle , userDisplayTextsForT…
  • packages/core/src/services/session-transcript-reader.ts:1359 — [review] The widened navigation-visibility predicate enumerates three of UserPromptRecordPayload 's four visibility-bearing reference fields but omits resourceLinks . Failure …
  • packages/core/src/services/sessionService.corruption.test.ts:273 — [review] The rewritten assertion leaves the test named "does not extract a uuid from a payload object inside a partial-tail fragment" ( :260 ) with no partial-tail fragment …
  • packages/core/src/services/sessionService.corruption.test.ts:319 — [review] The rationale paragraph above the aligned-boundary test still explains the case with "The fix peeks the byte before readStart to tell the two apart" — the rewritt…
  • packages/core/src/services/sessionService.corruption.test.ts:325 — [review] Every byte-geometry test of the new walk is pure ASCII ( 'x' , 'y' , 'A' , 'B' , 'C' ), so the one reason the walk carries spanParts as raw Buffer s instead …
  • …and 14 more (see the run report)

Mechanism health: this round did not close cleanly, so it withholds the incremental anchor — and the round it recovered had no anchor this round could use either — none at all, one with no certifier, one certified by an identity other than the one this round runs under, or one this round's fetch refused or resolved to the head — so the next review re-reads the whole diff unless recovery grafts an earlier own anchor that the round running it can use onto the complete work list this round leaves behind, and keeps doing so until a round's marker carries an anchor again or a graft lands that the round running it can use. (Stated, not acted on — this changes nothing about what the round posts.)

中文说明

仅完成部分审查,审查缺口已披露。

未审查(原文为英文):reverse audit — stopped after audit round 4 without two consecutive dry rounds: rounds 1-4 each reported new findings, so the loop was closed unconverged rather than dry.

未探索到全部深度(达到工具调用预算):"agent 1c":did not read the web-shell chat-message renderer to confirm whether an empty content: '' user message is filtered before paint (the blank-turn part of the sce…;"agent 1c":did not run the affected suites ( packages/core , packages/acp-bridge , packages/sdk-typescript ) to confirm the added tests pass at this head.。

收敛姿态下延后(第 24 轮,非阻断)——已记录,本轮不要求修改:共 34 条(原文未翻译,列表见上方英文部分)。

机制健康:本轮未能干净收尾,因而扣留了增量锚点,而它恢复到的那一轮也没有留下本轮可用的锚点——要么完全没有、要么没有认证者、要么由本轮运行身份之外的身份认证、要么被本轮的获取拒绝或解析为头提交——因此下一次评审将重读整个 diff,除非恢复流程把本轮能使用的更早自有锚点嫁接到本轮留下的完整工作清单上;并会一直如此,直到某一轮的标记重新带上锚点,或落地的嫁接能被运行该轮的评审使用。(仅陈述,不据此行动——这不改变本轮发布的任何内容。)

— deepseek-v4.1-flash-1017 via Qwen Code /review (v0.26.9)

@wenshao

wenshao commented Nov 7, 2026

Copy link
Copy Markdown
Collaborator

Watch round — review round 24 dispositioned; one disclosed gap closed with evidence

No code changed, nothing pushed. Head remains ce481fe9f9.

Review round 24 carries no actionable request, so this round makes no change. Its own ledger reads findings: [], posted: 0, fresh: 0, and all 34 listed items are marked "Deferred under the convergence posture (round 24, not a blocker) — recorded, not requested in this round". The reverse-audit note (stopped after round 4 without two consecutive dry rounds) and the anchor-withholding paragraph are mechanism-health statements about the review pipeline, not requests against this code.

Closing the disclosed test gap

The review discloses that it did not run the affected suites (packages/core, packages/acp-bridge, packages/sdk-typescript) to confirm the added tests pass at this head. Those suites were run at ce481fe9f9 during the takeover round, one file at a time, all green:

  • packages/core (14 files): sessionTitle 42, autoTitle 38, chatRecordingService 198, title-sidecar 19, session-title-sidecar 4, sessionService.title-sidecar 13, customTitle 24, message-deletion 16, session-transcript-reader 157, writer-lease 138 (+10 skip), corruption 76, sessionService 268, bun-pty 32 (+1 skip), client 624
  • packages/acp-bridge (5 files): 4 / 3 / 15 / 98 / 173
  • packages/sdk-typescript (4 files): 34 / 44 / 23 / 11

npm run build and npm run typecheck were additionally re-run on the pushed commit itself (both exit 0, zero error TS), as were client.test.ts (624), sessionTitle.test.ts (42) and session-transcript-reader.test.ts (157) — so the green is bound to this head, not to an earlier tree.

One deferred item verified as accurate, and why it is not acted on here

chatRecordingService.autoTitle.test.ts:846 — the claim that the Mod-prompt title exemption is tested only with a framed Mod prompt is correct: that file has a single provenance: 'mod_prompt' fixture (:834) and no asUser occurrence at all, so the isModPrompt && !payload?.asUser arm at chatRecordingService.ts:1773 is not pinned by that suite. This touches the merge invariant introduced when main was merged into this branch, so it is recorded here rather than dropped.

It is left for a follow-up rather than fixed in this round because the reviewer explicitly defers it as non-blocking, and pushing now would re-trigger the web-shell E2E Smoke lane that is currently failing on runner/daemon unavailability — producing more red without unblocking anything. It belongs with the residual suggestions already tracked in #13282.

Still outstanding for a person

Unchanged from the previous note: the web-shell E2E Smoke lane needs a decision (this watch's re-run budget on this head is spent; a person may re-run freely), and the PR needs 2 approving reviews.

中文说明

值守轮次 —— 第 24 轮评审已处置,并用证据关闭其披露的一个缺口

未改代码、未推送,head 仍为 ce481fe9f9。

第 24 轮评审不含可执行请求,故本轮不做改动。其自带台账为 findings: []、posted: 0、fresh: 0,所列 34 条全部标注「收敛姿态下延后(第 24 轮,非阻断)——已记录,本轮不要求修改」。反向审计说明(第 4 轮后未达两次连续干燥即停止)与扣留锚点那段,都是评审流水线的机制健康陈述,不是针对本代码的请求。

关闭其披露的测试缺口

评审披露其「未运行受影响套件(packages/core、packages/acp-bridge、packages/sdk-typescript)以确认新增测试在本 head 通过」。这些套件在接管轮次中确已于 ce481fe9f9 上逐文件运行,全绿:

  • packages/core(14 个文件):sessionTitle 42、autoTitle 38、chatRecordingService 198、title-sidecar 19、session-title-sidecar 4、sessionService.title-sidecar 13、customTitle 24、message-deletion 16、session-transcript-reader 157、writer-lease 138(+10 skip)、corruption 76、sessionService 268、bun-pty 32(+1 skip)、client 624
  • packages/acp-bridge(5 个文件):4 / 3 / 15 / 98 / 173
  • packages/sdk-typescript(4 个文件):34 / 44 / 23 / 11

此外,npm run build 与 npm run typecheck 已在已推送的提交本身上重跑(均退出码 0、零 error TS),client.test.ts(624)、sessionTitle.test.ts(42)、session-transcript-reader.test.ts(157)亦然——因此该绿绑定于本 head,而非更早的树。

一条延后项经核实属实,及本轮不处置的理由

chatRecordingService.autoTitle.test.ts:846 —— 「Mod-prompt 标题豁免仅以 framed Mod prompt 测试」这一主张属实:该文件只有一处 provenance: 'mod_prompt' 夹具(:834),且全文无 asUser,故 chatRecordingService.ts:1773 的 isModPrompt && !payload?.asUser 分支未被该套件钉住。此处触及把 main 合入本分支时确立的合并不变量,故在此记录而非略过。

本轮将其留作后续而非就地修复,理由是评审已明确将其延后为非阻断项,且此刻推送会重新触发当前因 runner/daemon 不可用而失败的 web-shell E2E Smoke lane——只增加红,不解开任何卡点。该项应与已记录在 #13282 的遗留建议一并处理。

仍待人工

与上一条说明一致:web-shell E2E Smoke lane 需要决定(本值守在本 head 的重跑预算已用完,人工可自由重跑),且本 PR 需要 2 个批准。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

@qwen-code-review-bot qwen-code-review-bot left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Partially reviewed — gaps disclosed.

Not reviewed: chunk 10 cross-file trace — stopped at the agent tool budget: the auditor did not finish tracing runtime.recording into the recorder's activate/constructor call site.

Not explored to full depth (tool budget reached): "agent reverse-audit (round 3)": the per-field rotation's "larger one" choice (transcript.ts:3042-3050, outside my chunk) compares .length while the ceiling is in bytes; I did not file it sep…; "agent reverse-audit (round 1)": could not walk the replay/emitter side ( transcript-replay.ts projection of embeddedResources / attachmentReferences and the SDK normalizer) — it lives outsi…; chunk 10: did not finish tracing runtime.recording into the recorder's activate /constructor call site — the grep was too noisy to read and the file belongs to another….

Mechanism health: this round did not close cleanly, so it withholds the incremental anchor — and the round it recovered had no anchor this round could use either — none at all, one with no certifier, one certified by an identity other than the one this round runs under, or one this round's fetch refused or resolved to the head — so the next review re-reads the whole diff unless recovery grafts an earlier own anchor that the round running it can use onto the complete work list this round leaves behind, and keeps doing so until a round's marker carries an anchor again or a graft lands that the round running it can use. (Stated, not acted on — this changes nothing about what the round posts.)

