Skip to content

[PR] fix: judge ACP confirmation marker icon from the first line only - #5388

Open
ranxianglei wants to merge 5 commits into
masterfrom
2026-11-01_visibility-marker-first-line
Open

ranxianglei wants to merge 5 commits into
masterfrom
2026-11-01_visibility-marker-first-line

Conversation

@ranxianglei

@ranxianglei ranxianglei commented Oct 31, 2026 •

Copy link
Copy Markdown
Owner

Closes #5387

What

buildVisibilityMarker (src/loop/core.ts) chose the ❌/tool icon by scanning every non-empty line of a tool receipt for FAILED / not found / is required. But success receipts embed model-authored content from line 2 onward (the compress summary fingerprint excerpt summaryFingerprintLine, decompress's restored history, search evidence snippets, acp_status block topics). Whenever that content merely mentions a keyword, a successful compress rendered as:

❌ [ACP] Compressed m00001–m00005 → 1 block(s), ~120 tokens saved.

Icon contradicts text. Display layer only — control flow never used this path (the #5384 classifier is a separate function that already judges line 1).

Fix — protocolized judgment (owner directive 2026-11-11)

The recurring class ("success/failure misjudged from receipt text") should not be patched with another text heuristic. The executors already return a structured outcome (#1875 ProxyToolResult.outcome: compress tri-valued applied/partial/refused, every other tool success/failure), and every marker call site can reach it — so the judgment is now protocolized:

Call-site inventory (all threaded; no legacy path left without the outcome):

Call site outcome source
src/loop/core.ts streaming emit + streamedMarkers the executeProxyToolAsync result object
src/loop/core.ts re-request coreMessages markers proxyResults entries (carry the outcome from the same execution)
5 wire adapters' emitMarker (google / anthropic / openai / responses / commandcode) interface param threaded from core
src/compress-loop-responses.ts surfaceProxyJson + re-request items toolResult.outcome; the exception path explicitly stamps outcome="failure"

What the protocolization catches that NO text scheme can: keyword-less failure receipts (pre-fix acp_read_original_image) and the [Compression REFUSED: …] refusal whose wording contains no FAILED at all (rendered a SUCCESS icon under every prior scheme).

Mirror defect fixed in the same PR: acp_read_original_image's three failure receipts now follow the [acp_read_original_image FAILED: …] convention (model-visible wording for other consumers; the icon itself is now driven by outcome="failure").

Behavior changes (old → new)

  1. Marker icon judgment basis: all-lines text scan → structured outcome (first-line text scan only as the absent-outcome fallback). Success receipts whose embedded content mentions a keyword: ❌ → correct tool icon. Refusals ([Compression REFUSED: …], gate refusals): 📦 (wrong — no keyword in the text) → ❌. Keyword-less failures (acp_read_original_image): 📦 (wrong) → ❌. Genuine FAILED failures: unchanged ❌.
  2. acp_read_original_image failure receipt wording: acp_read_original_image: <msg> → [acp_read_original_image FAILED: <msg>]. Model-visible text in plugin mode changes; no control-flow consumer scans this tool's result for keywords (all loop FAILED-scans are name-gated to compress/decompress/absorb).
  3. Nothing else client-visible moves. Marker lines stay human-readable <icon> [ACP] <first line> text; strip/replay contracts unchanged — MARKER_LINE (src/loop/tag-echo-filter.ts:409) matches <So> [ACP] … regardless of icon, and the anchors.ts shape convention + golden contracts do not pin icon+keyword combos.

Adjacent control-flow text scans (anyCompressFailed in loop/core.ts, retry/recovery classification) are deliberately untouched — that is #5384's territory.

Tests

Updated tests/issue5389-marker-icon-first-line.test.ts (8 tests, both directions per #3617 gate discipline):

  • real applyRanges success whose summary contains FAILED (fingerprint line carries it on a non-first line — asserted as test teeth) → 📦 via outcome="applied";
  • protocol matrix both ways: success outcome + keyword on line 1 → tool icon; failure outcome + keyword-free text → ❌; [Compression REFUSED: …] → ❌ with the outcome but 📦 without it (proof of the hole every text scheme had); partial → 📦;
  • fallback path (no outcome): embedded-keyword successes keep their tool icons (decompress / search_context / acp_status); every real failure shape still ❌ incl. the exception-path single line;
  • real partial fold (one valid + one rejected entry) → 📦 with the keyword in an embedded line; real no-op re-compress (Compression conflict banner (bili v0.1.205, node v24.19.0, win32-x64) #4521) → 📦 pinned;
  • mirror defect: invalid/missing refs → [… FAILED: …] + outcome="failure" + ❌, guidance substrings preserved;
  • strip compatibility: success and failure markers both removed by shape, icon-agnostic.

