Skip to content

fix: backport 0.12.1 fast batch v1b-2 (22 user-facing fixes) - #32864

Merged
buluoray merged 22 commits into
release/0.12.0from
crew/slack/0.12.1-fast-v1b-2
Nov 8, 2026
Merged

buluoray merged 22 commits into
release/0.12.0from
crew/slack/0.12.1-fast-v1b-2

Conversation

@buluoray

@buluoray buluoray commented Nov 8, 2026

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: Bring twenty-two fixes for problems a 0.12.0 user can hit, merged on main after v0.12.0, onto release/0.12.0, one commit each, in a PR of its own.

Why it matters

Each fix changes behaviour a 0.12.0 user can run into today. Every pick applied to release/0.12.0 cleanly, so the code each one replaces is on release.

PR Area What a 0.12.0 user sees today
#26371 apps Git clone auth refusal not recognised for some method lists
#30077 doctor Doctor passes when the selected Claude harness misses a component
#30493 a11y Personal Shopper text fields have no accessible names
#30835 sidebar In progress filter misses the open session's idle goal loop
#30941 dashboard Confusing 403 message for app-token requests
#31006 deploy Deploy confirmation reply does not say when it expires
#31244 i18n context breakdown sizes pluralized instead of a glued unit
#31493 apps registry trust banner reads as one translated message
#31325 shortcuts Alt+C/N/P/S panel rows explain they skip text fields
#31687 issue radar keyboard focus stays on refresh/retry buttons while busy
#32043 i18n Issue Radar settings names stay inside one translated sentence
#31471 UI Picking Overview in settings rail leaves the drill-in view
#32052 a11y focus stays in browser settings chip lists after removing one
#31884 a11y hover-only row controls appear on keyboard focus
#32292 chat blocked-link card shows allowed state everywhere after Allow
#32274 a11y focus stays on crew board Stop while request runs
#32186 a11y focus stays on memory retry buttons while they load
#32410 a11y/crewmate Focus lost on crewmate dashboard Retry while it runs
#31800 settings Clicking "More below" does not scroll the list
#32456 i18n Phased board column headings stay English
#32436 a11y/apps Focus lost when an app page reload Retry fails again
#32508 subagents Queued sub-agent chip keeps its wait reason after a page reload

Not a goal

  • No fix that is not already merged on main; nothing is changed only here.
  • No version bump, no tag, no CHANGELOG.md change.
  • Rows whose pick conflicts; they go in a later batch.

What changed (motivation → approach → change)

Each main squash commit was cherry-picked with -x onto release/0.12.0 @ aa7dfb4aae, in merge order on main, one commit each. Every pick applied cleanly, and every commit keeps its original author.

# PR main commit release commit
1 #26371 treat any SSH method list as a clone auth refusal ce40cec9c9 7f3dd2d1ed
2 #30077 fail when the selected claude harness is missing a component 857e6b3d16 635f8f9cd9
3 #30493 name the Personal Shopper text fields 888c8c6fea e648d9281b
4 #30835 count the open session's idle goal loop in the In progress filter 4efadc110c 3a11eff8b4
5 #30941 say no permission lets an app token make these requests 44166e6c74 47252d024b
6 #31006 state when a pending deploy confirmation expires ad90ef699d 54517d83e4
7 #31244 pluralize context breakdown sizes instead of gluing a unit a95ac0555d 7cb715654a
8 #31493 word the registry trust banner as one message 9597d17c6c 668f107388
9 #31325 name the text-field rule on the Alt+C/N/P/S panel rows af8341daf6 489afa3c69
10 #31687 keep focus on the remaining refresh and retry buttons while busy cd168b4a4b 43b0d44851
11 #32043 keep names inside one sentence on issue radar settings c814b3a0f6 ab231cbe30
12 #31471 leave the overview drill-in when the settings rail picks overview bb189b368a 24b4e122ee
13 #32052 keep focus in the browser settings chip lists after a remove f9c2082968 4af0bb3a0e
14 #31884 show hover-only row controls when the keyboard reaches them 6b1c426f2c a277cdc552
15 #32292 reflect allow across the whole blocked-link card e3a289cd4b 2459dc6ef9
16 #32274 keep focus on crew board stop while its request is out 2180fea484 8e6dce34e4
17 #32186 keep focus on memory retry buttons while they read 4ceb5659c6 9608cee00e
18 #32410 keep focus on crewmate dashboard retry while it runs 37f53e446b 15128e5bd5
19 #31800 scroll the list when "More below" is clicked b0ffbf69ac b3a573d8a7
20 #32456 translate the phased board's column headings 5c5c67735a 490cf0a885
21 #32436 put focus back on Retry when an app page reload fails again 7a2738a1b7 1cb47ed817
22 #32508 keep the queued sub-agent wait reason across a reload 64fade4229 36682e97fb

Three picks carried co-author trailers with internal addresses; each is rewritten as a plain "Co-author on main: " line, because the commit guard refuses the address.