中文说明

仅完成部分审查,审查缺口已披露。

未审查(原文为英文):chunk 10 cross-file trace — stopped at the agent tool budget: the auditor did not finish tracing runtime.recording into the recorder's activate/constructor call site.

未探索到全部深度(达到工具调用预算):"agent reverse-audit (round 3)":the per-field rotation's "larger one" choice (transcript.ts:3042-3050, outside my chunk) compares .length while the ceiling is in bytes; I did not file it sep…;"agent reverse-audit (round 1)":could not walk the replay/emitter side ( transcript-replay.ts projection of embeddedResources / attachmentReferences and the SDK normalizer) — it lives outsi…;chunk 10:did not finish tracing runtime.recording into the recorder's activate /constructor call site — the grep was too noisy to read and the file belongs to another…。

机制健康:本轮未能干净收尾,因而扣留了增量锚点,而它恢复到的那一轮也没有留下本轮可用的锚点——要么完全没有、要么没有认证者、要么由本轮运行身份之外的身份认证、要么被本轮的获取拒绝或解析为头提交——因此下一次评审将重读整个 diff,除非恢复流程把本轮能使用的更早自有锚点嫁接到本轮留下的完整工作清单上;并会一直如此,直到某一轮的标记重新带上锚点,或落地的嫁接能被运行该轮的评审使用。(仅陈述,不据此行动——这不改变本轮发布的任何内容。)

— deepseek-v4.1-flash-1017 via Qwen Code /review (v0.26.9)

The skeleton cascade sizes its byte-budget cut from an exact JSON-byte
measure, but the walk billed every code unit its worst-case escaped
width: a valid astral pair cost 12 against its real 4 bytes, and the
short escapes cost 6 against 2. A resource landing just over the 256 KiB
ceiling therefore lost most of a legitimate text, and replay presented
that over-cut value as correctly budgeted.

Bill each unit its exact escaped width and consume a valid surrogate
pair as one 4-byte unit so the cut never splits it. Every term remains
an upper bound (now an equality), so the convergence invariant and the
23-byte minimum trimmed form are unchanged.
@wenshao

wenshao commented Nov 7, 2026

Copy link
Copy Markdown
Collaborator

Watch round — Critical R25-31 fixed and pushed

Head ce481fe9f9 → 9493bb3aaa. One focused fix, two files, +89/−13. No base merge this round: there was no failed check to attribute, so merging main would only have widened the diff.

The CHANGES_REQUESTED review carried one request — Critical R25-31 at packages/sdk-typescript/src/daemon/ui/transcript.ts:2768, axis certifies-falsely. It is fixed and answered in its thread, with the billing change, the named floor constraint checked explicitly, and two mutation witnesses (worst-case billing restored → red at ~111 KB; a pair-splitting mutant → red on the prefix assertion).

Verified: whole packages/sdk-typescript suite 2432 passed / 6 skipped (no existing assertion retargeted — the test file is +63/−0), root npm run typecheck exit 0 with zero error TS, eslint and prettier clean. Independently re-checked on the pushed commit: target file 35/35, root typecheck exit 0, and an out-of-tree probe reproducing the over-cut (old 110,913 bytes vs fixed 262,113 against a 262,144 ceiling).

Left untouched, deliberately

  • The review's disclosed gaps (chunk 10 cross-file trace, the reverse-audit depth limits) and its mechanism-health / anchor-withholding paragraph are statements about the review pipeline, not requests against this code.
  • The unfiled observation in the review body is confirmed present: the field rotation at transcript.ts:3060 picks the "larger" field with mimeTypeValue.length >= skeleton.resource.uri.length, comparing UTF-16 code units while the ceiling is in real bytes, so a shorter-but-wider mimeType sorts second. The reviewer explicitly did not file it, so it was not changed this round; recording it here so it is not lost.
  • The 34 round-24 deferrals and the suggestions in Deferred review findings from PR #12585: fix(acp): persist embedded text resources for transcript replay #13282 were not reopened.

CI and approvals

This push starts a fresh CI evaluation on 9493bb3aaa, which also means the web-shell E2E Smoke lane runs again with a fresh per-head re-run budget. The three earlier needs_human entries were recorded against ce481fe9f9; if that lane fails again on the new head, the next round will triage it and can spend one re-run. The underlying attribution has not changed: this PR touches no packages/web-shell file, and the failures carried a runner/daemon-unavailability signature also seen on unrelated PRs.

The PR still needs 2 approving reviews. This watch does not approve, change review state, merge, or enable auto-merge.

中文说明

值守轮次 —— Critical R25-31 已修复并推送

Head ce481fe9f9 → 9493bb3aaa。一次聚焦修复,2 个文件,+89/−13。本轮未合并 base:没有需要归因的失败检查,合并 main 只会扩大 diff。

该 CHANGES_REQUESTED 评审只含一项请求——packages/sdk-typescript/src/daemon/ui/transcript.ts:2768 的 Critical R25-31,失败方向轴为 certifies-falsely。已修复并在其线程内回复,含计费改动、对点名下限约束的显式核实,以及两个变异见证(恢复 worst-case 计费 → 红于约 111 KB;切半变异 → 红于前缀断言)。

验证:packages/sdk-typescript 整包套件 2432 通过/6 跳过(无既有断言改靶——测试文件 +63/−0),根 npm run typecheck 退出码 0、零 error TS,eslint 与 prettier 干净。在已推送提交上独立复核:目标文件 35/35、根 typecheck 退出码 0,并用工作树外探针复现过度裁剪(旧 110,913 字节 vs 修复后 262,113,上限 262,144)。

有意未动的部分

  • 评审披露的缺口(chunk 10 跨文件追踪、反向审计深度限制)及其机制健康/扣留锚点段落,都是关于评审流水线的陈述,不是针对本代码的请求。
  • 评审正文中那条未立案的观察确实存在:transcript.ts:3060 的字段轮转用 mimeTypeValue.length >= skeleton.resource.uri.length 选「较大者」,比较的是 UTF-16 码元数而上限是真实字节,因此「码元更短但字节更宽」的 mimeType 会被排到第二位。评审明确未立案,故本轮未改动;在此记录以免丢失。
  • 第 24 轮的 34 条延后项与 Deferred review findings from PR #12585: fix(acp): persist embedded text resources for transcript replay #13282 中的建议均未重开。

CI 与批准

本次推送会在 9493bb3aaa 上启动全新 CI 评判,也意味着 web-shell E2E Smoke lane 重新运行并获得新的按 head 计的重跑预算。先前三条 needs_human 记录针对的是 ce481fe9f9;若该 lane 在新 head 上再次失败,下一轮会重新定性并可用掉一次重跑。底层归因未变:本 PR 不触及任何 packages/web-shell 文件,且失败带有在无关 PR 上同样出现的 runner/daemon 不可用签名。

本 PR 仍需要 2 个批准。本值守不 approve、不改评审状态、不合并、不开自动合并。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

@wenshao

wenshao commented Nov 7, 2026

Copy link
Copy Markdown
Collaborator

Watch round — stale lint gate refreshed by merging main

Head 9493bb3aaa → 4d8eba89e0e (merge of main a7a65e17833, 35 commits). No code change and no CI file was hand-edited.

Why the lane was red

Lint & Static (ubuntu-latest, Node 22.x) failed without running a single linter. The job lasted 48 seconds: step 11 Check lint gate freshness failed, and every step after it was skipped — Run ESLint, Run actionlint, Run shellcheck, Run yamllint, Run Prettier, the sensitive-keyword linter, the i18n check, the notices check and the helper tests.

The gate's own message named the cause:

The lint gate changed on 'main' after this branch last incorporated it: .github/workflows/ci.yml: 708ba347d355 feat(ci): add disposable external PR runners (#16513) … This lane checks out the branch head alone, so its green proves the branch passes the gate as the branch defines it — with the files above, that gate is stale. Merge or rebase 'main' into this branch and push to re-validate under the current gate.

So the remedy was a base merge, not a code fix. Taking main's ci.yml through the merge is what re-validation requires; it is not a weakening edit, and the PR's effective diff still contains no CI or gate file.

Gate freshness, verified

All 9 files in the checker's GATE_FILES were checked by asking whether main's last commit touching each one is an ancestor of the new head: 9/9 fresh, including .github/workflows/ci.yml → 708ba347d355, the exact commit the gate named. git diff a7a65e17833 HEAD -- .github/ is empty.

One caveat worth stating: running check-lint-gate-freshness.mjs locally exits 0 but prints BASE_REF, HEAD_SHA, GITHUB_REPOSITORY or GITHUB_TOKEN missing; skipping the check. That is a vacuous pass and is not used as evidence here — the ancestry check above is.

Merge safety

git merge-tree reported no textual conflict. Since the merge base, main changed 306 files and this PR 46, with a real overlap of exactly two: packages/acp-bridge/src/session-control-plane.ts and packages/cli/src/serve/hosted-harness-session.test.ts. Both sides were read in the merged tree and both survive — main's allowChannelSubscribers guard and reaper call site, its it.each retry-refused rewrite and two new cleanup tests; this PR's attachment-resource-index metadata handling and its approvalSelections injections. A clean textual merge is not treated as proof here: this PR previously had a semantic merge conflict that merged clean but broke tsc (finding L-1), so the typecheck below is the load-bearing evidence.