Targeted local suites green (209 tests across 18 files) + npm run typecheck clean; full-suite verdict = remote CI on this head.

No config surface changes. No kernel/ changes.

buildVisibilityMarker scanned EVERY receipt line for FAILED / "not found" /
"is required" when choosing the confirmation icon, so a SUCCESS compress
whose summary fingerprint excerpt (or a decompress's restored history, a
search evidence snippet, an acp_status block topic) merely mentioned a
keyword rendered as ❌ while its text said "Compressed …". The first
non-empty line is the authoritative status line for every proxy tool —
every failure receipt carries its keyword there; judge only that.

Mirror defect in the same mechanism: acp_read_original_image failure
receipts carried no keyword at all → failures rendered 📦. They now follow
the [tool FAILED: …] convention so both judgment schemes read them right.

Behavior changes (old → new):
- marker icon judgment: all non-empty lines → first non-empty line only.
  Success receipts with keyword mentions in embedded content: ❌ → correct
  tool icon. Real failures (keyword on line 1): unchanged.
- acp_read_original_image failure receipts: "acp_read_original_image: …"
  → "[acp_read_original_image FAILED: …]" (model-visible text in plugin
  mode; icon flips from wrong 📦 to correct ❌).

Regression tests: tests/issue5389-marker-icon-first-line.test.ts (both
directions per #3617 gate discipline).

Closes #5389
@github-actions

github-actions Bot commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-11-01_visibility-marker-first-line (b668c9d)

Latest build — install this exact version

npm install -g billion-context@0.0.9182-pr.5388.b190fb2

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
b668c9d billion-context@0.0.9182-pr.5388.b190fb2 build
297925c billion-context@0.0.7676-pr.5388.543eaf4 build
dbd5d64 billion-context@0.0.7672-pr.5388.b621197 build
655a3a3 billion-context@0.0.7611-pr.5388.9ccdebc build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr5388.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei ranxianglei changed the title fix: judge ACP confirmation marker icon from the first line only [PR] fix: judge ACP confirmation marker icon from the first line only Oct 31, 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(GitHub #5388 / ework #5390)是 issue #5389「成功压缩画成 ❌ 图标」的修复本体,一个 issue 一个 PR。相邻条目我都核过、均不同机制:#5384 改的是控制流分类器(只看首行状态行,与本函数正交)、#1029 是回执「单行即权威状态行」契约的来源、#1708 改的是 visibilityMarkers 三态开关门控(不碰图标判定)。open+closed 无第二个同机制(首行图标判定)PR/issue。

