Skip to content

fix(i18n): word three glued channel and agent sentences as one message - #31698

Merged
RohanK6 merged 1 commit into
mainfrom
crew/conductor/r19-separator-joins
Nov 7, 2026
Merged

RohanK6 merged 1 commit into
mainfrom
crew/conductor/r19-separator-joins

Conversation

@RohanK6

@RohanK6 RohanK6 commented Nov 6, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: The channel clear note, the agent backend could-not-check note and the unpinned model badge title each read as one translated message.

Three surfaces build one sentence from two catalog pieces joined in code with an English space or : :

  • ChannelPage.tsx: clear_context_done + space + clear_context_messages_deleted
  • AgentBackendTab.tsx (two sites): install_check_failed + space + can_still_switch
  • AgentTemplatesTab.tsx: model + : + model_auto (the badge title)

A language cannot word these sentences itself, and Japanese and Chinese read ASCII separators between full-width text.

Why it matters

These notes tell a person what a clear removed and why a switch button is still live. Stray English spacing in CJK reads as a broken string, and translators cannot reorder the pieces.

Not a goal

What changed (motivation → approach → change)

Symptom: ASCII separators between full-width sentences. Cause: two catalog pieces joined in code. Change: each site renders one whole catalog message.

New key en ja
pages.channelPage.clear_context_done_messages_deleted Context cleared. The channel's messages were deleted too. コンテキストを消去しました。チャネルのメッセージも削除されました。
pages.developer.agentBackendTab.install_check_failed_can_still_switch the two English sentences, as before joined with no space
pages.overview.agentTemplatesTab.model_auto_title Model: auto (backend default) モデル:auto(バックエンドのデフォルト)

Each locale value reuses that locale's own two pieces, so nothing is retranslated: ja and zh-CN join with no separator and use a full-width colon, and every other language keeps its space and : . English lives in en.manual.json, and en-XA.json is regenerated. The two fragment keys left with no reader, clear_context_messages_deleted and can_still_switch, are removed from all catalogs. clear_context_done, install_check_failed, model and model_auto keep their other readers.

The agent backend detail strip shows this note for a failed install probe, which errors-use-error-notice counts as an error. Because this PR edits that site, it now renders through ErrorNotice (inline, with no agent hand-off: the same detail can hold the Strands setup card, whose model and sign-in drafts are not saved yet, so a No hand-off comment names them). The row status line keeps the same sentence as plain status text.

Backwards compatibility

Compatible: only displayed text changes in ja and zh-CN; English and every spaced language read byte-identically, and no API, route, setting or stored value is touched.

Reader: website/src/pages/ChannelPage.tsx:clear acknowledgment -- entry: crew page -- the only reader of clear_context_messages_deleted, now reads the whole-message key (ChannelPage.clearContext tests)
Reader: website/src/pages/developer/AgentBackendTab.tsx:install-unknown note -- entry: crew page -- both readers of can_still_switch now read the whole-message key (AgentBackendTab tests)

Tests

  • ChannelPage.clearContext.test.tsx: the English deletion note reads as before, and the Japanese one runs on with no space. The Japanese test fails with origin/main and passes with this change.
  • AgentBackendTab.test.tsx: the Japanese could-not-check note reads as one sentence, inside an ErrorNotice alert with no hand-off button. It fails with origin/main and passes with this change. The existing English assertions still pass.
  • AgentTemplatesTab.test.tsx: the unpinned badge title reads Model: auto (backend default) in English and モデル:auto(バックエンドのデフォルト) in Japanese. The Japanese test fails with origin/main.

Proof: not provable -- frontend test (prove.py runs Python only)

Determinism: each test switches i18next language inside act and restores English in a finally; no clock, timer or network. tsc, eslint, i18n:check and the whole src/i18n suite including the per-locale style tests (both with I18N_BASE_REF=origin/main; the French value uses U+202F before its colon, as style/fr.md requires), and every ChannelPage, AgentBackendTab and AgentTemplatesTab test file (66 files, 1350 tests) pass. Local GPT review ran on gpt-5.6-sol (the brief pins gpt-6.1-sol, which was not available) and Opus 5.5: no findings.

Manual verification

N/A -- unit coverage sufficient: the change is which catalog key renders, and each rendered Japanese string is asserted through the real component.

Screenshots / video

Why no screenshot: the English text is unchanged and the capture host has no CJK font, so a screenshot would show no visible difference; the Japanese text is asserted by the tests.

Related Issues

Fixes #31669
Deferred: #31685

Sibling sweep

Swept: line scan of website/src for two i18nT calls joined by a literal space, colon or {" "}
Sibling: website/src/pages/ChannelPage.tsx -- changed here
Sibling: website/src/pages/developer/AgentBackendTab.tsx -- changed here
Sibling: website/src/pages/overview/AgentTemplatesTab.tsx -- changed here
Sibling: website/src/pages/settings/AboutPanel.tsx -- tracked in #31208
Sibling: website/src/pages/settings/RemoteCrewPanelLaunch.tsx -- tracked in #31208
Sibling: website/src/apps/ops-mission-control/SignalsPanel.tsx -- tracked in #31685
Sibling: website/src/apps/ops-mission-control/HandoverPanel.tsx -- tracked in #31685
Sibling: website/src/apps/auto-research/ResearchLabPage.tsx -- tracked in #31685
Sibling: website/src/pages/settings/McpManagementHero.tsx -- tracked in #31685

Pattern harvest

Rule candidate: lint
Pattern: two i18nT() calls joined by a literal separator in a template or JSX, which the catalog cannot word as one sentence.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

Contribution License Agreement

Contributed under the project Apache-2.0 license.

