Skip to content

[PR] compress: accept flat single-range summary-in-content calls, guard unrecognized refs separators, unmask reject reasons - #4253

Merged
ranxianglei merged 7 commits into
masterfrom
fix/compress-input-contract
Oct 22, 2026
Merged

ranxianglei merged 7 commits into
masterfrom
fix/compress-input-contract

Conversation

@ranxianglei

@ranxianglei ranxianglei commented Oct 22, 2026 •

Copy link
Copy Markdown
Owner

Closes #4252.

What was broken (live session on v0.1.204): the compress tool schema advertises two forms — "PREFERRED: content = ONE plain string holding ALL ranges" and the flat single-range {startId,endId,summary,topic}. A model doing single-range compression naturally merges them into {content: "<summary text>", startId, endId, topic} — which the kernel parser rejected as kind=content-not-array, so the model retried the same valid-intent call in a loop (6 identical rejects in 60ms observed, plus refold churn at 0% cache hit all night). Two sibling defects in the same path: a U+2212 minus dash silently narrowed m00150−m00168 to m00150–m00150 (end ref swallowed into the topic — wrong range folded), and header-only string entries reported a blanket content-not-array that hid the real "missing summary" reasons (#362 visibility).

Fix (root cause — the parser, not the callers):

  • kernel/src/parse-compress-input.ts parseObjectValue: string content with NO refs headers + top-level bounds ⇒ recovered as ONE flat range (diagnostics flag flatSummarySalvage). A JSON-encoded {summary,topic} inside the string is unwrapped; an explicit top-level summary field wins. Content that carries refs headers keeps its line-form reading — never hijacked (tested both ways).
  • Refs separator class now includes U+2212 minus and U+FF0D fullwidth hyphen; a ref pair with an unrecognized separator (m00150 => m00168) is rejected with "refs separator not recognized — refusing to narrow a ref pair to its first ref" instead of silently folding one message.
  • Header-only line entries: kind=no-valid-ranges + per-entry reasons; prose without refs stays content-not-array (existing classification tests unchanged).
  • kernel/src/compress-tools.ts: schema text documents the accepted flat shape on both the tool and the content param.
  • src/compress-tool.ts: proxy lane logs [acp-compress-input] flat-summary salvage: recovered N range(s)… (compress: literal backslash-n in string content still fails as kind=content-not-array, hiding the real missing-summary reason #3780-style observability).

Tests: +10 kernel parse tests (kernel/tests/parse-compress-input.test.ts), +3 proxy tests (tests/compress-reject-reasons.test.ts). Full kernel suite 1428/1428, proxy compress sweep 179/179 locally.

Not covered (separate follow-ups, noted in the issue): gate:none-resolved churn from refs aimed above the allocated ceiling; the refold/estimate economics.

kernel-change: #4252 compress input-contract fix in kernel/src/parse-compress-input.ts (accept the flat single-range shape {content:"

",startId,endId}, U+2212/U+FF0D refs separators plus unrecognized-separator guard, unmask header-only line-entry reject reasons) and matching schema text in kernel/src/compress-tools.ts. Host side cannot solve it: the lenient parser and the tool schema live in acp-kernel and are shared by every lane (proxy/plugin/standalone extensions converge on this one parser, acp-kernel#108).

…ntent; guard unrecognized refs separators; unmask line-form reject reasons

- kernel parseObjectValue: a string content with NO refs headers plus
  top-level bounds ({content, startId, endId, topic}) is the two
  schema-advertised forms merged — recover it as one range (JSON-encoded
  {summary,topic} in the string unwrapped; explicit top-level summary
  wins). Line-form content with refs headers is never redirected.
- REF_PAIR_IN_LINE / LINE_ENTRY_REFS_RE: accept U+2212 minus and U+FF0D
  fullwidth hyphen as range separators; a pair with an unrecognized
  separator is rejected with a precise reason instead of silently
  narrowing to the first ref.
- Header-only line entries now report kind=no-valid-ranges with their
  per-entry missing-summary reasons instead of a blanket
  content-not-array (#362 visibility).
- Schema text documents the accepted flat shape; proxy logs a
  flat-summary salvage warn line (#3780 style).
- Tests: 10 kernel parse tests + 3 proxy reject-reason tests; full
  kernel suite 1428/1428, compress sweep 179/179.
@ranxianglei ranxianglei changed the title compress: accept flat single-range summary-in-content calls, guard unrecognized refs separators, unmask reject reasons [PR] compress: accept flat single-range summary-in-content calls, guard unrecognized refs separators, unmask reject reasons Oct 22, 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 PR 对应 #4252(compress 调用被拒 + U+2212 破折号静默缩窄区间),检索 open/closed issue 与近期 PR 均无同机制条目;#362/#3780 是可见性与可观测性先例,非同缺陷。