我独立重读了 diff 与全部回执产出点(未照抄 #5389 floor 2/3 的表述)。以下为我的审核结论。

结构化八问

  1. 是什么问题:一次成功的 compress/decompress/search_context/acp_status,只要回执第 2 行起内嵌的模型内容(压缩摘要指纹摘录、decompress 恢复正文、search 证据摘录、acp_status 块主题)恰好提到 FAILED/not found/is required,确认标记就画成 ❌,与文字里的 “Compressed …” 自相矛盾。
  2. 什么原因:buildVisibilityMarker(src/loop/core.ts:78)判图标时扫了回执所有非空行;而全仓回执契约是「首个非空行即权威状态行」(ACP 状态标记被拼接进用户消息文本块;跨会话代际 stale refs 压缩整批失败 #1029),第 2 行起全是模型自撰内容,天然易含这些字样——拿内容当状态分类就是错层。关键佐证:该函数本来就只渲染首行(非 acp_status 分支 inner = lines[0]),修完后“图标判定的行 == 它显示的行”,天然自洽。
  3. 影响范围 / 严重度:纯展示层。仅 visibilityMarkers !== false 的车道可见(pi/opencode 默认可见;Hermes 生产 false 不可见)。控制流不受影响(见第 5 问 + 下方独立发现)。p3。
  4. 怎么修的:图标判定改为只看首个非空行。顺带修同机制镜像缺陷:acp_read_original_image 三条失败回执原本不带任何关键字 → 失败被画成 📦,现对齐为 [acp_read_original_image FAILED: …](可操作指引逐字保留)。
  5. 回归审查(我逐项核过):
    • 每个真实失败回执的关键字都在首行 → 仍 ❌,没有任何真实失败被改判成功图标。逐一核对:compress(stream.ts:792/1009/1013/1017/1062/1297/1518 + external-summary-compress.ts 五处 [Compression FAILED…])、decompress(decompress-shared.ts 十余处 [decompress FAILED…] / [Block bN not found])、search([search_context FAILED…])、absorb(absorb.ts 三处)、acp_cache(cache-ledger.ts:2437)、retrieve(store.ts:205/219)、异常路径(compress-loop-responses.ts:126 单行 ❌ [ACP] <tool> FAILED: …)。不带关键字的回执(rules 校验失败、unknown tool)前后都走默认图标,不变。
    • 剥离侧 icon-agnostic:MARKER_LINE(tag-echo-filter.ts:409)= /^\p{So}(?:[ \t])?\[ACP\][^\n]*\n?/,匹配任意符号 + [ACP],📦/❌ 都照剥;anchors.ts 形状约定与 golden 契约均未钉死“图标+关键字”组合 → 无误剥风险。
    • acp_status 无文本内失败形态:handleAcpStatus 只返回 toolOk(acp-status.ts:191/591),异常走单行 ❌ 前缀路径 → 统一进首行规则安全;还顺带修掉报告体/块主题提关键字时 📊→❌ 的同族误判。
    • 两种压缩模式下该 marker 都是客户端可见确认、构建路径不分模式 → 无模式相关回归;无 fail-fast 变吞错、无阈值/时序/输出格式漂移。
  6. 是否破坏缓存:不涉及(不动缓存/重试/游标/去重键)。
  7. 是否改设计方向:不涉及,在既有“状态行为准”契约内收敛。
  8. 是否引入新配置:不涉及。

行为变更披露(旧→新,已写入 PR 正文)

  • 图标判定范围:回执所有非空行 → 仅首个非空行。成功回执内嵌内容提关键字:❌→正确工具图标;真实失败不变。
  • acp_read_original_image 三条失败回执措辞:acp_read_original_image: <msg> → [acp_read_original_image FAILED: <msg>](插件模式模型可见文本变化;图标由错画 📦→正确 ❌)。已核实无控制流按该工具结果扫关键字。

测试有效性(④)

新 tests/issue5389-marker-icon-first-line.test.ts 按 #3617 双向纪律、断言有牙:真实 applyRanges 成功压缩且摘要含 FAILED,并带牙断言确认关键字确在非首行(否则用例自废)→ 📦 无 ❌;单元双向矩阵覆盖带嵌入关键字的成功(decompress/search/acp_status)与各真实失败形态(含异常路径单行,输入形状与 compress-loop-responses.ts:126 一致);镜像缺陷对真实 executeReadOriginalImage 断言 […FAILED…]+❌+指引子串保留;剥离兼容验证成功/失败 marker 均按形状剥离。无恒真断言,能兜底本 bug 回归。

diff 整洁度 / 代码质量(⑤/⑤b)

3 文件 +167/-8,单 commit,基于当前 master(e209109bf),无越界改动、无空白/锁文件 churn。注释讲“为什么”(引 #5389/#1029 契约并列出各失败形状)而非复述代码;命名/风格一致;image-rescale 改动逐字保留指引。达标。

我的独立发现(不在本 PR 范围,建议另行跟踪)

审控制流时发现同机制的隐性缺陷:core.ts 有 5 处直接扫 pr.result.includes("FAILED")(覆盖整条回执而非首行)——:1461 & :1572(anyCompressFailed gate hideConsumedCompressCalls)、:1482(fold-refresh gate)、:1687(compress committed 计数)、:1694(absorb committed 计数)。触发条件比展示 bug 窄(只认大写 FAILED,不含 not found/is required):一次成功的 compress/decompress/absorb 若其内容含大写 “FAILED”,会被误判为失败 → 跳过 hide-consumed、跳过折叠视图刷新(下一轮退回压缩前视图)、少计节省量。后果多为单轮自愈/遥测偏差,非数据丢失。