The previous round's R25-31 fix is intact: transcript.ts and daemon-embedded-resources.test.ts are byte-identical to 9493bb3aaa, with end += paired ? 2 : 1 still present and the astral/CJK case still in the suite.

Verification

npm run build exit 0; npm run typecheck exit 0 with zero error TS; 17 vitest suites run one file at a time from inside each package directory, 3420 passed / 11 skipped — including both overlapping files' suites (acp-bridge/bridge.test.ts 1060, cli/hosted-harness-session.test.ts 711) and the R25-31 pin (daemon-embedded-resources.test.ts 35).

Independently re-run in the maintainer session on the pushed merge commit: root npm run typecheck exit 0, zero error TS, and daemon-embedded-resources.test.ts 35/35.

One infrastructure red was diagnosed rather than suppressed: core/client.test.ts first failed with a vitest-worker V8 heap OOM at the local 4144 MB default (v8 coverage is on by default outside CI) followed by ERR_IPC_CHANNEL_CLOSED, with zero tests executed and no assertion failure; re-run at 6144 MB — the repository's own build heap setting — it passed 624. No assertion was changed or loosened.

The push re-validates the lane under the current gate; only CI can show it green, and no green lint lane is claimed here.

Still outstanding

The PR needs 2 approving reviews. The web-shell E2E Smoke lane re-evaluates on this new head with a fresh per-head re-run budget; the three earlier needs_human entries were recorded against ce481fe9f9. This watch does not approve, change review state, merge, or enable auto-merge.

中文说明

值守轮次 —— 合并 main 以刷新失效的 lint 门禁

Head 9493bb3aaa → 4d8eba89e0e(合并 main a7a65e17833,35 个提交)。未改任何代码,未手工编辑任何 CI 文件。

该 lane 为何红

Lint & Static (ubuntu-latest, Node 22.x) 一步 linter 都没跑就失败了。job 仅 48 秒:第 11 步 Check lint gate freshness 失败,其后全部跳过——Run ESLint、Run actionlint、Run shellcheck、Run yamllint、Run Prettier、敏感词检查、i18n 检查、notices 检查与 helper 测试。

闸门自身指明了原因:main 上的 .github/workflows/ci.yml 已由 708ba347d355(#16513,add disposable external PR runners)改动,而本分支尚未纳入;该 lane 只检出分支 head,因此其绿色只证明分支通过了「分支自己定义的」门禁,而那份门禁已失效。处置是合并 main 后推送,在当前门禁下重新验证。

因此本轮是合 base,而非改代码。通过合并取得 main 的 ci.yml 正是重新验证所要求的,不属于削弱门禁;本 PR 的有效 diff 仍不含任何 CI 或 gate 文件。

门禁新鲜度已核实

对检查器 GATE_FILES 中全部 9 个文件,逐个判定 main 侧最后触碰它的提交是否为新 head 的祖先:9/9 新鲜,含闸门点名的 .github/workflows/ci.yml → 708ba347d355。git diff a7a65e17833 HEAD -- .github/ 为空。

需声明一点:本地直接运行 check-lint-gate-freshness.mjs 虽退出 0,但输出 BASE_REF, HEAD_SHA, GITHUB_REPOSITORY or GITHUB_TOKEN missing; skipping the check,属空过,未作为证据;上述祖先判定才是证据。

合并安全性

git merge-tree 报告无文本冲突。自 merge base 起,main 改动 306 个文件、本 PR 改动 46 个,真实重叠恰好两个:packages/acp-bridge/src/session-control-plane.ts 与 packages/cli/src/serve/hosted-harness-session.test.ts。已在合并后的树中读过两侧,双方意图均存活——main 的 allowChannelSubscribers 守卫与 reaper 调用点、其 it.each retry-refused 重写与两个新 cleanup 用例;本 PR 的附件资源索引元数据处理与 approvalSelections 注入。此处不把「文本合并干净」当作证明:本 PR 先前出现过文本干净但 tsc 断裂的语义合并冲突(发现 L-1),故下方的 typecheck 才是承重证据。

上一轮的 R25-31 修复完好:transcript.ts 与 daemon-embedded-resources.test.ts 与 9493bb3aaa 逐字节相同,end += paired ? 2 : 1 仍在,astral/CJK 用例仍在套件中。

验证

npm run build 退出 0;npm run typecheck 退出 0、零 error TS;17 个 vitest 套件在各包目录内逐文件运行,3420 通过/11 跳过——含两个重叠文件的套件(acp-bridge/bridge.test.ts 1060、cli/hosted-harness-session.test.ts 711)与 R25-31 钉子(daemon-embedded-resources.test.ts 35)。

维护者会话在已推送的合并提交上独立重跑:根 npm run typecheck 退出 0、零 error TS,daemon-embedded-resources.test.ts 35/35。

一处基础设施红已定位而非压制:core/client.test.ts 首跑因 vitest worker 在本机 4144 MB 默认堆下 V8 OOM(非 CI 时配置默认开启 v8 coverage)继而 ERR_IPC_CHANNEL_CLOSED,零用例执行、无断言失败;以仓库自身 build 所用的 6144 MB 重跑后 624 通过。未改动或放宽任何断言。

本次推送会在当前门禁下重新验证该 lane;只有 CI 能给出其绿色,此处不声称 lint lane 已绿。

仍待处理

本 PR 需要 2 个批准。web-shell E2E Smoke lane 将在这个新 head 上重新评判并获得新的按 head 计的重跑预算;先前三条 needs_human 记录针对的是 ce481fe9f9。本值守不 approve、不改评审状态、不合并、不开自动合并。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

@qwen-code-review-bot

Copy link
Copy Markdown
Collaborator

Qwen Code review did not complete successfully. Failed to write qwen review log. A transient error is retried automatically; if you are seeing this, retry with @qwen-code /review. See workflow logs.

@wenshao

wenshao commented Nov 7, 2026

Copy link
Copy Markdown
Collaborator

Watch round — the review-pr failure is the review pipeline being terminated, not a finding

No code changed, nothing pushed. Head remains 4d8eba89e0e.

Answering the review fallback notice. Its stated reason, Failed to write qwen review log, is a consequence, not the cause. The job's own sequence is:

Time (UTC) Event
18:47:18 the review issues its 21-agent fan-out call
18:51:38 Operation cancelled.
18:51:44 Terminated
18:51:44 Failed to write qwen review log.
18:51:49 Process completed with exit code 1

So the review process was cancelled and terminated mid-run, and the log write failed because the process was already gone.

What it was not:

  • Not a job timeout. The job ran 17:39:44 → 18:52:13, about 72 minutes, against a QWEN_REVIEW_JOB_TIMEOUT_MINUTES budget of 480. The workflow's own classifier agrees: FAILURE_KIND was not timeout, which is why the posted notice used the generic branch rather than the "review timed out" one.
  • Not the supersede watcher. That path ends an attempt when the head moves; the head has not moved since this run started.
  • Not a finding against this PR. The review never reached composition; no finding was produced by this run.

What this watch did

It attempted the sanctioned re-run of run 42396891347. The helper deferred it: the run is still in progress, and GitHub only re-runs a run once it has ended. The check is therefore left unrecorded, and the next gate pass reports it again once the run finishes, at which point the re-run can be spent.

The notice suggests retrying by commenting @qwen-code /review. That was deliberately not done: dispatching the review through a comment is outside what this watch may do, and the re-run path above achieves the same thing within its budget.

Elsewhere on this head

web-shell E2E Smoke shards 1/4, 3/4 and 4/4 are green; shard 2/4's re-run is still in flight. The PR still needs 2 approving reviews. This watch does not approve, change review state, merge, or enable auto-merge.

中文说明

值守轮次 —— review-pr 的失败是评审流水线被终止,不是一条发现

未改代码、未推送,head 仍为 4d8eba89e0e。

此处回应评审失败通知。其给出的理由 Failed to write qwen review log 是后果而非根因。job 自身的时序为:

时刻(UTC) 事件
18:47:18 评审发出 21 个 agent 的扇出调用
18:51:38 Operation cancelled.
18:51:44 Terminated
18:51:44 Failed to write qwen review log.
18:51:49 Process completed with exit code 1

即评审进程在运行中被取消并终止,写日志失败是因为进程已不在。

它不是以下情形:

  • 不是 job 超时。 该 job 从 17:39:44 跑到 18:52:13,约 72 分钟,而 QWEN_REVIEW_JOB_TIMEOUT_MINUTES 预算为 480。workflow 自身的分类器也一致:FAILURE_KIND 不是 timeout,故通知走了通用分支而非「评审超时」分支。
  • 不是 supersede watcher。 该机制在 head 移动时提前结束尝试;本 run 启动以来 head 未移动。
  • 不是针对本 PR 的发现。 评审尚未进入撰写阶段,本 run 未产出任何发现。

本值守的处理

已尝试对 run 42396891347 执行规程内的重跑。helper 将其延后:该 run 仍在进行中,而 GitHub 只能重跑已结束的 run。因此该检查未予记录,门禁会在其收尾后再次上报,届时可用掉这次重跑。

通知建议以评论 @qwen-code /review 重试。此处有意未照做:通过评论派发评审超出本值守的权限范围,而上述重跑路径在其预算内达成同样效果。

本 head 的其他情况

web-shell E2E Smoke 的 shard 1/4、3/4、4/4 已绿;shard 2/4 的重跑仍在飞。本 PR 仍需要 2 个批准。本值守不 approve、不改评审状态、不合并、不开自动合并。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

@wenshao

wenshao commented Nov 7, 2026

Copy link
Copy Markdown
Collaborator

Watch round — web-shell E2E Smoke budget now spent on a second head; this needs a person

No code changed, nothing pushed. Head remains 4d8eba89e0e. The re-run spent on run 42396893001 has finished, and the helper now reports "was re-run on this head already; treat the failure as real", so shard 2/4 and the summary job are recorded as needs_human. This is the second head on which this lane has exhausted the watch's budget.