标签: bug pi scope/kernel p1(缺陷类;事故现场是 pi 长会话;改动主体在 kernel 解析器;p1 依据 owner 2026-10-19 修订——缓存命中崩坏导致成本爆炸:~25 次 compress burst、0% cache hit、一夜 ~2.6M tokens 无归因回补)。

审核进行中,先报一个已确认的 CI 红灯:

  • kernel-guard 失败(其余 gate/静态检查全绿,test 分片因依赖被 skip):PR 正文缺少 Kernel Boundary 要求的 kernel-change: <reason> 声明。这是机械性缺口,我会在 PR 分支上补齐(正文追加声明行 + 按版本纪律补 kernel/package.json 版本号),然后继续逐行审代码与回归面。

后续回复给结构化八问结论。

@ranxianglei ranxianglei added bug Something isn't working p1 pi scope/kernel labels Oct 22, 2026
awork added 2 commits October 22, 2026 11:14
… text

The compress tool description and content param description now document the accepted flat single-range form {content: "<summary text>", startId, endId, topic?}. Golden regeneration via scripts/update-wire-contract-goldens.ts; the diff is exactly those two description strings across the four wire shapes (tool + param) in kernel.json/bili.json/manifest-*.json — no other surface moved.
Kernel Boundary rule 2 (CI kernel-guard): kernel/src changes require a version bump in kernel/package.json so accumulated changes can ship as an independent acp-kernel release later. Regenerates both lockfiles (version lines only).
@github-actions

github-actions Bot commented Oct 22, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: fix/compress-input-contract (8259905)

Option A — Install this exact build (recommended)

npm install -g billion-context@0.0.5074-pr.4253.f5ff3ed

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

Option B — Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr4253.tgz
npm install -g package

Install instructions are refreshed on each push.

The compress tool description/param text is injected into every outbound
body (compress.injectTool=true in the golden drive), so the schema-text
change shifts all pinned per-turn digests by the ~173-byte tool-JSON delta.
Regenerated via scripts/update-storage-render-goldens.ts (digest-only diff,
matches the CI failure pair 2a7d4d40 -> f98b8102 on chat turn 0).
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

收到,判断接受。storage-render 那两次「golden 重生」(#4253 f771c71、#4255 b6f9a884d)是止血不是修复——我照着历史上几条 "regen storage-render goldens" 的 commit 先例走了,但那些场景是生产 wire 字节合法变化(nudge 低水位戳、effort ledger 字段),这次漂移来自测试驱动没走生产接线:tests/golden-storage-drive.ts 直接 new ProxyServer(opts) 起服务,applyKernelConfig(注入 compress.injectTool=true)没跑,tool JSON 每个出站 body 都带一遍。把漂移后的 digest 重新钉死 = 把测试假象固化成基线,下次动文案还会再炸一次。这层我漏判了:只验证了「digest-only diff + 套件绿」,没问「这些字节本来该不该出现在被钉的 wire 上」。

分支状态我刚实际核对过(git fetch --prune 后逐 ref 查):

接下来怎么处理:

  1. [PR] compress: accept flat single-range summary-in-content calls, guard unrecognized refs separators, unmask reject reasons #4253:等 28c940e58 落地后我拉下来做定向验证(storage-engine-proofs + golden-storage-render + wire-contract-golden + parse-compress-input 套件 + typecheck/build),结果回帖并盯 CI 收尾。在此之前我不再往该分支推任何东西。
  2. [PR] fix: accept merged flat compress form, recognize U+2212/FF0D dashes, report honest rejection reasons (#4252) #4255:头 a10751733 带着同样的隐患(schema 文案 vs storage-render pin,而且它的 test lane 从未完整跑过一轮)。建议按先来后到关为 [PR] compress: accept flat single-range summary-in-content calls, guard unrecognized refs separators, unmask reject reasons #4253 的 duplicate(你也已经把修复投到了 [PR] compress: accept flat single-range summary-in-content calls, guard unrecognized refs separators, unmask reject reasons #4253);如果你想保留两个实现做对比,我就把同样的 drive 接线修复移植过去。你拍板,我不擅自关你的 PR。
  3. 本地 review 分支会同步到 origin 的新头;全程不 force-push。

