Skip to content

[PR] fix: fail open when a marker strip would blank the latest user turn (#5812) - #5816

Merged
ranxianglei merged 2 commits into
masterfrom
2026-11-06_marker-user-slot-fail-open
Nov 10, 2026
Merged

ranxianglei merged 2 commits into
masterfrom
2026-11-06_marker-user-slot-fail-open

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

What & why

stripTextSlot() (src/marker/engine.ts) replaced a text slot with a single space whenever the marker-line strip left it blank — without checking whether that marker line was the message's only content. A user message whose entire text is one line starting with a symbol + [ACP] (e.g. pasting a bug report like ❌ [ACP] stream error: ... 这是什么报错?) was delivered to the model as a single space, so the model saw an empty turn while the client UI still showed the original text.

Root cause

The raw-stage strip already skipped the status-body path for the latest user turn (the current turn is live input, not echoed history — the #2815 paste guard). But the marker-LINE path ran on every user+assistant slot regardless; when it blanked a slot it unconditionally set " ".

Fix

Extend the same "latest user turn" predicate to the marker-line blanking: if stripping the latest user turn collapses it to whitespace, return the original bytes instead of a space (counting nothing as stripped). Older user slots and assistant slots are unchanged — bili injects its own visibility markers into older user-role slots (src/compress-loop-responses.ts), so only the latest turn is safe to fail open. Renamed the option noStatusBody → latestUserTurn since it now governs both guards.

Behavior change (old → new)

  • Latest user turn whose whole text is one marker line: old → delivered as " " ; new → original bytes preserved (fail open); nothing counted as stripped.
  • Older user slots / assistant slots that are entirely a marker line: UNCHANGED — still degrade to " " (this is what keeps bili's own injected markers from leaking upstream).
  • Applies to both compression modes (the raw strip stage runs in both).

Config surface

N/A — no config field added, renamed, or re-semantics'd. No env var. No CLI flag. No persistence/wire-shape change.

Tests

Regression tests in tests/marker-lane-coverage.test.ts:

  • a marker-shaped LATEST user turn survives intact on all four wires (anthropic / openai / google / responses);
  • boundary: an OLDER marker-only user slot + assistant marker slots still degrade to a space while the latest user turn fails open;
  • non-degenerate: a marker line WITH prose in the latest user turn strips the marker and keeps the prose.

Also updated the existing KDD#10 control assertion (a latest-user status-marker slot now stays intact rather than becoming a space).

Local: typecheck clean; tests/marker-lane-coverage.test.ts 12/12, tests/fastpath.test.ts 19/19, plus 7 other marker-subsystem files 78/78. Full-suite verdict = remote CI on this head.

Fixes #5812

@github-actions

github-actions Bot commented Nov 6, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-11-06_marker-user-slot-fail-open (e322097)

Latest build — install this exact version

npm install -g billion-context@0.0.8926-pr.5816.d4c5665

This exact version is immutable and reproducible. The rolling pr tag tracks the newest test build across all PRs.

Recent builds (commit → version)

commit version build
e322097 billion-context@0.0.8926-pr.5816.d4c5665 build
2c9de23 billion-context@0.0.8912-pr.5816.fdc40d1 build
9d2bfad billion-context@0.0.8374-pr.5816.2b4aefc build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr5816.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei ranxianglei changed the title fix: fail open when a marker strip would blank the latest user turn (#5812) [PR] fix: fail open when a marker strip would blank the latest user turn (#5812) Nov 6, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

查重: 不重复 —— 本条即 GitHub issue #5812 的修复 PR(GitHub PR #5816)的镜像;open [PR] 列表里没有第二个针对同一主题(marker strip 把最新用户轮打成空格)的条目。

标签: bug + scope/wiring + p2(缺陷类;marker 剥离属请求管道/wire 改写面;p2 = 特定条件下有问题——仅当整条最新消息就是一行 marker 形态文本时模型看到空轮,有绕行不影响核心链路)。

这是 [PR] 镜像(agent 已完成的工作),按规矩不重做,改为核对代码 + CI。目前进展:

  1. PR 状态: [PR] fix: fail open when a marker strip would blank the latest user turn (#5812) #5816 open、mergeable、单 commit(9d2bfadc)压在最新 master 上,diff 只动 src/marker/engine.ts(+43/-17)与 tests/marker-lane-coverage.test.ts(+61),与描述一致,无跑题文件/无 lockfile 噪音。
  2. 代码核对(已通读全 diff):修法是把 Inbound markers: one declarative strip engine for all four wires (#2789, #2798, #2811 plan B) #2815 已有的「最新用户轮」谓词从 status-body 路径延伸到 marker-line 路径——stripTextSlot() 里剥离后若塌缩为空白且是最新用户轮,返回原始字节且不计数;四个 wire walker(anthropic/openai/google/responses)只是把选项 noStatusBody 改名 latestUserTurn,谓词本身(i === lastUserIdx)未动。旧用户槽/assistant 槽的降级为空格行为不变(边界测试钉住了)。fail-open 分支放在计数之前,「nothing counted as stripped」成立。
  3. 本地定向验证(PR head worktree): tests/marker-lane-coverage.test.ts + tests/fastpath.test.ts 共 31/31 通过。typecheck 与其余 marker 子系统文件正在补跑。
  4. CI: 已完成的 lane 无红(18 success / 12 skipped),其余 shard 仍在跑,最终判定以 CI 为准。

完整八问结论稍后补在本线程。

@ranxianglei ranxianglei added bug Something isn't working p2 scope/wiring labels Nov 6, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

核对 + 验证全部完成。结论:可合并(合并留给你:#5816)

八问结论

  1. 是什么问题:用户最新消息的整段文本恰好是一行 marker 形态文本(比如粘贴 ❌ [ACP] stream error... 这是什么报错? 这种报错行)时,出站 raw 层的 marker 剥离把该槽位塌缩成一个空格——模型看到空轮,客户端 UI 却显示原文,用户体感是"我明明贴了报错,模型却像没看见"。
  2. 什么原因导致的:stripTextSlot()(src/marker/engine.ts)有两条会产出「剥完为空」的路径——status-body 和 marker-line。Inbound markers: one declarative strip engine for all four wires (#2789, #2798, #2811 plan B) #2815 的粘贴守卫只覆盖了 status-body 路径(跳过最新用户轮);marker-line 路径对所有 user+assistant 槽照跑,剥出空就无条件降级 " ",没检查那行 marker 是不是消息的全部内容。
  3. 影响范围与严重程度:p2。触发面窄——最新用户消息的全部内容必须恰好是一整行 marker 形态;该条件下当轮输入对模型为空(单轮功能失效),无数据丢失,不影响其他槽位/旧历史/assistant 槽。
  4. 怎么修复的:把 Inbound markers: one declarative strip engine for all four wires (#2789, #2798, #2811 plan B) #2815 已有的「最新用户轮」谓词延伸到第二条路径——最新用户轮若剥离后塌缩为空白,fail open 返回原始字节(不计数);旧用户槽与 assistant 槽保持降级为空格(这正是防止 bili 自己注入的可见性 marker 泄漏到上游的机制)。选项改名 noStatusBody → latestUserTurn(现同时管两个 guard)。修在决策点(根因),不是输出侧打掩码。
  5. 是否有回归:无。逐项审过触碰路径的既有行为:status-body 消费(旧槽)不变;Inbound markers: one declarative strip engine for all four wires (#2789, #2798, #2811 plan B) #2815 的 skip 行为不变;marker-line 部分剥离(留正文)逻辑语义逐字等价;旧 user/assistant 空槽→空格不变(新边界测试钉住,negative 方向齐备);nudge-echo 整条移除路径未动且 fail-open 内容不可能匹配 nudge 文本(形状不同);thinking/redacted/reasoning/thought 签名部件仍在 strip walk 之前显式跳过(KDD#10 字节不变断言保留——我核过该 fixture,messages[1] 确是唯一/最新 user 轮,控制断言改动正确,不是恒真);panel 移除未动。唯一变更即 PR 描述已披露的 old→new。测试有牙:精确字符串 + 计数断言,gate 双向(fail-open 该触发时触发 / 不该触发时仍降级)+ 非退化 case(marker 行带正文→剥 marker 留正文)。
  6. 是否破坏缓存:否。只有最新用户轮的字节发生变化,该位置每轮本来就在变、位于一切 cache seam 之外;旧槽前缀字节逐位不变;重试/增量游标/去重键不受影响。
  7. 是否改了设计方向:否——与 Inbound markers: one declarative strip engine for all four wires (#2789, #2798, #2811 plan B) #2815 同一哲学(最新轮 = live input,不是 echo 历史),把既有谓词延伸到同 stage 的兄弟 guard。已知取舍(测试注释里写明):raw 层无法凭内容区分「用户粘贴的 marker 行」与「伪造 marker」——但伪造进最新轮在 status-body 路径上本 PR 之前就已可能,此修复是让两条路径对齐,不扩大信任面。
  8. 是否引入新配置:无。无 config 字段/env/CLI flag/持久化/wire 形状变化。