Where the lane stands on this head

Shard First run After the re-run
1/4 success success
2/4 18 failed, 27 flaky, 67 passed (29.7m) 26 failed, 24 flaky, 62 passed (35.0m)
3/4 success success
4/4 success success

The re-run made shard 2/4 worse, not better, and it landed on a different runner (hz1-3, previously hz1-28).

The evidence, and one pattern worth acting on

The failure signature is the E2E connection handshake, not an assertion about this PR's behaviour. In the re-run's full job log: 135 waitForConnection hits, 128 references to utils/sseTransport.ts:176, 64 TimeoutError: page.waitForFunction: Timeout 20000ms exceeded, 171 connect ECONNREFUSED 127.0.0.1:4170, and 46 element(s) not found from pages that never mounted. Line 176 is the body of async waitForConnection(sessionId, { timeout = 20_000 }), and the comment directly above it in the repository reads: "handshake loses the same race against first paint on a busy pool host."

The pattern: in this very run, shard 2/4 took 36.8m of wall clock against 13.9m (1/4), 18.2m (3/4) and 10.3m (4/4) — 2 to 3.6 times the others — and across its recent runs its Playwright time has been 32.4m, 32.6m, 29.7m and 35.0m. It is also the one that keeps failing, on two different heads and four runs, while 1/4, 3/4 and 4/4 have all gone green. A longer shard spends more wall clock exposed to the busy-host handshake race, which would explain both the persistence and why the counts move with host load (18 failures on hz1-28, 26 on hz1-3). Offered as a hypothesis consistent with the measurements, not as a proven cause.

This PR still changes zero packages/web-shell files, and an unrelated PR (feat/core-pr-view) showed the same signatures in the same window.

What a person has to decide

The limit that stopped this round is this watch's own per-head budget, not GitHub's — a person can re-run the lane freely. The options are:

  1. Relieve the environment and re-dispatch — the daemon backend refusing connections on port 4170 and the busy-pool handshake race are the measured cause; re-running onto a quieter host is the cheapest test of that.
  2. Raise the E2E connection handshake budget or split shard 2/4 — if the hypothesis above holds, the 20 s waitForConnection timeout and the shard's runtime are the levers. That is a change to the E2E harness, outside this PR's scope, so it was not made here.
  3. Treat it as real and investigate shard 2/4 — its failures concentrate in web-shell.files.spec.ts and web-shell.managed-progress.spec.ts, both of which need a live daemon connection.

This watch will not spend more re-runs on 4d8eba89e0e, and it does not approve, change review state, merge, or enable auto-merge. Separately, the review-pr lane's own re-run is still deferred because its run has not ended, and the PR needs 2 approving reviews.

中文说明

值守轮次 —— web-shell E2E Smoke 在第二个 head 上耗尽预算,需要人工

未改代码、未推送,head 仍为 4d8eba89e0e。run 42396893001 上用掉的重跑已收尾,helper 现报「was re-run on this head already; treat the failure as real」,故 shard 2/4 与汇总 job 记为 needs_human。这是该 lane 在第二个 head 上耗尽本值守预算。

本 head 上该 lane 的现状

Shard 首次 重跑后
1/4 success success
2/4 18 failed、27 flaky、67 passed(29.7m) 26 failed、24 flaky、62 passed(35.0m)
3/4 success success
4/4 success success

重跑使 shard 2/4 更差而非更好,且落在不同 runner(hz1-3,此前为 hz1-28)。

证据,以及一个值得据以行动的规律

失败签名是 E2E 连接握手,而非关于本 PR 行为的断言。重跑的完整 job 日志中:waitForConnection 135 次、utils/sseTransport.ts:176 出现 128 次、TimeoutError: page.waitForFunction: Timeout 20000ms exceeded 64 次、connect ECONNREFUSED 127.0.0.1:4170 171 次、页面未挂载导致的 element(s) not found 46 次。第 176 行正是 async waitForConnection(sessionId, { timeout = 20_000 }) 的函数体,而仓库中其上方注释写着:「handshake loses the same race against first paint on a busy pool host」。

规律: 就在本次 run 内,shard 2/4 的墙钟为 36.8m,而 1/4 为 13.9m、3/4 为 18.2m、4/4 为 10.3m——是其余的 2 至 3.6 倍;其近期各次的 Playwright 时长为 32.4m、32.6m、29.7m、35.0m。它也正是持续失败的那个,跨两个 head、四次运行,同时 1/4、3/4、4/4 均已转绿。更长的 shard 在墙钟上暴露于繁忙主机握手竞态的时间更多,这既解释了持续性,也解释了计数随主机负载波动(hz1-28 上 18 条失败,hz1-3 上 26 条)。此为与测量一致的假设,非已证成因。

本 PR 仍改动 零 个 packages/web-shell 文件;无关 PR feat/core-pr-view 在同一时间窗呈现相同签名。

需要人工决定的内容

拦住本轮的上限是本值守自己的按 head 预算,不是 GitHub 的限制——人工可自由重跑。可选项为:

  1. 缓解环境后重新派发——已测得的成因是 daemon 后端在 4170 端口拒连与繁忙池上的握手竞态;换到较空闲主机重跑是检验该点最便宜的方式。
  2. 提高 E2E 连接握手预算或拆分 shard 2/4——若上述假设成立,waitForConnection 的 20 秒超时与该 shard 的运行时长就是杠杆。这属于 E2E 测试框架的改动,超出本 PR 范围,故本轮未做。
  3. 当作真实缺陷排查 shard 2/4——其失败集中在 web-shell.files.spec.ts 与 web-shell.managed-progress.spec.ts,两者都需要活的 daemon 连接。

本值守不会在 4d8eba89e0e 上再花重跑,也不 approve、不改评审状态、不合并、不开自动合并。另外,review-pr lane 自身的重跑仍因其 run 未收尾而延后;本 PR 需要 2 个批准。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

@qwen-code-review-bot qwen-code-review-bot left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

[Critical] Blocking finding(s) follow.

Partially reviewed — gaps disclosed.

Unresolved, please confirm:

  • [Critical] packages/core/src/services/session-transcript-reader.ts:3578 — existing Critical R21-1 (cold restore now reads every user turn record of the whole session to keep the title arrays index-aligned): the read set is confirmed at HEAD and docume…

Not reviewed: reverse audit — closed after audit round 2 without two consecutive dry rounds: rounds 1 and 2 each reported new findings, so the loop ended unconverged rather than dry.

Not reviewed: reverse audit round 2 candidates — 16 further candidates in 14 chunks, recorded in the run report, not verified this round.

Deferred under the convergence posture (round 26, not a blocker) — recorded, not requested in this round:

  • packages/acp-bridge/src/embedded-resource-replay.test.ts:38 — [review] The suite's only copy-isolation assertion is top-level…
  • packages/acp-bridge/src/embedded-resource-replay.test.ts:183 — [review] The exemption test pins only the "not retained" half of…
  • packages/acp-bridge/src/embedded-resource-replay.test.ts:193 — [review] The only test of readDaemonAttachmentReferences …
  • packages/cli/src/acp-integration/session/Session.test.ts:17586 — [review] The test named does not retain text resources that exceed…
  • packages/cli/src/ui/commands/renameCommand.ts:191 — [review] The new ternary is dead code, and its comment asserts a…
  • packages/core/src/core/client.ts:4446 — [review] The added displayProjection: true marker is…
  • packages/core/src/services/session-transcript-reader.ts:370 — [review] The title-projection policy is now implemented twice —…
  • packages/core/src/services/sessionService.corruption.test.ts:455 — [review] The three-part test asserts the same lower bound as the…
  • packages/core/src/services/sessionService.corruption.test.ts:494 — [review] The new regression test sizes its corrupt-tail fixture…
  • packages/sdk-typescript/test/unit/daemon-embedded-resources.test.ts:951 — [review] This is the only test in the file that reaches the stored…

Mechanism health: this round did not close cleanly, so it withholds the incremental anchor — and the round it recovered had no anchor this round could use either — none at all, one with no certifier, one certified by an identity other than the one this round runs under, or one this round's fetch refused or resolved to the head — so the next review re-reads the whole diff unless recovery grafts an earlier own anchor that the round running it can use onto the complete work list this round leaves behind, and keeps doing so until a round's marker carries an anchor again or a graft lands that the round running it can use. (Stated, not acted on — this changes nothing about what the round posts.)