@RohanK6
RohanK6 requested review from a team and KiroCrewCaptain as code owners November 6, 2026 19:21
@RohanK6
RohanK6 requested a review from Zedmor November 6, 2026 19:21
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Nov 6, 2026
@github-actions

github-actions Bot commented Nov 6, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5) — ✅ PASS (a BLOCK with no Trigger and no Outcome does not block)

Design-level review of 1c1b49e1dfb3248480c5222ea98ec212a07d63c1 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

cannot evaluate: missing screenshot of the agent backend detail strip's could-not-check note. The note changes from neutral text to a red ErrorNotice beside a live Use button, and no rendered evidence was attached.

Blockers

The could-not-check note is restyled as an error, and there is no screenshot of it
Trigger: a backend probe returns installed === 'unknown', and the detail strip now renders <ErrorNotice message={…install_check_failed_can_still_switch} variant="inline" /> where it used to render <p className="text-[13px] … text-text">.
Outcome: the sentence becomes 12px text-danger with an AlertTriangle icon → this goes against the file's own recorded UX reasoning ("a bright Use button under a 'could not check' line reads as a mistake the reader refuses to touch") → users may stop trusting a switch the design deliberately leaves live. The evidence list has 0 images and 0 recordings, and the "Why no screenshot" note ("the English text is unchanged") skips the style change, so I could not judge the shape.
Fix: do the ErrorNotice migration in its own PR with a capture, or keep the <p> here and render only the merged key. That would match this PR's i18n-only goal.
Clears when: the description carries a screenshot of THIS revision showing the unknown-probe detail strip with its Use button, or the ErrorNotice change is reverted out of this PR.

Notes

  • The ErrorNotice restyle is outside the stated goal ("each site renders one whole catalog message") and is the only part of the diff a person sees in English.

[DESIGN-REVIEWED] 1c1b49e

Description read

[DESCRIPTION-READ] fe88f7fb8a7173efeb2c580fb1e6c7fc6320cb543e65011b51b578e467624525

That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Nov 6, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5) — ✅ PASS (a BLOCK with no Trigger and no Outcome does not block)

UX-level review of 1c1b49e1dfb3248480c5222ea98ec212a07d63c1 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

UX-Verdict: PASS

cannot evaluate: missing a screenshot of the agent backend "could not check" detail, which now renders in red with a warning icon. The PR says nothing visible changes, but in English this note goes from neutral 13px text to the text-danger inline ErrorNotice.

Evidence gaps

  • Agent backend detail strip, installed === 'unknown' state (AgentBackendTab.tsx): the plain <p className="text-[13px] … text-text"> becomes <ErrorNotice variant="inline">, which renders at 12px in text-danger with an AlertTriangle icon. The English words stay the same, but the colour, size and icon are new in every language. The PR says "the English text is unchanged… a screenshot would show no visible difference", and that is not true at this site. The blind read was not performed because the PR supplies no images. To close the gap, attach an English screenshot of Developer → Agent backend with a backend whose install check returned unknown and is highlighted, showing the detail note next to its Use button.

Notes

  • The file's own comment above this branch warns that a "could not check" line under a bright Use button "reads as a mistake the reader refuses to touch". Turning that line red with a warning triangle makes this more likely, not less. Check the screenshot for this. If it reads as alarming, render the note as the neutral status text it was before, or use a warning tone, rather than the danger style.

[UX-REVIEWED] 1c1b49e

Description read

[DESCRIPTION-READ] 11a87bdf25536cadb73ab3ac8d5cd2bb64d72c58e92e1dff616b3ae46b805f1e

That sha256 names everything this verdict read: the PR description, plus the 3 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Nov 6, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5) — ✅ PASS

Premise-level review of 1c1b49e1dfb3248480c5222ea98ec212a07d63c1 — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

First-Principles-Verdict: PASS

Nothing blocks. Each new key has its reader, and the two removed fragment keys have no other reader. The ErrorNotice switch is required by website/AGENTS.md ("Errors shown to the user render through ErrorNotice").

What this change ships

Inventory (6 items) — 6 justified

Intent: CJK readers see the channel-clear note, the agent backend could-not-check note and the unpinned model badge title as one natural sentence. This is a FIX.