验证证据

  • diff:单 commit 9d2bfadc,基于最新 master,仅 2 文件(src/marker/engine.ts +43/-17、tests/marker-lane-coverage.test.ts +61/-1),无跑题文件、无 lockfile 噪音;
  • 本地定向(PR head worktree):typecheck clean;marker 子系统 5 个套件(marker-lane-coverage / fastpath / acp-status-marker-block / issue1029-acp-marker-stale-refs / marker-render-contract)53/53 通过;
  • CI(head 9d2bfadc):47 lanes = 35 success + 12 skipped,0 红。

判定:可合并。

…5812)

stripTextSlot replaced a text slot with a single space whenever the marker-line strip left it blank, without checking whether that marker line was the user's only content. A pasted marker-shaped error line (a bug report whose first line starts with a symbol + [ACP]) was delivered to the model as one space.

The raw-stage strip already skipped the status-body path for the latest user turn (the current turn is live input, not echoed history), but the marker-line path ran on every slot. Extend the same predicate to marker-line blanking: if stripping the latest user turn collapses it to whitespace, return the original bytes instead of a space, counting nothing as stripped. Older user slots and assistant slots are unchanged - bili injects its own visibility markers into older user-role slots, so only the latest turn is safe to fail open.

Rename the option noStatusBody -> latestUserTurn to reflect that it now governs both guards.

Fixes #5812
@ranxianglei
ranxianglei force-pushed the 2026-11-06_marker-user-slot-fail-open branch from 9d2bfad to 2c9de23 Compare November 10, 2026 07:55
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Conflict resolved, branch rebased onto current master, full CI green on the new head. Nothing else was red this run — the only blocker was the merge conflict.

What happened this run

No design decision was needed, no authorization gate was hit (diff stays in src/marker/ + tests/; no kernel/, no config surface), and no new issue/PR was warranted. Merging remains human-only: #5816


七问(本轮 CI 处置)

  1. 是什么问题:PR 分支停在旧 master(0aace2f203)上,与当前 master(4dea18eabf)冲突(dirty),无法合并;红项仅此一项。
  2. 什么原因导致的:master 侧唯一触碰同文件区间的提交是 stripRawStage rescans all history text every turn — a fourth O(history) sync path beyond #4528/#5744 #6018(dd4013850b)——它在 stripTextSlot() 里给 marker-line 剥离加了 t.includes("[ACP]") 字面量门(性能优化);本 PR 在同一代码块内加了最新用户轮 fail-open 分支,两边改了同一块文本,产生内容冲突。
  3. 影响范围与严重程度:仅本 PR 的合并被阻塞(特定条件下有问题——不解决就合不了);不影响任何已发布行为、不影响其他 PR。
  4. 怎么修复的:rebase 到当前 master,冲突取「双方意图并集」——保留 master 的字面量门、把本 PR 的 fail-open 放进门内。语义上等价于 PR 原意(无 [ACP] 字面量时正则必不命中,该路径不可能把槽位剥空,fail-open 无需触发),同时保住 stripRawStage rescans all history text every turn — a fourth O(history) sync path beyond #4528/#5744 #6018 的性能优化;测试文件自动合并,两侧测试共存。
  5. 是否有回归问题:无。逐项核过:master 侧 stripRawStage rescans all history text every turn — a fourth O(history) sync path beyond #4528/#5744 #6018 的行为(字面量门正/负行、strippedMs 计时与慢日志)原样保留且其测试全部通过;本 PR 侧 fail-open 四 wire 回归行通过;旧槽降级为空格、部分剥离留正文等既有语义均未动。本地定向 87/87 + typecheck clean;CI 全量 30 lanes 0 红。rebase 本身未引入任何新行为变更(PR 已披露的 old→new 不变)。
  6. 是否破坏缓存:否。未触碰任何缓存/重试/游标/去重逻辑;出站字节行为与 PR 原描述一致(仅最新用户轮在剥空时保留原文)。
  7. 是否改了设计方向:否。纯机械冲突消解,两条已评审的设计(stripRawStage rescans all history text every turn — a fourth O(history) sync path beyond #4528/#5744 #6018 性能门 + marker strip blanks a user message whose entire text is a [ACP] line — the model receives a single space #5812 fail-open)按各自原意叠加,无架构/边界/数据流转向;无新增配置。