中文说明

仅完成部分审查,审查缺口已披露。

未决,请确认:共 1 条(原文未翻译,列表见上方英文部分)。

未审查(原文为英文):reverse audit — closed after audit round 2 without two consecutive dry rounds: rounds 1 and 2 each reported new findings, so the loop ended unconverged rather than dry.

未审查(原文为英文):reverse audit round 2 candidates — 16 further candidates in 14 chunks, recorded in the run report, not verified this round.

收敛姿态下延后(第 26 轮,非阻断)——已记录,本轮不要求修改:共 10 条(原文未翻译,列表见上方英文部分)。

机制健康:本轮未能干净收尾,因而扣留了增量锚点,而它恢复到的那一轮也没有留下本轮可用的锚点——要么完全没有、要么没有认证者、要么由本轮运行身份之外的身份认证、要么被本轮的获取拒绝或解析为头提交——因此下一次评审将重读整个 diff,除非恢复流程把本轮能使用的更早自有锚点嫁接到本轮留下的完整工作清单上;并会一直如此,直到某一轮的标记重新带上锚点,或落地的嫁接能被运行该轮的评审使用。(仅陈述,不据此行动——这不改变本轮发布的任何内容。)

— deepseek-v4.1-flash-1017 via Qwen Code /review (v0.26.9)

@wenshao

wenshao commented Nov 8, 2026

Copy link
Copy Markdown
Collaborator

Watch round — confirming R21-1 at HEAD: the retention half is closed, the read set stands by design

No code changed, nothing pushed. Head remains 4d8eba89e0e.

Round 26 asks to confirm one carried-forward Critical, R21-1 at packages/core/src/services/session-transcript-reader.ts. Its own ledger reads findings: [], posted: 0, fresh: 0, prevPosted: 1 — nothing new was posted — and the thread it comes from (PRRT_kwDOPB-92c6wg2h4) is resolved and outdated on GitHub. All 81 review threads are resolved. Confirming it against the code rather than against either status flag:

R21-1 had two halves. They now have different answers.

The retention half is closed

The finding's acute complaint was that cold restore retained full ChatRecord objects, inline base64 media included, in a titleProjectionRecords array living until deriveTitleProjectionState, just to derive two scalar fields. That array no longer exists at this head. In its place:

  • TitleProjectionAccumulator (session-transcript-reader.ts:351) holds only scalars — a Array<string | null | undefined>, a Array<{ projection: boolean; displayText: string | undefined }>, one boolean and one string. It retains no record.
  • dispatchRecord feeds it one record at a time (titleProjection.add(record) at :3434, under the user-turn guard at :3433); add reads what it needs off the record and keeps nothing else.

So the heap-pinning mechanism the finding named is gone, and the class comment states the reason in the same terms.

The read set does still stand, and it is load-bearing

Confirmed as the review says: selectedRuntimeUuids still admits every turn's user record via index.byUuid.get(uuid)?.turnHint.turnParentUuid !== undefined (:3590), so cold restore reads user turn records that a compression checkpoint removed from the model set.

That clause is deliberate, and the comment directly above it says why: "Title-projection state is index-aligned with every turn boundary, so each turn's user record must be read even when a compression checkpoint removed it from the model set." turnTitleStates is pruned by turn index on rewind and must stay aligned with turnParentUuids; dropping the clause would restore a short array, which is the misalignment the length-equality guard was added to reject. Removing it to save reads would reintroduce a correctness bug, not fix one.

Disposition

Declined for this round, with the read set confirmed as designed. The finding's axis is fails-closed — it certifies no wrong result; what remains is read amplification on cold restore, which is the cost of the alignment invariant this feature needs. The reviewer requested confirmation, not a specific change, and posted no new finding.

A follow-up worth recording rather than doing here: persist the per-turn title scalars in the journal so cold restore can rebuild the aligned arrays without reading each turn's payload. That removes the read amplification without touching the invariant, but it is a record-format change and belongs with the residual work in #13282, not in this PR.

The 10 items round 26 lists under the convergence posture are marked "not a blocker … not requested in this round" and were not reopened. The disclosed reverse-audit gaps (closed unconverged after round 2; 16 further candidates unverified) and the anchor-withholding paragraph are statements about the review pipeline, not requests against this code.

Still outstanding for a person

Unchanged: the web-shell E2E Smoke lane needs a decision on this head (the watch's re-run budget is spent; a person may re-run freely), and the PR needs 2 approving reviews. This watch does not approve, change review state, merge, or enable auto-merge.

中文说明

值守轮次 —— 在 HEAD 上确认 R21-1:保留那半已闭,读取集按设计保留

未改代码、未推送,head 仍为 4d8eba89e0e。

第 26 轮要求确认一条结转的 Critical,即 packages/core/src/services/session-transcript-reader.ts 的 R21-1。其自带台账为 findings: []、posted: 0、fresh: 0、prevPosted: 1——未发布任何新发现——且其来源线程 PRRT_kwDOPB-92c6wg2h4 在 GitHub 上为 resolved 且 outdated;全部 81 条评审线程均已 resolved。以下以代码而非任一状态标记为准进行确认。

R21-1 包含两半,如今答案不同。

保留那半已闭

该发现的尖锐指控是:冷启动为推导两个标量字段,把**完整 ChatRecord 对象(含内联 base64 媒体)**保留在 titleProjectionRecords 数组中,直到 deriveTitleProjectionState 才释放。该数组在本 head 上已不存在,取而代之的是:

  • TitleProjectionAccumulator(session-transcript-reader.ts:351)只持有标量——一个 Array<string | null | undefined>、一个 Array<{ projection: boolean; displayText: string | undefined }>、一个布尔与一个字符串,不保留任何记录。
  • dispatchRecord 逐条喂入(:3434 的 titleProjection.add(record),其上 :3433 为 user turn 守卫);add 从记录上取所需值,其余一概不留。

因此该发现点名的堆占用机制已消失,类注释也以同样措辞说明了原因。

读取集确实仍在,且是承重的

如评审所述确认:selectedRuntimeUuids 仍通过 index.byUuid.get(uuid)?.turnHint.turnParentUuid !== undefined(:3590)纳入每个 turn 的 user 记录,故冷启动会读取被压缩检查点移出模型集的 user turn 记录。

该条款是刻意的,其上方注释说明了原因:「标题投影状态与每个 turn 边界索引对齐,故即使压缩检查点把该 turn 移出模型集,其 user 记录仍必须读取。」 turnTitleStates 在 rewind 时按 turn 索引裁剪,必须与 turnParentUuids 保持对齐;去掉该条款会恢复出一个短数组,而这正是长度相等守卫所要拒绝的错位。为省读取而移除它会重新引入正确性缺陷,而非修复缺陷。

处置

本轮不予改动,并确认读取集属设计如此。 该发现的轴为 fails-closed——不认证任何错误结果;剩下的是冷启动的读放大,而这是该功能所需的对齐不变式的代价。评审要求的是确认而非某项具体改动,且未发布新发现。

值得记录而非在此实施的后续:把每 turn 的标题标量持久化进 journal,使冷启动无需读取各 turn 载荷即可重建对齐数组。这能在不触碰不变式的前提下消除读放大,但属记录格式变更,应与 #13282 的遗留工作一并处理,不放在本 PR。

第 26 轮在收敛姿态下列出的 10 条均标注「非阻断……本轮不要求修改」,未重开。其披露的反向审计缺口(第 2 轮后未收敛即关闭;另有 16 条候选未验证)与扣留锚点段落,都是关于评审流水线的陈述,不是针对本代码的请求。

仍待人工

不变:web-shell E2E Smoke lane 在本 head 上需要决定(本值守重跑预算已用完,人工可自由重跑),且本 PR 需要 2 个批准。本值守不 approve、不改评审状态、不合并、不开自动合并。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

@wenshao

wenshao commented Nov 8, 2026 •

Copy link
Copy Markdown
Collaborator

Verdict: merge-ready — 37/37 scripted assertions passed. Verified head 4d8eba89e0e43bf0ad38830738062cd6a74a1492 (the head has not moved since the last watch round; base merge-base a7a65e1783, current main e4e28d7ee0). Round 8 (follow-up): the central claim still holds at this head and on a fresh trial merge into today's main; no new blocking finding.