  1. The channel-clear note now comes from one whole-sentence catalog entry (ja/zh-CN have no stray space) — justified
  2. The agent backend row status line now reads the could-not-check note as one message — justified
  3. The agent backend detail strip shows the could-not-check note as an inline ErrorNotice alert with no hand-off — justified
  4. The unpinned model badge title comes from one message and uses a full-width colon in ja/zh-CN — justified
  5. The fragment keys clear_context_messages_deleted and can_still_switch are removed from every catalog (0 remaining readers in website/src) — justified
  6. New tests cover the English and Japanese wording at all three sites — justified

[FIRST-PRINCIPLES-REVIEWED] 1c1b49e

Description read

[DESCRIPTION-READ] 66f113e7488f3a5a37c759a8699170021dd52a5ef2facbb91fcfb62984f83638

That sha256 names the PR description this verdict read. Recompute it with .github/scripts/pr-description-capture.sh; if it differs, the description changed after this verdict was formed and any finding drawn from the description is unproven.

@github-actions

github-actions Bot commented Nov 6, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed 1c1b49e1dfb3248480c5222ea98ec212a07d63c1 — this comment is updated in place on each push.

Review details

I found nothing blocking. Neither of the two candidate findings held up when I checked the code.

  • The missing "Ask the agent" hand-off on a failed install check: the new ErrorNotice carries a No hand-off comment, and the comment names a real draft. The Strands setup card renders in the same detail. The errors-use-error-notice rule tells reviewers to block only when that decision is missing, and never to overrule which way it went.
  • The new alert being announced on plain row navigation: the rule says a failed check is an error and must render through ErrorNotice. The candidate named no wrong outcome beyond that intended styling, so this is a design preference, not a defect.

I also confirmed that the three new translation keys exist where the code looks them up.

No findings.

[OPUS-REVIEWED] 1c1b49e

Verdict parsed from the review's SHA-scoped output markers for commit 1c1b49e1dfb3248480c5222ea98ec212a07d63c1.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 1c1b49e1dfb3248480c5222ea98ec212a07d63c1 false-positive: <reason, 20 to 500 characters>

@github-actions

github-actions Bot commented Nov 6, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of 1c1b49e1dfb3248480c5222ea98ec212a07d63c1 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 1c1b49e

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 1c1b49e1dfb3248480c5222ea98ec212a07d63c1 false-positive: <reason, 20 to 500 characters>

@RohanK6
RohanK6 force-pushed the crew/conductor/r19-separator-joins branch 3 times, most recently from 20fa981 to e48344f Compare November 6, 2026 22:33
@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Nov 7, 2026
The channel clear acknowledgment, the agent backend could-not-check
note and the unpinned model badge title each joined two catalog pieces
in code with an English space or colon. Translators could not word
them as one sentence, and Japanese and Chinese read ASCII separators
between full-width text. Each is now one catalog message that reuses
the locale's own pieces, so English reads exactly as before; the two
fragment keys left with no reader are removed.
@RohanK6
RohanK6 force-pushed the crew/conductor/r19-separator-joins branch from e48344f to 1c1b49e Compare November 7, 2026 01:29
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Nov 7, 2026
@RohanK6
RohanK6 enabled auto-merge (squash) November 7, 2026 03:31

@bolichen97 bolichen97 left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

Approving head 1c1b49e. The gate shows 99 checks passed. The 8 red entries are cancelled duplicate runs, and each of those checks has a newer successful run on this head.

  • Three glued sentences become one key each: clear_context_done_messages_deleted, install_check_failed_can_still_switch and model_auto_title. A locale can now punctuate the whole sentence, and the Japanese tests pin that no English space or colon is left. The old keys can_still_switch and clear_context_messages_deleted are removed with their last callers, and every locale file in the diff gets the new keys.
  • One visible change: a failed install probe now shows through ErrorNotice (inline, role="alert") instead of a plain paragraph. The comment and the test say why it has no hand-off button: that detail can hold unsaved Strands setup drafts.

No product regex (the one pattern is in a test), and no new restriction.

@RohanK6
RohanK6 merged commit e68ca0a into main Nov 7, 2026
158 of 168 checks passed
@RohanK6
RohanK6 deleted the crew/conductor/r19-separator-joins branch November 7, 2026 06:31
@github-actions github-actions Bot removed the readiness: passed Eligible automated validation passed for the current revision label Nov 7, 2026
buluoray pushed a commit that referenced this pull request Nov 8, 2026
#31698)

The channel clear acknowledgment, the agent backend could-not-check
note and the unpinned model badge title each joined two catalog pieces
in code with an English space or colon. Translators could not word
them as one sentence, and Japanese and Chinese read ASCII separators
between full-width text. Each is now one catalog message that reuses
the locale's own pieces, so English reads exactly as before; the two
fragment keys left with no reader are removed.

(cherry picked from commit e68ca0a)
buluoray pushed a commit that referenced this pull request Nov 8, 2026
#31698)

The channel clear acknowledgment, the agent backend could-not-check
note and the unpinned model badge title each joined two catalog pieces
in code with an English space or colon. Translators could not word
them as one sentence, and Japanese and Chinese read ASCII separators
between full-width text. Each is now one catalog message that reuses
the locale's own pieces, so English reads exactly as before; the two
fragment keys left with no reader are removed.

(cherry picked from commit e68ca0a)
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.

Channel, agent backend and template pages join translated pieces with English separators

2 participants