这是 #5389 同一根机制(拿全文做状态分类 vs 首行权威)的另一 facet,但症状不同(控制流 vs 图标)、触发集更窄。没有并入本 PR:本 PR 是纯展示小修(可独立合并),把 5 处 load-bearing 的折叠/缓存信用逻辑塞进来会膨胀 diff、抬高回归面,且需独立的两种压缩模式分析+测试。按 loop-guard(#5389 系自动立项 lineage)我不擅自开新 issue,在此上报请 owner 定夺:并入 #5389 作子项,或按其更窄触发单开。修法方向与 #5389 一致(改判首行即可)。
补充精确化:PR 正文说“control flow never used this path / 用 #5384 分类器看首行”——就 buildVisibilityMarker 不被控制流调用这点属实,但这 5 处是直接 .includes 扫描,属同模式的独立实例,措辞上宜区分。

CI

head 655a3a3:30 项 check,0 failed,6 个 ubuntu test shard 仍在跑;已完成项(static-gates×2、mysql-contract、gate、plan、windows test 全 5 shard)全绿,e2e 类按门控 skipped。等 6 个 shard 收口即全绿。

收工判定

可合(mergeable)——逻辑正确、无真实失败误判、测试有牙、diff 干净;等 6 个 ubuntu shard 转绿即可由 owner 合并。控制流 facet 不阻塞本 PR,建议按上单独跟踪。

@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)

APPROVE — independent review of the full diff (3 files, single commit vs master e209109; no kernel/, no config surface).

What I verified myself (not taken from the issue thread)

  • Every real failure receipt carries its keyword on line 1. Walked all emission sites: [Compression FAILED: …] (src/stream.ts:792/1009/1013/1017/1062/1297/1518), [decompress FAILED: …] / [Block bN not found] (src/decompress-shared.ts:106–600, all sites), [search_context FAILED: …] (:830/:990), [absorb FAILED: …] (src/absorb.ts:89/104/109), acp_cache (src/cache-ledger.ts:2437), retrieve (src/store.ts:205/219 — the multi-line export-failure form has FAILED on L1), and the exception path ❌ [ACP] <tool> FAILED: … (src/compress-loop-responses.ts:126, single line). None regresses to a success icon.
  • The first-line-is-authoritative contract holds for success receipts too: decompress success always leads with the controlled header [Block bN content — N item(s)] (decompress-shared.ts:246/442/597) before any embedded restored body; the compress fingerprint excerpt starts on line 2. So judging line 1 cannot misread a success either.
  • acp_status unification is sound: handleAcpStatus (src/acp-status.ts) has exactly two return paths, both toolOk; zero toolFail/throw in the file — there is no in-text failure shape to preserve, so dropping its all-lines special case loses nothing (and fixes 📊→❌ on report bodies mentioning keywords).
  • Strip side is icon-agnostic: MARKER_LINE (src/loop/tag-echo-filter.ts) matches \p{So} [ACP] for ANY symbol; anchors.ts:16 cites both shapes as prose examples only; golden contracts pin no icon+keyword combo → no mis-strip, no downstream dependency on the old judgment.
  • Control flow untouched by the image-tool wording change: all loop FAILED-scans are name-gated to compress/decompress/absorb (src/loop/core.ts:1461/1482/1572/1687/1694); nothing scans acp_read_original_image results. Display-only, as disclosed.
  • Tests have teeth and cover both directions (#3617): the real applyRanges success case asserts the keyword actually sits in a NON-first line (so the test would go blind if the fixture drifted); the unit matrix covers success-keeps-icon ×4 and failure-stays-❌ ×6 incl. the exception shape; mirror-defect cases assert [… FAILED: …] + preserved guidance substrings; strip-compat asserts removal by shape for both icons.
  • CI on head 655a3a3: mirror cache shows 30 checks, 0 failed, 0 pending (15 success / 15 skipped).

Existing-behavior audit (touched paths)

Behavior Verdict
Failure marker renders ❌ (all tools + exception path) preserved — keyword always on L1
Success marker renders tool icon preserved; false-❌ on keyword-mentioning embedded content eliminated (the fix)
Marker stripping from replayed history unchanged — regex is icon-agnostic
visibilityMarkers tri-state gating / dedup (seenMarkers) untouched
Model-visible text of acp_read_original_image failures INTENTIONAL change, disclosed in PR body: acp_read_original_image: <msg> → [acp_read_original_image FAILED: <msg>]
No-keyword failures ([Unknown proxy tool: X], rules-feature validation) render tool icon not ❌ pre-existing, unchanged by this PR — noted, not a regression

No fail-fast→swallow, no threshold/timing/output-format drift beyond the two disclosed items.