八问

  1. 是什么问题:PR 改了 compress 工具的 schema 文案;测试驱动因未走生产接线把 tool JSON 注入每个出站 body,钉死的 digest 整体偏移 ~173B;我用「重生 golden」把它掩盖了。
  2. 什么原因导致的:根因在 tests/golden-storage-drive.ts 绕过 applyKernelConfig 直建 ProxyServer(测试假象);我的错是把历史 regen 先例套用到不适用的场景,没有追问「这些字节该不该在被钉的 wire 上」。
  3. 影响范围与严重程度:仅测试基线层。不重生 → CI 假红;重生 → 基线被污染,未来真正的 wire 回归可能被假基线吸收。无运行时用户影响。
  4. 怎么修复的:按你的方案——让 drive 走 applyKernelConfig / 真实启动路径反映生产接线,而不是重钉 digest(等 28c940e58 落地后由我验证)。
  5. 是否有回归问题:wire-contract golden 重生(d5848cf1a)不受此影响——它快照的就是工具 schema 本身,文案变更正是本 PR 的目的,保留;kernel 版本 bump + 双 lockfile 同步同样保留。
  6. 是否破坏缓存:不涉及(纯测试基建)。
  7. 是否改了设计方向:不改;drive 向生产接线对齐本来就是回归既有设计意图。
  8. 是否引入新配置:不涉及。

收工判定:#4253 等你的 commit 落地后我做验证收尾;#4255 等你拍板关或留。

@ranxianglei

Copy link
Copy Markdown
Owner Author

Golden regeneration added: the compress schema text is carried on the wire, so the doc change shifts per-turn digests — storage-render (4 wires) + wire-contract (4 manifests) regenerated and verified locally (golden-storage-render 5/5, wire-contract 22/22, storage-engine-proofs 12/12 wire parity). Re-running the failed legs.

ework-agent added 2 commits October 22, 2026 12:30
…NE plain string

Per review decision on #4252: the multi-form advertisement (array-of-objects
entries + flat single-range calls + summary-in-content hybrids) is what led
models to blend forms into shapes the parser rejected. The schema now
teaches ONLY the string form; every legacy shape stays functional in
parseCompressInput (open object schema, no required / additionalProperties —
flat calls still validate), it is just no longer documented.

- kernel compress-tools.ts: content + tool descriptions trimmed to the
  string form; flat-form top-level properties dropped; header contract
  comment rewritten for the new policy
- tests/wire-contract.test.ts: COMPRESS_FORM_PROPS guard reduced to
  [topic, content] with the rationale
- kernel 0.0.152 -> 0.0.153 (kernel-guard rule: schema change ships a bump)
- goldens regenerated: storage-render (4 wires) + wire-contract (4 manifests)

Verified: kernel 1428/1428, wire-contract 22/22, golden-storage-render +
compress-reject-reasons 12/12.
@ranxianglei

Copy link
Copy Markdown
Owner Author

Per the review decision: the schema now advertises exactly ONE compress form — content = ONE plain string holding ALL ranges (7e104d0). Legacy shapes (array-of-object entries, flat single-range calls, summary-in-content hybrids) remain functional in the parser — the schema is still an open object schema so flat calls validate — they are simply no longer taught. Wire-contract guard updated (COMPRESS_FORM_PROPS → [topic, content]), kernel bumped 0.0.152 → 0.0.153, goldens regenerated. Local: kernel 1428/1428, wire-contract 22/22, golden + reject-reason suites 12/12.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

核对完成(拉分支 head 7e104d0e2 逐条比对,未改动代码)。评论里的说法与分支实际状态一致,CI 关键门禁全绿。三件我独立复核过的关键点先说清:

  • 原红灯已解:kernel-guard 之前红是因为 PR 正文缺 kernel-change: 声明;现已补上(正文尾部有该段),kernel-guard / version-guard / lock-sync 均绿。
  • golden 重生合法性(此前 floor5 质疑 vs floor7 反驳):我独立裁定 owner 的解释成立。storage-render golden 是 digest-only,15 个 turn 每个恰好 −1020 字节(23307→22287、32017→30997… 完全均匀的常数差)——这正是「同一个 compress schema 对象序列化进每个出站 body、现在变小了」的指纹,不是 harness 注入假象。且 tests/golden-storage-drive.ts 本 PR 未改(harness 没动),旧基线本来就含 tools-per-body,delta 纯粹来自 compress 文案变化。wire-contract golden 的 delta 100% 落在 compress 工具描述文本上,非 compress 的工具名零改动([CI/testing] wire-contract anti-regression system: validation-parity fake upstreams + golden schema snapshots + constraint ledger + staged 4-protocol x 3-mode client smoke matrix #1304 台账没有缩水)。结论:重生合法,没有把测试假象固化进基线。
  • 「legacy 仍可用」这条保证是被钉死的:既有 kernel/tests/compress-schema-contract.test.ts 用真 validate(COMPRESS_PARAMETERS,…) 双向断言——flat {topic,startId,endId,summary} acceptedByBoth(过 schema 且解析成功)、trap shape rejectedByBoth、空壳调用「schema 保持宽松」。本 PR 没动这个文件,删掉 advertised props 后 CI 仍绿 = 证明删属性没破坏开放对象语义。Gate 双方向纪律满足。

八问

  1. 是什么问题:compress schema 同时教了「content=一个纯字符串」和 flat {startId,endId,summary} 两种形状;模型做单区间压缩时自然把它们混成 {content:"<摘要>",startId,endId},解析器当 content-not-array 拒掉 → 模型同参重试循环(60ms 内 6 次相同拒绝)+ 整夜 0% 命中的 refold 抖动。同路径两伴生缺陷:U+2212 减号把 m00150−m00168 静默缩成 m00150–m00150(折错区间)、header-only 字符串报一刀切 content-not-array 盖住真实原因。
  2. 什么原因导致的:根因在解析器不在调用方——它把「两种形状被合并」的输入直接判死而非恢复真实意图(单区间);refs 分隔符类不全(漏 U+2212/U+FF0D),遇未知分隔符时选了「缩成第一条 ref」这种会丢数据的兜底而不是拒绝。
  3. 影响范围与严重程度:波及所有走 ACP compress 的车道里模型自发单区间压缩的场景;p1——非不可用,而是触发后陷重试循环 + 缓存命中崩到 0%,成本爆炸(一夜 ~2.6M tokens 无归因回补)。属 owner 2026-10-19 修订的「缓存命中崩坏」类 p1。
  4. 怎么修复的:解析层根因修——(a) 只救「无 refs 头字符串 + 顶层 bounds」这一种被合并形状,恢复成单个 flat 区间(flatSummarySalvage),带 refs 头的行式内容绝不劫持(两向都有测试);(b) 分隔符类补 U+2212/U+FF0D;(c) 未识别分隔符直接拒绝并给精确原因(refusing to narrow a ref pair to its first ref),不再静默折一条;(d) header-only 条目细分 no-valid-ranges + 逐条原因。schema 按评审决定只教一种形状但保持开放对象,legacy 仍可用只是不再教。
  5. 是否有回归问题:逐项过既有行为——advertised schema 收窄=有意变更(PR 正文已披露,见下);legacy 接受度保留(开放对象 + 解析 salvage,既有 schema-contract 测试双向钉死);refs 识别增强(U+2212/U+FF0D 由误缩窄→正确区间,修 bug);未识别分隔符由静默折一条→拒绝报错(修 bug,旧行为是数据丢失方向);reject-reason 细化=纯可观测性不改折叠结果;golden 位移仅来自 compress 文案(非无关台账缩水)。无意外回归。
  6. 是否破坏缓存:机制本身没坏。schema 文案随 tools 数组进请求前缀,升级后每会话后续 turn 有一次性 prefix 重算(任何工具 schema 变更的固有代价,非常态破坏);存储侧 digest 位移是预期基线更新。
  7. 是否改了设计方向:架构/边界/数据流不变。「parser 是唯一执法点、schema 保持宽松超集不杀合法调用」(acp-kernel PR #172 (Layer 1) not in stable 0.0.46 — #362 data-loss/flooding still live until pin bump #374/[bug] 0.1.148 Claude 请求因 compress 顶层 anyOf schema 被 Anthropic 400 拒绝 #1299 契约)不动;唯一转向是「schema 从教多形状收敛为只教一种」=评审拍板的教学策略调整。
  8. 是否引入新配置:不涉及,无新 env/config/默认值。