Left out of this batch: #30189 and #30110 now conflict, in a test file and a spec doc only, so they go to the docs/tests batch. #32417 picks cleanly but needs a useEffect import that release does not have, so tsc fails; it stays on main only.

Backwards compatibility

Compatible: each change is the one already merged on main, and its compatibility note applies unchanged.

Removes nothing: no setting, key or caller is removed.

Tests

Proof: not provable -- other: a backport of main commits whose proofs ran on their own PRs

  • CI does not run the backend suite on PRs into release/*, so the direct test modules were run here: the 12 Python test files these picks touch gave 451 passed.
  • Frontend: tsc -p tsconfig.app.json --noEmit passes; the 25 vitest files these picks touch gave 446 passed; lint:i18n passes at the same count as clean release.
  • flake8, isort, the black gate and mypy src/kiro_crew pass. i18n:check without a base reports the same findings class as a clean release/0.12.0 on this host, and fewer of them.

Manual verification

N/A - each fix was verified on its own main PR. The 0.12.1 RC soak is the end-to-end check.

Screenshots / video

Why no screenshot: the frontend picks carry the same rendered change as their main PRs, which show it.

Related Issues

Backport to the 0.12 line for 0.12.1 of #26371, #30077, #30493, #30835, #30941, #31006, #31244, #31493, #31325, #31687, #32043, #31471, #32052, #31884, #32292, #32274, #32186, #32410, #31800, #32456, #32436, #32508.

Sibling sweep

Siblings: none -- each source PR carried its own sibling sweep; this PR only carries those fixes.

Pattern harvest

Not generalizable: this PR only carries fixes already on main. The source PRs have their own harvest.

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

Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.

mbriones98 and others added 22 commits November 8, 2026 02:50
A credential-free app clone that an SSH server refuses got the
credentials explanation only when the refusal read
"Permission denied (publickey...". OpenSSH prints the server's own
method list in the server's order, so a server answering
"Permission denied (keyboard-interactive,publickey)." got a bare
"git clone failed".

The SSH refusal is now recognized by a small character-level scanner:
after "permission denied (", the text must be a comma-separated list of
method names that start with a-z and use only a-z, 0-9 and "@._+-",
closed by ")". The scanner reads the list once and stops at the first
character that cannot continue it, so its work stays linear in the
length of git's output. A local filesystem "Permission denied" has no
such list, so it still does not count as an auth failure. The same
classifier decides when an anonymous registry clone is shown as needing
sign-in, so that state now follows any method order too.

Fixes #26365

## Pattern harvest

Rule candidate: review-prompt -- matching an SSH refusal by a fixed
prefix of its method list, which the server orders as it likes.

Co-author on main: Matt Briones
(cherry picked from commit ce40cec)
…nent (#30077)

kirocrew doctor printed the Claude Code row as an optional backend
whatever the selection, so with agent.acp_backend or
agent.member_acp_backend set to claude and claude-agent-acp or the
claude CLI absent it reported ready and exited 0 while every session
on that role failed to spawn.

The row now takes the issues list and a required flag, decided by a
positive test on the two selected roles. Required and MISSING prints
a failing row naming the components and records an issue; unselected,
the row keeps its optional wording. UNKNOWN or a raising probe is
never an issue. The config is loaded once before the Dependencies
section so the row and the later sections read the same selection.

Closes #30050

(cherry picked from commit 857e6b3)
…ess filter (#30835)

The sidebar "In progress" filter and the board's state lanes derive goal-loop
membership from the Redux automation collection
(selectSidebarAutomationRunningKeys). The cold seed (seedAutomations) is the one
place that feeds that collection on every connect: it fetches the authoritative
loop list and reconciles it against live frames and per-slot cold reads.

A per-slot read in flight when the snapshot lands used to "protect" its slot
purely on a fetch-state change, with no resolved data behind it. That dropped
the slot's authoritative list record from the seeded records AND left the
carry-across nothing to carry (the per-slot cache held no record yet), so the
loop never reached Redux. The open session hits this on every connect, because
its composer re-reads the per-slot key: its armed loop fell out of the
In-progress count and list while the composer above it still showed the loop.

Protect a slot from the snapshot only when its per-slot read has actually
RESOLVED (a record or a null tombstone), never while it is still pending. A
pending read is not fresher than the list, so the list record now lands. This
changes only whether the open slot's already-armed loop is present in the set;
every other reader (composer, board) and every other reconcile path is
unchanged.

Co-authored-by: kirocrew-agent <agent@kirocrew.local>
(cherry picked from commit 4efadc1)
…ts (#30941)

Four requests are refused to every app token whatever its app.json
declares: the question-card routes under /api/ask-question, the mobile
sign-in link, dismissing a subagent run the gateway no longer tracks,
and stopping all of a chat's subagents. Their 403 said "app token not
permitted for this endpoint", "app token not allowed" or just
"app_token_forbidden". The first is the same text as the API scope
refusal, which a permissions.api entry clears, so an app developer
would add an entry that can never help.

app_token_refusal builds the 403 for all four. error names the action,
says no permission in app.json lets an app token do it, and says who
can do it instead. For question cards it points an app at its own
session's /api/chat/slots/{slot}/question routes, which need
permissions.sessionApproval and a permissions.api entry. The code
stays app_token_forbidden.