结构化八问

  1. 是什么问题:成功的压缩/decompress/search/acp_status,只要回执第 2 行起的内嵌模型内容(摘要摘录、恢复正文、证据摘录、块主题)提到 FAILED / not found / is required,确认标记就画成 ❌,与成功文字自相矛盾。
  2. 什么原因导致:buildVisibilityMarker 拿回执全文做状态分类;而全仓回执契约是首个非空行即权威状态行(#1029 注释文档化),第 2 行起是模型自撰内容——拿内容当状态是错层。
  3. 影响范围与严重程度:纯展示层;visibilityMarkers 默认可见的车道(pi/opencode 等)在特定条件下图标画错,Hermes 生产 false 不可见;控制流不受影响。p3。
  4. 怎么修复的:判定只看首个非空行(与 #5384 同原则);acp_status 经核实无文本内失败形态,统一进同一规则;同机制镜像缺陷(image 工具三条失败回执无关键字→失败画 📦)一并对齐到 […] FAILED: … 约定。
  5. 是否有回归问题:逐项审过——全部真实失败回执关键字都在首行(仍 ❌);剥离正则与图标无关;控制流扫描均 name-gated 不碰该工具;既有 ❌ 钉住测试在新判定下全部成立(CI 绿)。无 fail-fast 变吞错、无阈值/时序漂移。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及,收敛于既有「状态行为准」契约。
  8. 是否引入新配置:不涉及。

收工判定:可合并(人工 merge)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b) Review 完成:判定 APPROVE —— 独立核验了全部失败回执产出点(关键字均在首行)、acp_status 无文本内失败形态、剥离正则与图标无关、控制流扫描均 name-gated 不碰 image 工具;测试双向有牙(#3617);CI head 655a3a3 全绿(30 checks,0 failed,镜像缓存读取)。无 kernel/ 改动、无新配置、行为变更已在 PR 正文披露。GitHub 不允许 approve 自己的 PR,故以 COMMENT 态评审发出(https://github.com/ranxianglei/billion-context/pull/5388#pullrequestreview-5724980308),可人工合并。

ci-check added 2 commits November 1, 2026 16:11
…l receipt (#5389)

Review follow-up for this PR's first-line-only marker judgment: executeRule's
multi-operation refusal was the one reachable toolFail site whose first line
carries no keyword, so a real refusal renders the tool icon instead of the
crossed-out one under either judgment scheme. Prefix it with the
[acp_rule FAILED: ...] receipt convention (same shape as the image-rescale
receipts fixed here); update the exact-string constant in tests/rules.test.ts.
@ranxianglei

ranxianglei commented Nov 1, 2026 •

Copy link
Copy Markdown
Owner Author

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

Reviewed this against the live branch (head 655a3a3) rather than from the description:

  1. First-line-only judgment — verified correct. I re-audited every producer feeding buildVisibilityMarker: all failure receipts across compress/decompress/absorb/search_context/acp_status/image-rescale carry a keyword on line 1 (kernel error text contains no literal "FAILED", so partial-fold receipts keep theirs on line 1 too); success receipts whose later lines hold model-authored content (summaries, restored bodies) now render the per-tool icon instead of ❌. The both-direction test file pins exactly this boundary, per the gate-test discipline (Test Blind Spot Retrospective: Reinforcing Negative Assertion Discipline, Tightening Fake Upstream Strictness, Thin Inst | 测试盲区复盘:补反向断言纪律、假上游严格化、薄安装冒烟、形态矩阵与生命周期覆盖 #3617).

  2. Residual-gap follow-up — retracted and reverted. My earlier commit dbd5d64 prefixed the refusal at src/rules-feature.ts:72 with [acp_rule FAILED: …]; that was wrong. The design note directly above that return states validation failures are normal outcomes returned verbatim WITHOUT a FAILED marker ("they tell the model how to fix the input, they are not proxy errors"), and none of the five FAILED consumers name-gate acp_rule anyway, so the receipt never touched control flow. Reverted by commit 297925c (rules + marker suites 23/23 on the pushed head); the branch is back to exactly the author's original behavior.