判定:冲突已解、CI 全绿,可合并(合并操作留给你)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

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

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

⚠️ Posted as a comment review: GitHub rejects APPROVE on the author's own PR (HTTP 422 self-approval ban). The verdict below stands.

APPROVE — independent end-to-end review of head e322097c09 (fix commit 2c9de23124 + my one-line comment repair).

Direction (original issue). Issue #5812 is real and correctly diagnosed at the root: stripTextSlot() degraded any marker-line strip that blanked a slot to " " unconditionally, while the existing #2815 paste guard exempted only the status-body path for the latest user turn. The fix extends that same predicate to the marker-line path at the decision point — the producer of the empty slot — not a mask downstream. All four wire walks share this one function, so the same root cause cannot resurface per-wire.

Rebase interaction (#6018). The fail-open branch sits inside master's new t.includes("[ACP]") literal gate (src/marker/engine.ts:80-90). The gate cannot swallow the fix: MARKER_LINE requires the literal [ACP], so where it is absent nothing matches and no slot can be blanked there anyway. Verified hunk by hunk.

Regression audit — existing behaviors on every touched path:

  • older user slots & assistant slots still degrade to " " (flag false → old code path; pinned by the new boundary test incl. tool_calls survival);
  • partial strips (marker line + prose) keep byte-equivalent semantics (the stripped = out.trim() refactor preserves the old ternary);
  • status-body consumption unchanged (rename only; #2815 skip preserved);
  • KDD#10 signed slots (thinking/redacted/reasoning/thought/thoughtSignature) are still skipped before the strip walk; the flipped control assertion is correct — I checked the fixture: messages[1] is the sole/latest user turn, not a tautology;
  • nudge-echo splice runs after stripTextSlot; a fail-open'd pasted line cannot match nudge text (different shape) — no interaction;
  • fold-reconcile's markerStripped input now excludes fail-open'd lines from the artifact count — more accurate (they are user input, not bili artifacts) and stable per pass, so no churn/misdiagnosis introduced;
  • fastpath watermark lane + fork-adoption lanes run the same engine with lastUserIdx computed inside their fragment — consistent, no double handling;
  • both compression modes covered: the raw stage runs unconditionally in every prepare-* lane before mode-dependent projection (verified all four call sites).

Tests. Real assertions with teeth: exact-byte preservation + markerLines === 0 count assertions on all four wires; boundary rows prove the gate does NOT over-fire (older/assistant slots still blank); non-degenerate row proves marker+prose still strips the marker and keeps the prose. Gate discipline (#3617) satisfied in both directions.

CI. Mirror cache at 2c9de23124: 30 checks, 0 failed / 0 pending. Local targeted re-run after my comment commit: marker-lane-coverage + fastpath 33/33 green. The comment-only commit re-triggers CI; no behavior delta possible.

Conventions. Rename complete (zero residual noStatusBody repo-wide); diff clean (2 files, every line relevant); behavior change disclosed old→new in the body; config-surface N/A correctly declared; no kernel/ touch → no authorization gate; Fixes #5812 present.

One adjacent finding is routed separately as 🔶 需要决策 (panel/report whole-message splice can still remove a pasted latest turn wholesale — pre-existing, different class, a semantic scope call): see the follow-up comment.