行为变更披露(旧→新):① advertised schema:删 startId/endId/startRef/endRef/summary 五个属性 + 缩短 tool/content 描述 → 只剩 topic+content(有意,legacy 仍可解析);② U+2212/U+FF0D 分隔符:静默缩窄成单条 → 正确识别为区间边界(修 bug);③ 未知分隔符:静默折一条消息 → 拒绝并报精确原因(修 bug)。

CI 现状:全部 guard/gate/static-gates(ubuntu24+win22)/build/version/lock/docs/lane/kernel-guard 绿;test 分片 ubuntu-22×3 全绿;ubuntu-24×1 + windows-22×2 仍在跑,零红(约 7 分钟,慢在 windows/node24 车道)。本地未装 node_modules,按本地测试纪律不重复跑 CI 已在跑的全量套件。

标签:bug pi scope/kernel p1(沿用 floor3 判定,无需改)。

收工判定:代码正确、关键门禁全绿、golden 分歧已裁定为合法、legacy 可用性有测试兜底。这是 kernel 改动 → 无论 CI 结果如何都属人工合并档(§7.4 / Kernel Boundary),不自动合、我不代合也不 force-push。 等最后 3 个慢分片落地绿即可由你合并;若其间转红,系统会再次转发给我定向排查。

@ranxianglei ranxianglei added the ci:full Run the FULL CI matrix (all e2e lanes + 12-leg test matrix) on this PR label Oct 22, 2026
… ranges

beyondFrontierNote (#2146) only fires when EVERY requested endpoint sits
above the session's highest mapped ref. The incident shape — start inside,
end above (allocation lags rendering: refs visible in the transcript are
not yet minted in the store) — fell through to generic guidance, and the
model blind-retried against a moving ceiling (session 01a15993: 5×
gate:none-resolved inside 8 minutes, compress-loop breaker armed).

frontierClampNote now fires for exactly that partial case: names the
highest stored ref, says numbers visible in the transcript may not be
ingestible yet, and gives the actionable clamp (Retry with endId ≤
highest). Mutually exclusive with the all-above note; b-ref endpoints and
empty states never fire; the above-frontier endpoint stays verbatim in
the receipt (typo-correctable, per the #2146 mixed-case contract).

Verified: issue2146 suite 9/9, e2e-update-state + compress-reject-reasons
+ preflight-compress 11/11.
@ranxianglei

Copy link
Copy Markdown
Owner Author

Folded the first follow-up into this PR (8259905): partially-past-frontier ranges now get frontierClampNote — names the highest stored ref and the actionable clamp (Retry with endId ≤ m…) instead of generic guidance. This was the incident's 5× gate:none-resolved loop: start inside the frontier, end above it (allocation lags rendering), so the all-above note (#2146) never fired. The refold/estimate economics follow-up stays out — it's tuning work, not a defect fix.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

增量核对完成(新 head 8259905d2,只动了 src/stream.ts + tests/issue2146-compress-loop-breaker.test.ts,host 侧、无 kernel 改动)。核心部分我上一轮已验证过,这轮只审折叠进来的这个 follow-up。

改了什么 & 对不对:新增 frontierClampNote——当某个区间「端点在会话最高已存 ref 之上、但并非全部之上」(start-in/end-out)时,点名天花板(最高已存 ref)并给可执行钳位(Retry with endId ≤ m…),替代原来的通用提示。这正是事故的 5× gate:none-resolved 循环形状:分配滞后于渲染,模型看到已渲染的 ref 但 store 还没铸号,旧 beyondFrontierNote(#2146) 只在「全部之上」才触发,所以这种半跨边界的区间一路漏到通用提示、对着移动的天花板盲重试。逻辑我逐条对过:

范围合规:这属于本 PR 同一机制族(compress 拒绝引导,#2146 血脉,同在 src/stream.ts),按「连贯修复不拆片」折进来是对的;refold/estimate economics 明确留作调优不混入缺陷修复,切分干净。host-only → 无需再 bump kernel 版本,失败回执不进 wire/storage golden,也无需重生 golden(diffstat 确认没动 golden)。

八问

  1. 是什么问题:单区间压缩的 start 落在会话 frontier 内、end 越过 frontier(分配滞后渲染)时,旧的「全之上」提示不触发,模型拿到的是通用引导,对着一个还在涨的 ref 天花板反复盲重试(事故里 8 分钟 5 次 gate:none-resolved、breaker 都武装了)。
  2. 什么原因导致的:beyondFrontierNote 的触发条件是「每个端点都高于最高 ref」,而半跨界区间只有部分端点越界,落进了没有专门分支的空档 → 只能落到泛化提示,缺一句「你的上界最多到 m…」的可执行信息。
  3. 影响范围与严重程度:波及 pi/dsh 等 ACP 车道里「引用了刚渲染但尚未铸号的 ref」做压缩的场景;严重度 p1(同主 issue,缓存命中崩坏/成本爆炸家族),本 follow-up 是把该路径从「盲重试循环」修成「一次收敛」。
  4. 怎么修复的:加一条与全之上提示互斥的分支,只对「有端点越界但非全越界」的情形点名最高已存 ref + 给出 endId ≤ m… 钳位,让模型一次重试就收敛;不动压缩判定本身。
  5. 是否有回归问题:逐项过——只改失败回执文本(诊断面),不碰折叠/拒绝决策;其余输入形态回执逐字节不变;与 dsh内部问题 出现这个错误后,重启dsh可以发消息 但陷入无限压缩失败 #2146 提示互斥无双报;严格边界保留天花板可摄入;b-ref 排除。无意外回归。
  6. 是否破坏缓存:不涉及。失败回执不进请求前缀/tools,不影响上游 prefix cache 或存储 digest。
  7. 是否改了设计方向:不改。仍是「parser/kernel 管判定、host 侧负责把拒绝原因讲成可执行动作」的既有分工,只是补了一个漏掉的分支。
  8. 是否引入新配置:不涉及。

行为变更披露(旧→新):半跨界区间的失败回执,由通用引导 → 「点名最高已存 ref + Retry with endId ≤ m…」(有意,让重试一次收敛);提示串顺序在 ${beyond} 之后插入 ${clamp}。

CI 现状(新 head 8259905):全部 guard/gate/static-gates(ubuntu22/24 + win22/24)/build/version/lock/docs/lane/kernel-guard 绿;因本次动了 src/stream.ts(请求管道),CI 自动扩跑了全套真实客户端 e2e(e2e、hermetic registry ubuntu+win、real codex/dsh/opencode/pi、image-billing、V2-WebSocket)——全部绿,这正是 AGENTS.md 要求管道改动必跑的回归面,端到端验证了这个改动;零红,仅剩 14 个 unit-test 分片在后台跑。

收工判定:这个增量修复正确、测试到位、范围干净,与上一轮已验证的核心合并后整个 PR 从我的角度可合。仍属人工合并档(同一 PR 内含 kernel 改动 → §7.4 / Kernel Boundary 不自动合),我不代合也不 force-push。等 14 个 unit 分片落地绿即可由你合并;其间若任一转红会再次转发给我定向排查。

@ranxianglei
ranxianglei merged commit 94f75cc into master Oct 22, 2026
55 checks passed
@github-actions

Copy link
Copy Markdown
Contributor

📦 Merged Build Published

This merge to master was published as 0.0.735-master.94f75cc (rolling master tag).

npm install -g billion-context@0.0.735-master.94f75cc  # 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

bug Something isn't working ci:full Run the FULL CI matrix (all e2e lanes + 12-leg test matrix) on this PR p1 pi scope/kernel

Projects

None yet

Development

Successfully merging this pull request may close these issues.

compress calls rejected when the summary text rides in content next to top-level startId/endId — plus U+2212 dash silently folds the wrong range

1 participant