  3. Numbering mismatch to fix before merge. The body says "Closes [PR] docs: New blog post — What to do when billion-context is frequently compressed? | docs: 新博客文章——billion-context频繁压缩该怎么办? #5389", but GitHub issue [PR] docs: New blog post — What to do when billion-context is frequently compressed? | docs: 新博客文章——billion-context频繁压缩该怎么办? #5389 is a closed blog-docs PR. The marker-icon issue actually fixed here is GitHub Successful compression renders the wrong ❌ confirmation icon when its summary text contains FAILED #5387 (its mirror thread is numbered 5389 — same title/timestamp, verified via API). As written, the merge closes nothing; change the reference to Successful compression renders the wrong ❌ confirmation icon when its summary text contains FAILED #5387.

Composition check: #5384 + this PR merge cleanly together and pass 46/46 across plugin-protocol + marker + rules suites on the composed tree — the two halves are independent and merge order is free.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Review 收工确认(完整审核已以 COMMENT 态评审在档:pull/5388#pullrequestreview-5724980308,GitHub 禁止 approve 自己的 PR):

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Follow-up 三项逐条处置完毕:

  1. 配对确认:收到——两 PR 同机制族互补、各自自洽可绿,合并顺序自由,无异议。

  2. dbd5d64 → 撤回:该 commit 落地时我独立核验过(diff + 审计 rules-feature 其余 toolFail 点均为 kernel validation、按设计不带关键字)。随后看到 PR 线程里的撤回说明与 revert commit 297925cd0,认同撤回理由:rules-feature 的回执按文件自身设计注记就是「输入形状反馈、非 proxy 错误、刻意不带 FAILED 标记」,且五处 FAILED 消费方均 name-gated 到 compress/decompress/absorb,本就不触碰控制流——撤回后该回执渲染中性图标是有意设计而非残留缺陷,此项关闭。我侧的连带清理:① 我准备的一个 docs-comment 对齐 commit(把设计注记改成「多操作拒绝带 FAILED 约定」)已丢弃——revert 之后原注记重新与代码一致,那个 commit 反而会写错事实;② PR 正文里我曾为 dbd5d64 补的第 3 条行为变更披露已删除,正文现与最终 diff(3 文件 +167/-8)严格对应。

  3. Closes 引用已修正:PR 正文首行 Closes #5389 → Closes #5387(已核实 GitHub Successful compression renders the wrong ❌ confirmation icon when its summary text contains FAILED #5387 = open 的 marker-icon 缺陷 issue;[PR] docs: New blog post — What to do when billion-context is frequently compressed? | docs: 新博客文章——billion-context频繁压缩该怎么办? #5389 = closed 的 blog-docs PR)。合并即自动关闭 Successful compression renders the wrong ❌ confirmation icon when its summary text contains FAILED #5387。

CI 状态(final head 297925cd0):30 checks 中 29 条绿/跳过,唯一红条 test (ubuntu-latest, 22, 1) 失败用例仅一条——tests/issue2129-restart-measurement-state.test.ts「PFA stamp must be on disk after restart」,即 #4599 在档的 #2129 restart-e2e 低频负载 flake(TAP pass 3090 / fail 1),与本 PR diff 零交集。已触发 rerun-failed-jobs(run 41288910870,HTTP 201)并在 #4599 记档。

判定维持 APPROVE(GitHub 禁止 approve 自己的 PR,完整审核以 COMMENT 态评审在档:pull/5388#pullrequestreview-5724980308)。rerun 转绿后即可人工合并。

ci-check added 2 commits November 11, 2026 16:19
The confirmation-marker icon was chosen by scanning receipt TEXT for
FAILED / "not found" / "is required" — a heuristic over model-influenced
bytes. Success receipts embed model-authored content (summary fingerprint
excerpts, restored history, evidence snippets, report topics) where those
words appear as ordinary prose, and some failure wordings carry no keyword
at all ([Compression REFUSED: …], pre-fix acp_read_original_image).

The judgment is now protocolized: the executor's structured outcome
(#1875 ProxyToolResult.outcome) is authoritative — failure/refused → ❌,
applied/partial/success → the tool icon — threaded through every marker
call site (loop/core.ts streaming + re-request paths, all five wire
adapters' emitMarker incl. commandcode, the responses lane incl. its
exception path). Receipt text is scanned only as a fallback when outcome
is absent (first non-empty line only; no executor today omits it).

Closes #5387

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

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Successful compression renders the wrong ❌ confirmation icon when its summary text contains FAILED

1 participant