中文摘要

结论:merge-ready(37/37 脚本断言通过)。 被验证 head:4d8eba89e0e(与上一值守轮相同,未移动)。

  • A/B(本机、真实模块、三臂):把「已持久化的用户记录(含受限嵌入式文本资源)」喂给真实的 transcript 回放机,再经真实的 daemon UI SDK 归一化与 transcript store:head 8/8 通过;trial-merge 臂 8/8 通过;base 只通过 2 格(负向对照与「无资源」对照),其余 6 格失败——base 上资源在回放中完全消失,纯资源轮次在刷新后连用户块都不出现。见下表与证据图 01-ab-replay-base-merge-head.png。
  • 变异矩阵:8 行全杀(含 1 行阳性对照)。记录侧(payload 丢弃资源 / 丢弃截断标记)、保留侧(256 上限、native 位置豁免、旧元数据同 URI 歧义、qwen.daemon.* 元数据剥离)、回放侧(截断占位)、SDK 侧(user.resource.delta 归一化)各自都有点名失败的用例。见 02-mutation-matrix.png。
  • 定向门禁:head acp-bridge 195/195、core 527/527、SDK 35/35;trial-merge core 393/393、acp-bridge 195/195、SDK 35/35。
  • trial merge:4d8eba89 合入今日 main e4e28d7e 无冲突,合并树上 A/B 与定向套件全绿。
  • findings:无新增阻断项。结转项见下表:C-1 已修且仍成立;C-2 在 head/merge 上复测为「有文字时轮次可见、占位 0」;R21-1 由值守轮在本 head 以代码确认(保留那半已闭、读取集属设计),本轮未做行为复测。
  • 未覆盖:真实 daemon HTTP session/load 端到端、>256 KiB 原生附件经 daemon 的端到端、浏览器刷新验收、内联 blob 资源持久化、Windows/Linux 本地运行、R21-1 的 RSS 量化、逐提交归因。

Previous-finding status (follow-up protocol)

This round re-measures the carried-forward items at 4d8eba89. Round 7 (merge-ready, 214/214) verified ed5fb4339; the head has since moved only by merges of main plus the review-round fixes recorded in the watch rounds.

# Finding Severity Status at this head
C-1 = L-2 PR body describes a rejection path the code does not have suggestion FIXED, stands: body still says admitted + live + retention-only drop; measured by ab3 (marker, no rejection)
C-2 Partial truncation drops user content with no client-visible trace suggestion STANDS, narrowed, re-measured: text+oversized prompt replays with placeholderChunks=0 and the text visible (head and merge); truncation-only prompt emits the marker (ab3)
R2-#4 Intra-turn block order not preserved on replay note STANDS, not re-measured this round (documented accepted shape; no code change in the replay ordering)
R21-1 Cold restore retains per-turn displayText/cloned apiHistory; read set amplification note retention half CLOSED, read set by design: TitleProjectionAccumulator holds scalars only; the user-record read set is required for turn alignment (verified by code trace at this head, last watch round; no RSS measurement this round)
M12 Incremental accumulator has no committed test note not re-run: unchanged coverage observation, not a defect
L-3 Base let any ACP caller inject forged attachmentReferences security closed: acpAgent.ts re-injects both provenance keys only for trusted parents; unchanged since round 7
C-4 / R2-#3 / M0–M3 Count cap, URI encoding, replay liveness, byte budget suggestion re-measured: mutation.* rows for the 256 cap, native exemption, legacy ambiguity and qwen.daemon.* strip all killed

Central claim and A/B

Central claim: a bounded ACP embedded text resource, recorded with its owning user prompt, is replayed as a typed user chunk on that same prompt (acp-bridge replay → daemon UI SDK transcript), without URI fetching and without migrating to a later turn; an oversized direct text resource is admitted live and dropped only from durable retention, with embeddedResourcesTruncated.

The harness (ab-replay.mts) drives the real createTranscriptReplayMachine, the real normalizeDaemonEvent, and the real createDaemonTranscriptStore of the arm it runs in, with the same record payloads on every arm. It uses no API that exists only on the head, so base runs it unchanged.

Cell Oracle base a7a65e17 trial-merge b893bb99 head 4d8eba89
ab1 replay: resource chunk count user_message_chunk with content.type=resource 0 1 1
ab1 replay: uri/text preserved resource.uri / resource.text equal the recorded block none equal equal
ab1 replay: no URI fetch no absolutePath/fetch artefact in the chunk pass pass pass
ab1 SDK: typed user resource block transcript user block embeddedResources.length 0 1 1
ab2 SDK: resource-only turn visible user blocks after reload 1 (turn lost) 2 2
ab2 SDK: not migrated to the next turn embeddedResources on the later text-only block none none none
ab3 replay: truncation marker visible replay text for a truncation-only record 0 updates marker marker
control: no resource ⇒ no chunk resource chunks for a plain prompt 0 0 0

base fails 6 of 8 cells; the two it passes are the negative control (no URI fetch) and the absence control. That is the load-bearing proof: the same persisted record produces a typed resource on head/merge and nothing on base — and base loses the resource-only turn entirely (1 user block instead of 2).

A/B arms

Corrections

None this round. Round 7's correction (the PR body's rejection wording) still matches the shipped behaviour by measurement.

Findings

No blocking finding. Three observations for the record:

  1. The web-shell E2E Smoke lane is red on this head (shard 2/4 failed after 36m48s, and the aggregate job exits 1 in 9s). The PR changes no packages/web-shell/** file; every other lane on this head is green, including all eight Linux unit-test shards and Integration Tests. This is the same undecided lane the watch rounds recorded; it needs a human re-run/decision, not a code change from this PR.
  2. Truncation of a direct resource is silent whenever the prompt also has visible text (placeholderChunks=0, text visible). This is the accepted C-2 shape and the body documents the retention-only drop, so it is a tradeoff rather than a defect — recorded because it is the only case where a dropped resource leaves no trace.
  3. The 256 KiB budget is charged in serialized-block bytes (Buffer.byteLength(JSON.stringify(block))), so the durable payload can exceed 256 KiB of text (JSON escaping of a hostile uri/meta counts too). The bound is on the stored block, which is what the SDK's own estimateDaemonTranscriptBlockBytes measures; noted only so a future reader does not assume "256 KiB of text".

Not covered

  • Real daemon HTTP session/load after a restart: the round drives the real replay machine and SDK projection in-process, not POST /session/:id/load over HTTP. The PR's own body lists this as unvalidated.
  • > 256 KiB native attachment through a live daemon and browser refresh acceptance: not run (no daemon/browser rig this round).
  • Inline blob resources: persistence explicitly out of scope in the body; not probed.
  • Windows/Linux local runs: macOS host only. CI covers Linux unit tests on this head.
  • R21-1 RSS quantification: read-amplification cost not measured; only the retention mechanism was inspected.
  • Head CLI Session.test.ts as a whole suite: the run was still in flight at the end of the round and was stopped; the four embedded-resource cases in it were exercised by the mutation rows above (all four killed the payload mutant), and CI's eight Linux unit-test shards are green on this head.
  • Per-commit attribution: production commits since round 7 were not mapped one by one; the matrix is per-guard, covering the guards the recent commits touch.
  • Base-side full suites: only the targeted files were run at base; the base arm's red cells are the A/B, not a suite comparison.

Methodology

Host: macOS, Node 24.18.1, pnpm 11.24.0. Three detached worktrees under tmp/pr12585-verify-20261108-183437/ (base = merge-base a7a65e17, head = 4d8eba89, merge = local merge of the head into origin/main e4e28d7e, b893bb99). Each arm ran corepack pnpm install --frozen-lockfile --ignore-scripts and built core/acp-bridge/sdk-typescript from its own sources; workspace symlinks were confirmed to resolve into each arm (readlink -f node_modules/@qwen-code/qwen-code-core), so no arm borrowed another arm's code. Builds of web-templates/channels and packages/cli/src/generated/git-commit.ts were copied from the host checkout because the PR does not touch those paths and the fresh-worktree build fails on the generated-UI step for reasons unrelated to this change. The A/B harness runs npx tsx ab-replay.mts from each arm's packages/sdk-typescript; raw output is ab-runs.log, the C-2 probe c2-partial-truncation.log. The mutation matrix (mutations/, driver run-mutations.sh) mutates one production anchor at a time in the head tree, asserts the anchor matched exactly once, runs the named suite, and restores with git checkout HEAD --; the control row mutates the 256 constant and is killed by two tests, proving the runner can fail. Every number in this report is parsed from those logs by build-assertions.mjs, which writes assertions.json.

Evidence

…pr-12585

Resolve packages/core/src/services/chatRecordingService.ts so both sides
survive. Main (#14654) invalidates the cached last-user-prompt time after a
message-group deletion; keep that notification. Main also re-derives the
title texts and Mod keys in an explicit loop after rebuildTurnBoundaries,
because on main that method only rebuilds the boundaries. This PR moved the
re-derivation into rebuildTurnBoundaries itself (clearing the ring, the
per-turn states and the Mod keys first), so keeping main's loop would track
every retained prompt a second time. Drop the loop, keep the notification.
@wenshao

wenshao commented Nov 8, 2026

Copy link
Copy Markdown
Collaborator

Watch round — conflict with main resolved

Head 4d8eba89e0e → 6940d16460 (merge of main 25e69c0f695, 72 commits). Push 4 of this watch's 10.

The conflict and how it was resolved

git merge-tree named exactly one conflicting file, packages/core/src/services/chatRecordingService.ts, in deleteMessageGroups. The two sides disagreed for a structural reason rather than a textual one:

  • main (feat(daemon): expose last user prompt time and session sorting #14654, feat(daemon): expose last user prompt time and session sorting) added notifyLastUserPromptRecorded(undefined, this.getSessionId()) after a message-group deletion, and still carried the merge-base's explicit title re-derivation loop — because main's rebuildTurnBoundaries only rebuilds boundaries, it does not track titles.
  • this PR had already deleted that loop, because it moved the re-derivation into rebuildTurnBoundaries, which clears the ring, the per-turn states, the projection flag and the Mod keys and then calls trackTitleText per turn.