八问(中文,供人工复核)

  1. 是什么问题:用户最新消息整条恰好是一行「符号 + [ACP]」形态的文本(典型场景:把 bili 自己的报错行原样贴进输入框求助)时,出站 raw 层的 marker 行剥离把整个槽位剥空后无条件填一个空格——模型收到空轮,客户端 UI 却显示原文。
  2. 什么原因导致的:stripTextSlot()(src/marker/engine.ts)的 marker-line 路径对「剥完变空」一律降级 " ",不判断这一行是不是消息的全部内容;#2815 已有的「最新用户轮 = live input」谓词此前只豁免 status-body 路径,marker-line 这条兄弟分支漏在守卫外。
  3. 影响范围与严重程度:p2、触发面窄——最新一条 user 消息的全部内容必须恰好是一整行 marker 形态;四条 wire × 两种压缩模式均受影响;该条件下当轮输入对模型为空(答非所问/反问),无数据丢失、会话可继续、有绕行。
  4. 怎么修复的:把既有「最新用户轮」谓词延伸到 marker-line 空白化这一步——剥空且是最新用户轮时返回原始字节(fail open)、不计剥离数;旧 user 槽与 assistant 槽保持降级为空格(保住 bili 自注入 marker 不外泄的既有不变量)。选项改名 noStatusBody → latestUserTurn 反映其现管两个守卫。修在决策点(根因),不是输出侧打掩码。
  5. 是否有回归问题:无。逐项审过触碰路径的既有行为:旧槽降级不变(边界测试钉住);部分剥离语义逐字等价;status-body 消费仅改名;KDD#10 签名槽仍在 walk 前显式跳过(控制断言翻转经 fixture 核对正确);nudge 整条移除与 fail-open 内容形状不冲突;fold-reconcile artifact 计数排除 fail-open 行更准确且逐轮稳定;fastpath/fork-adoption 走同一引擎行为一致;rebase 与 #6018 字面量门的交互逐 hunk 核过(门内无 [ACP] 字面量时正则必不命中,不可能吞掉修复)。唯一行为变更即 PR 已披露的 old→new。
  6. 是否破坏缓存:否。变化只发生在最新用户轮位置——该位置每轮本来就在变、位于一切前缀缓存 seam 之外;旧槽前缀字节逐位不变;重试/增量游标/去重键不受影响。
  7. 是否改了设计方向:否。是 #2815 paste-guard 哲学在同一 stage 的延伸(最新轮 = live input,不是回显历史),不改架构/边界/数据流。已知取舍:raw 层无法凭内容区分「用户粘贴」与「伪造 marker」——伪造进最新轮的暴露面在本 PR 之前经 status-body 路径就已存在,本 PR 是让两条路径对齐,不扩大信任面。
  8. 是否引入新配置:无。无 config 字段/env/CLI flag/持久化/wire 形状变化。

判定:可合并(合并留给人工)。相邻缺口(panel 整条 splice 仍可能删掉粘贴的最新轮)已按 🔶 需要决策 流程单独留言,不阻塞本 PR。

@ranxianglei

Copy link
Copy Markdown
Owner Author

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

已顺手修:PR 编辑把 stripTextSlot 头注里的 render-tag 示例整段吞掉了(src/marker/engine.ts:54 变成无语义的 "Render-tag spans () are",KDD#2 角括号工具坑)——已按仓库惯例用 hex escape 恢复为 (\x3cacp …\x3e)。纯注释变更,改后定向测试 marker-lane-coverage + fastpath 33/33 绿。+ commit e322097

@ranxianglei

Copy link
Copy Markdown
Owner Author

🔶 需要决策

(ai: vllm-qwen/qwen3.8-27b) Review 中发现一个同失败域的相邻缺口——非本 PR 引入、超出 #5812 范围;是否/如何扩展粘贴守卫是语义范围判断,需要你拍板,故单独列出、不动代码。

缺口:入站 raw 层的两类整条移除仍可能把「最新用户轮」整体删掉(比 #5812 的空格更重——模型直接看不到这一轮),而 #2815/#5812 的粘贴守卫未覆盖它们:

  1. panel/report 整条 splice:stripAcpPanelMessages(src/acp-panel.ts:102-116)对任何全文匹配 panel 语法(box panel / fallback panel / [acp-cache]…[/acp-cache] / [acp-rule]…[/acp-rule])的 user 消息直接 splice——包括最新一条,无豁免;responses 变体 stripAcpPanelResponsesInput(src/acp-panel.ts:267-283)与 google walk 的整条移除(src/marker/engine.ts:174-186)同样。触发场景:用户把一份 [acp-cache]/[acp-rule] 报告或面板框原样整条贴进输入框求助(「这个报告什么意思?」)——当轮对模型完全消失。
  2. nudge-echo splice:要求消息以 bili 自己的 nudge 长文精确结尾,自然输入命中概率≈0——列此仅为完整,可忽略。

为何不在本 PR 顺手修:把守卫扩到这两类是语义范围决策且有真实反风险——若某宿主把报告形态的上下文块追加成尾部 user 消息,fail-open 会让它每轮泄漏到上游。#2815 的先例是按类逐次扩展、以现场证据驱动;panel 类目前无现场报告。本 PR 的范围(marker-line 类)本身正确,不应由 reviewer 越权扩大。

选项:

  • A(推荐):维持现状(status-body + marker-line 已守卫),panel 类记为已知残留,出现首个现场报告再按同类修法扩展;
  • B:预防性把最新轮守卫扩展到 panel/report splice 三类(全 wire);
  • C:折中——只对纯文本报告包装([acp-cache]/[acp-rule],最可能被原样粘贴)加守卫,box/fallback panel 维持严格。

代价对比:B/C 在未证实前给怪宿主多开一条泄漏面;A 留一个低概率 p2 形空洞。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b) Review of PR #5816 done: full independent review posted on the PR with verdict APPROVE / 可合并 (posted as a COMMENT review — GitHub 422-bans self-approval on the author's own PR; human approval click still required), incl. 八问 and per-path regression audit; 已顺手修: restored the render-tag example stripped from the stripTextSlot header comment at src/marker/engine.ts:54 (commit e322097c09, pushed, targeted tests 33/33 green); one adjacent pre-existing gap (panel/report whole-message splice can still delete a pasted latest user turn wholesale) routed as 🔶 需要决策 with options A/B/C on-thread + needs-decision label — non-blocking. CI mirror at head: 30 checks, 0 failed.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) Task wrap-up for PR #5816 review lane:

Done & verified:

  • Full independent review posted on the PR with verdict APPROVE / 可合并 (regression audit per touched path + 八问). Note: GitHub rejects APPROVE on the author's own PR (HTTP 422 self-approval ban), so it landed as a COMMENT review — the human approval click still has to come from a person.
  • Small finding fixed on the spot: the PR edit had stripped the render-tag example from the stripTextSlot header comment (src/marker/engine.ts:54) — restored with hex escapes per KDD#2 house style. Commit e322097c09 pushed to the PR branch; targeted tests (marker-lane-coverage + fastpath) 33/33 green after the change.
  • CI mirror at head: 30 checks, 0 failed / 0 pending.

Pending owner decision (non-blocking): one adjacent pre-existing gap found during review — panel/report whole-message splices (src/acp-panel.ts:102-116 / :267-283, src/marker/engine.ts:174-186) can still delete a latest user turn wholesale when the user pastes an [acp-cache]/[acp-rule] report or panel box verbatim. Routed as 🔶 需要决策 on the PR with options A (recommended: keep scope, act on first field report) / B / C; needs-decision label applied. Code untouched; PR merge not blocked by this.

(The immediately preceding thread comment was the syncer echoing my own decision comment back — no new input there.)

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) 此条是同步器把我自己的「已顺手修」回执(commit e322097c09)回显到本线程,无新输入、无新决策,状态不变。待办仍只有一项:panel/report 整条 splice 缺口的 A/B/C 决策等你拍板(详见 PR #5816 上的 🔶 需要决策 评论)。其余均已收口。

@ranxianglei
ranxianglei merged commit 1fbcdcf into master Nov 10, 2026
84 checks passed
@github-actions

Copy link
Copy Markdown
Contributor

📦 Merged Build Published

This merge to master was published as 0.0.1164-master.1fbcdcf (rolling master tag).

npm install -g billion-context@0.0.1164-master.1fbcdcf  # this exact build
npm install -g billion-context@master  # always the newest merged build

npm view billion-context@master prints the current version at any time.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

marker strip blanks a user message whose entire text is a [ACP] line — the model receives a single space

1 participant