Keeping both would have tracked every retained prompt twice. The resolution keeps main's notification and drops the duplicated loop. Verified in the merged tree: rebuildTurnBoundaries still clears and re-tracks (:4143), and deleteMessageGroups calls it and then main's notification (:2520–2521) with no second loop. Both features survive; neither is double-counted.

Scope of the merge

Since the merge base main changed 504 files and this PR 46, with a real overlap of 17 files — so 16 auto-merged textually and still needed checking. This PR has twice been bitten by a textually clean merge that broke tsc (finding L-1), so the typecheck below is the load-bearing evidence, not the absence of conflict markers.

Discriminative markers are unchanged (44 matching lines / 45 occurrences before and after): trackTitleText, turnTitleStates, displayProjectionForTitle, modPromptKeysForTitle, the Mod-prompt invariant (!isModPrompt &&, isModPrompt && !payload?.asUser) and its mirror in TitleProjectionAccumulator, and the R25-31 fix (end += paired ? 2 : 1 plus the astral/CJK case). No gitlink change; the commit's diff against main is exactly this PR's 46 files.

Verification

npm run build exit 0; npm run typecheck exit 0 with zero error TS; prettier and eslint clean on the staged set. Thirteen vitest suites run one file at a time inside each package, 5745 passing, including main's own new wiring witness control-plane-last-user-prompt.test.ts. No assertion was retargeted and no test was weakened.

Independently re-run in the maintainer session on the pushed merge commit: root npm run typecheck exit 0, zero error TS; chatRecordingService.test.ts 203; control-plane-last-user-prompt.test.ts 2 — the check that main's last-prompt-time feature still works after the resolution; and daemon-embedded-resources.test.ts 35, the R25-31 pin.

Outstanding

This push starts a fresh CI evaluation on 6940d16460, including a fresh per-head re-run budget for web-shell E2E Smoke. The five earlier needs_human entries were recorded against ce481fe9f9 and 4d8eba89e0. The PR still needs 2 approving reviews; this push dismisses any approval given on the previous head under main's stale-review rule. This watch does not approve, change review state, merge, or enable auto-merge.

中文说明

值守轮次 —— 与 main 的冲突已解决

Head 4d8eba89e0e → 6940d16460(合并 main 25e69c0f695,72 个提交)。本值守第 4 次推送(上限 10)。

冲突及其解决

git merge-tree 指出唯一冲突文件为 packages/core/src/services/chatRecordingService.ts,位置在 deleteMessageGroups。两侧分歧是结构性的而非文字性的:

  • main(feat(daemon): expose last user prompt time and session sorting #14654,feat(daemon): expose last user prompt time and session sorting)在消息组删除后新增了 notifyLastUserPromptRecorded(undefined, this.getSessionId()),并仍带着 merge-base 那段显式标题重推导循环——因为 main 的 rebuildTurnBoundaries 只重建边界、不 track 标题。
  • 本 PR 已删除该循环,因为它把重推导移进了 rebuildTurnBoundaries:后者先清空 ring、per-turn states、投影标志与 Mod 键,再逐 turn 调用 trackTitleText。

两者都留会把每条留存 prompt 重复计入一次。解决方式为保留 main 的通知、丢弃重复循环。已在合并后的树中核实:rebuildTurnBoundaries(:4143)仍先清空再重推;deleteMessageGroups(:2520–2521)先调它、再调 main 的通知,无第二个循环。两个特性均存活,且无双重记账。

合并范围

自 merge base 起,main 改动 504 个文件、本 PR 改动 46 个,真实重叠 17 个——即其中 16 个文本自动合并但仍需检查。本 PR 已两次栽在「文本干净但 tsc 断裂」的合并上(发现 L-1),故下方的 typecheck 才是承重证据,而非「没有冲突标记」。

判别性标记合并前后不变(44 行匹配/45 次出现):trackTitleText、turnTitleStates、displayProjectionForTitle、modPromptKeysForTitle,Mod-prompt 不变式(!isModPrompt &&、isModPrompt && !payload?.asUser)及其在 TitleProjectionAccumulator 中的镜像,以及 R25-31 修复(end += paired ? 2 : 1 与 astral/CJK 用例)。无 gitlink 变更;提交相对 main 的差异恰好是本 PR 的 46 个文件。

验证

npm run build 退出 0;npm run typecheck 退出 0、零 error TS;暂存集 prettier 与 eslint 均干净。13 个 vitest 套件在各包目录内逐文件运行,5745 通过,其中包含 main 自己新增的接线见证 control-plane-last-user-prompt.test.ts。未改靶任何断言,未削弱任何测试。

维护者会话在已推送的合并提交上独立重跑:根 npm run typecheck 退出 0、零 error TS;chatRecordingService.test.ts 203;control-plane-last-user-prompt.test.ts 2——这一项正是验证解决冲突后 main 的 last-prompt-time 特性仍然工作;以及 daemon-embedded-resources.test.ts 35,即 R25-31 的钉子。

仍待处理

本次推送会在 6940d16460 上启动全新 CI 评判,web-shell E2E Smoke 也随之获得新的按 head 计的重跑预算。先前五条 needs_human 记录针对的是 ce481fe9f9 与 4d8eba89e0。本 PR 仍需要 2 个批准;按 main 的陈旧评审规则,本次推送会作废在上一 head 上给出的任何批准。本值守不 approve、不改评审状态、不合并、不开自动合并。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

@qwen-code-review-bot qwen-code-review-bot left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Partially reviewed — gaps disclosed.

Not reviewed: three low-confidence Critical candidates (title-projection window units; web-shell userBlockBoundaryKey not folding embeddedResources; SideTaskPanel hasUserInteraction not counting embeddedResources) — the mechanism of each was traced to exact lines but none could be executed: qwen review scratch-tree returned available:false because the repo's local git config carries an includeif content-filter the command cannot screen, so no probe or mutation ran and the triggers rest on reads alone.

Not reviewed: build-and-test — Integration Tests (CLI, No Sandbox) was skipped in CI at this head and its suite did not run locally, so the daemon/session replay path was exercised only by package unit tests.

Not reviewed: build-and-test — Test (macos-latest, Node 22.x), Test (windows-latest, Node 22.x) and macOS unit tests were skipped in CI at this head and their suites did not run locally; the PR's own Tested-on table also records Windows and Linux as not tested by the author.

Not explored to full depth (tool budget reached): chunk 6: full-file Session.test.ts run (1451 tests) hit the 600 s wall-clock timeout before printing results, so I could not confirm the file as a whole; the seven add….

Deferred under the convergence posture (round 27, not a blocker) — recorded, not requested in this round:

  • packages/acp-bridge/src/embedded-resource-replay.test.ts:39 — [review] The only independence assertion is top-level identity, so…
  • packages/acp-bridge/src/embedded-resource-replay.test.ts:130 — [review] Every budget case is dominated by resource.text (two 200…
  • packages/acp-bridge/src/embedded-resource-replay.test.ts:134 — [review] Every retention fixture in this file is pure ASCII and…
  • packages/acp-bridge/src/embedded-resource-replay.test.ts:149 — [review] The implementation deletes *every* key starting with…
  • packages/acp-bridge/src/embedded-resource-replay.test.ts:183 — [review] This is the only test of native-position exclusion and its…
  • packages/acp-bridge/src/embedded-resource-replay.test.ts:193 — [review] The new public validator is 40 lines of per-item shape…
  • packages/acp-bridge/src/session-control-plane.ts:11804 — [review] provenance-by-position is only tested native-first, so the…
  • packages/acp-bridge/src/transcript-replay.ts:1513 — [review] The truncation placeholder's guard only counts projections…
  • packages/cli/src/acp-integration/session/Session.test.ts:17587 — [review] Both new "empty projection" tests pair the empty…
  • packages/cli/src/acp-integration/session/Session.test.ts:17684 — [review] The test named "does not retain text resources that exceed…
  • packages/cli/src/acp-integration/session/Session.test.ts:38063 — [review] All three added deferred- /advisor tests (and the…
  • packages/cli/src/ui/commands/renameCommand.test.ts:147 — [review] The only test that exercises the recorder's projection…
  • packages/cli/src/ui/commands/renameCommand.ts:191 — [review] The whole point of this hunk — that the recorder's…
  • packages/core/src/services/chat-recording-message-deletion.test.ts:175 — [review] Both new tests (this rewind one and the delete one at…
  • packages/core/src/services/chatRecordingService.autoTitle.test.ts:750 — [review] The new tests named after the title projection *marker*…
  • packages/core/src/services/session-transcript-reader.ts:373 — [review] TitleProjectionAccumulator.add duplicates…
  • packages/core/src/services/session-transcript-reader.ts:523 — [review] Three of the four title-restore fields the reader…
  • packages/core/src/services/session-transcript-reader.ts:1415 — [review] navigationKindForRecord enumerates embeddedResources/embedd…
  • packages/core/src/services/session-writer-lease.test.ts:3427 — [review] The test's stated invariant ("the flag must follow the…
  • packages/core/src/services/sessionService.corruption.test.ts:488 — [review] The test and its three-line comment assert a pre-fix…
  • …and 11 more (see the run report)

Mechanism health: this round did not close cleanly, so it withholds the incremental anchor — and the round it recovered had no anchor this round could use either — none at all, one with no certifier, one certified by an identity other than the one this round runs under, or one this round's fetch refused or resolved to the head — so the next review re-reads the whole diff unless recovery grafts an earlier own anchor that the round running it can use onto the complete work list this round leaves behind, and keeps doing so until a round's marker carries an anchor again or a graft lands that the round running it can use. (Stated, not acted on — this changes nothing about what the round posts.)

中文说明

仅完成部分审查,审查缺口已披露。

未审查(原文为英文):three low-confidence Critical candidates (title-projection window units; web-shell userBlockBoundaryKey not folding embeddedResources; SideTaskPanel hasUserInteraction not counting embeddedResources) — the mechanism of each was traced to exact lines but none could be executed: qwen review scratch-tree returned available:false because the repo's local git config carries an includeif content-filter the command cannot screen, so no probe or mutation ran and the triggers rest on reads alone.

未审查(原文为英文):build-and-test — Integration Tests (CLI, No Sandbox) was skipped in CI at this head and its suite did not run locally, so the daemon/session replay path was exercised only by package unit tests.

未审查(原文为英文):build-and-test — Test (macos-latest, Node 22.x), Test (windows-latest, Node 22.x) and macOS unit tests were skipped in CI at this head and their suites did not run locally; the PR's own Tested-on table also records Windows and Linux as not tested by the author.

未探索到全部深度(达到工具调用预算):chunk 6:full-file Session.test.ts run (1451 tests) hit the 600 s wall-clock timeout before printing results, so I could not confirm the file as a whole; the seven add…。

收敛姿态下延后(第 27 轮,非阻断)——已记录,本轮不要求修改:共 31 条(原文未翻译,列表见上方英文部分)。

机制健康:本轮未能干净收尾,因而扣留了增量锚点,而它恢复到的那一轮也没有留下本轮可用的锚点——要么完全没有、要么没有认证者、要么由本轮运行身份之外的身份认证、要么被本轮的获取拒绝或解析为头提交——因此下一次评审将重读整个 diff,除非恢复流程把本轮能使用的更早自有锚点嫁接到本轮留下的完整工作清单上;并会一直如此,直到某一轮的标记重新带上锚点,或落地的嫁接能被运行该轮的评审使用。(仅陈述,不据此行动——这不改变本轮发布的任何内容。)

— deepseek-v4.1-flash-1017 via Qwen Code /review (v0.27.0)

@wenshao

wenshao commented Nov 8, 2026

Copy link
Copy Markdown
Collaborator

Watch round — review round 27 dispositioned; two of its disclosed gaps closed with evidence

No code changed, nothing pushed. Head remains 6940d16460.

Round 27 posted nothing — its ledger reads findings: [], posted: 0, fresh: 0, prevPosted: 0 — and its 31 listed items are marked "not a blocker … not requested in this round". So there is no request to act on. Two of its disclosed gaps can nevertheless be closed here, and a third is located precisely rather than left dangling.

Closed: the Session.test.ts whole-file gap

The round discloses that its full-file Session.test.ts run (1451 tests) hit the 600 s wall-clock timeout before printing results. That file was run whole during the merge round on this head's content, and it finished well inside that limit:

  • Test Files 1 passed (1), Tests 1451 passed (1451)
  • Duration 271.06s (tests 264.34s)
  • runner summary line: END cli-Session exit=0

So the 600 s timeout was the review environment's, not the suite's. The file is confirmed green as a whole, not just its added cases.

Located: two of the three unverified Critical candidates are outside this diff

The round names three low-confidence candidates it could not execute. Two of them are web-shell consumption questions — userBlockBoundaryKey not folding embeddedResources, and SideTaskPanel's hasUserInteraction not counting them. Both symbols exist, in packages/web-shell/client/components/artifacts/SideTaskPanel.tsx and packages/web-shell/client/daemon/session/DaemonSessionProvider.tsx. This PR changes zero packages/web-shell files (its diff against the merged main is 46 files, none there).

That is not a claim that the candidates are wrong — it places them: they concern a downstream client not yet reading the new typed SDK field, which the PR body already declares out of scope ("Downstream clients must render the new typed SDK field to display a card"). They belong in follow-up work, not in this diff.

The third candidate — title-projection window units — is in this PR's area and remains unverified by either side. Recorded as an open lead with its anchor: MAX_TITLE_USER_DISPLAY_TEXTS = 20 (chatRecordingService.ts:126) is used for the Mod-key set size (:1782), the display-text ring length (:1793) and the per-turn window slice (:1826). No probe was run against it here either, so no conclusion is claimed.

Why the round could execute nothing

It reports qwen review scratch-tree returned available:false because the repo's local git config carries an includeif content-filter the command cannot screen. Worth recording for whoever owns the review pipeline: neither this worktree nor the main checkout has any includeif (git config --local --get-regexp '^includeif' is empty in both). The earlier review-pr job log shows the runner's own config carrying includeif.gitdir:/github/workspace/.git… pointing at a /github/runner_temp/git-credentials-*.config that no longer exists. So this is a CI-runner config artifact blocking the review's probe step, not a property of the repository or of this PR.

Not closed

The skipped lanes stay skipped and are not claimed as covered: Integration Tests (CLI, No Sandbox), Test (macos-latest), Test (windows-latest) and macOS unit tests did not run at this head, and the PR's own Tested-on table records Windows and Linux as untested by the author. The 31 deferred items were not reopened.

Still outstanding for a person

The PR needs 2 approving reviews. The five earlier needs_human entries belong to the two previous heads; web-shell E2E Smoke is being re-evaluated on this head with a fresh re-run budget. This watch does not approve, change review state, merge, or enable auto-merge.

中文说明

值守轮次 —— 第 27 轮评审已处置,并用证据闭合其两个披露缺口

未改代码、未推送,head 仍为 6940d16460。

第 27 轮未发布任何内容——台账为 findings: []、posted: 0、fresh: 0、prevPosted: 0——所列 31 条均标注「非阻断……本轮不要求修改」,故无可执行请求。但其披露的缺口中有两个可在此闭合,第三个予以精确定位而非悬置。

已闭合:Session.test.ts 全文件缺口

该轮披露其「Session.test.ts 全文件运行(1451 用例)撞到 600 秒墙钟上限,未打印结果」。该文件在合并轮次中确已就本 head 的内容整份跑过,且远在该上限之内:

  • Test Files 1 passed (1)、Tests 1451 passed (1451)
  • Duration 271.06s(其中 tests 264.34s)
  • 运行器汇总行:END cli-Session exit=0

故那 600 秒超时属评审环境,而非套件本身。该文件作为整体已确认绿,不只是新增用例。

已定位:三个未验证 Critical 候选中有两个不在本 diff 内

该轮点名三个无法执行的低置信候选。其中两个是 web-shell 的消费问题——userBlockBoundaryKey 未折叠 embeddedResources,以及 SideTaskPanel 的 hasUserInteraction 未计入它们。两个符号确实存在,位于 packages/web-shell/client/components/artifacts/SideTaskPanel.tsx 与 packages/web-shell/client/daemon/session/DaemonSessionProvider.tsx。本 PR 改动零个 packages/web-shell 文件(相对所合并 main 的差异为 46 个文件,无一在该目录)。

这不是断言候选有误,而是给它们定位:它们关乎下游客户端尚未读取新的类型化 SDK 字段,而 PR 正文已将其声明为范围外(「Downstream clients must render the new typed SDK field to display a card」)。应归后续工作,不属本 diff。

第三个候选——title-projection window 的单位——确在本 PR 范围内,且双方均未验证。在此作为未决线索记录并给出锚点:MAX_TITLE_USER_DISPLAY_TEXTS = 20(chatRecordingService.ts:126)同时用于 Mod 键集合大小(:1782)、显示文本 ring 长度(:1793)与 per-turn 窗口切片(:1826)。此处同样未对其跑探针,故不下任何结论。

该轮为何一无所能执行

它报告 qwen review scratch-tree 返回 available:false,原因是仓库本地 git config 带有该命令无法筛查的 includeif content-filter。为评审流水线的维护者记录一点:本工作树与主 checkout 都没有任何 includeif(两处 git config --local --get-regexp '^includeif' 均为空)。先前 review-pr 的 job 日志显示,是 runner 自身的 config 带着 includeif.gitdir:/github/workspace/.git…,指向一个已不存在的 /github/runner_temp/git-credentials-*.config。因此这是 CI runner 的配置残留挡住了评审的探针步骤,而非仓库或本 PR 的属性。

未闭合的部分

被跳过的 lane 仍属未覆盖,不作已覆盖声称:Integration Tests (CLI, No Sandbox)、Test (macos-latest)、Test (windows-latest) 与 macOS unit tests 在本 head 未运行,PR 自带的 Tested-on 表也记录作者未测 Windows 与 Linux。31 条延后项未重开。

仍待人工

本 PR 需要 2 个批准。先前五条 needs_human 属于此前两个 head;web-shell E2E Smoke 正在本 head 上以新的重跑预算重新评判。本值守不 approve、不改评审状态、不合并、不开自动合并。

— qwen-latest-series-invite-202610-m9 via Qwen Code (v0.26.8)

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

Labels

review/self-reported The linked issue was opened by the PR author (self-reported)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

bug(acp): embedded resource prompt blocks are lost from transcript replay

8 participants