(cherry picked from commit 44166e6)
GET /api/deploy/pending now returns expires_at_epoch on each entry
(created_at_epoch + 1 hour, derived, not stored), and both deploy_artifact
replies (preview and scan-blocked) say to confirm within 1 hour, by a UTC
time, worded apart from the site TTL. The 1-hour window is unchanged.

Part of #17022

Co-author on main: Zejiang Guo
(cherry picked from commit ad90ef6)
…#31244)

The context breakdown shows sizes as a formatted number next to the
single-form unit_chars or unit_tokens word: the start-row picker, the
engine rows and chars caption, and the hover readout. One therefore
read "1 characters" or "1 tokens", and no translation could choose
a plural form or place the number.

Read each size from one plural key (n_chars, n_tokens) with the
formatted number inside and the plural chosen from the raw count, the
pattern WorldsPage already uses. unit_chars and unit_tokens stay for
the caption under the large total, which shows no number of its own.

(cherry picked from commit a95ac05)
The banner shown after adding a trusted registry was five catalog
pieces joined in code with English spaces. No language could order
the sentence its own way, and Japanese and Simplified Chinese got
ASCII spaces between full-width text. It is now one catalog message
per host count, with the host list as a placeholder, in all shipped
catalogs. English reads the same.

(cherry picked from commit 9597d17)
#31325)

The handler returns before the panel map while a text field has focus, so
Alt+C/N/P/S do nothing from the composer and the shortcuts list did not say
so. Add "(outside text fields)" to the four panel row values in every shipped
catalog, the same wording the Alt+arrow and prompt-step rows use. Handler
behaviour is unchanged.

Fixes #28422

Co-author on main: Jacob Largent
(cherry picked from commit af8341d)
…2043)

The General and repo settings intros built a sentence from two catalog
fragments with the provider or repo name glued between them, so no
language could reorder the sentence and zh-CN spelled GitHub into its
fragment. Each is now one key with {provider} and {repo} placeholders,
and the four fragment keys are gone.

(cherry picked from commit c814b3a)
#31471)

A rail selection lands on a tab's root, but Overview keeps its drill-in
in ?view=/?store=/?focus=, which a tab change copied along, so clicking
Overview from the Memory view kept the Memory view. A tab can now
declare its drill-in params and every rail selection drops them; a
re-click that drops the shown tab's drill-in asks the pane's leave
guard first, since it now unmounts that pane.

(cherry picked from commit bb189b3)
…ve (#32052)

Removing a sign-in domain or a local origin chip in Settings > Browser
unmounts the focused remove button with it, so focus fell to the page.
Focus now moves to the editor's add field.

(cherry picked from commit f9c2082)
#31884)

Four controls are drawn only on mouse hover (opacity-0 with
group-hover:opacity-100): the copy-command button on an app's detail
page, the delete-key button on Settings > Developer > Storage, a
Library folder row's menu, and the edit-theme pencil. Tab still lands
on each one, so a keyboard user focuses something they cannot see.
Each now also shows on keyboard focus, the way the session and history
rows already do: focus-visible on a button, has-[:focus-visible] on a
wrapper that holds one. A source guard keeps any new hover-only control
of this shape from shipping without the focus reveal; it skips an
adopted app's wheel copy, whose fix lands in the app's own repository.

(cherry picked from commit 6b1c426)
After Allow for this host from one entry of a blocked-link card, the
other entries kept a live Allow button and the card kept its "blocked"
title, and a reply with no message time lost its Allowed / Undo line
when the card closed.

The card now reads the host's one allow outcome: no entry offers Allow
once the host is allowed, and the title says the host is allowed. A
reply with no message time keeps that outcome in its own provider, so
it survives the card closing and reopening. Undo's clear-out timer
clears only the Undone line, so a quick re-Allow keeps its Undo.

(cherry picked from commit e3a289c)
Stop and Take over took native disabled while the action request was
pending, which blurs the pressed button to the body. A keyboard user
lost their place before the failure notice rendered. Hold them with
aria-disabled and refuse the second press in the handler instead.

Fixes #32272

(cherry picked from commit 2180fea)
The Retry buttons on the Overview memory cards lost keyboard focus as
soon as they were pressed. Each one was disabled while its read ran,
and each rendered only while the query held an error, which React Query
clears when a query with no data refetches. So the pressed button left
the tab order and then unmounted for the whole retry.

Keep the failure surface up through the retry with a small
queryReadFailed helper (the rule CreatorFilter already applies), and
mark the busy state with aria-disabled plus a guard in the click
handler, as the Apps store Refresh does.

(cherry picked from commit 4ceb565)
The four Try again buttons in the Dashboard tab error bands took
native disabled while their re-ask was out. Each band stays on screen,
so the pressed button blurred to the body and a keyboard user lost
their place. Hold them with aria-disabled and refuse a second press in
the handler instead.

Fixes #32406

(cherry picked from commit 37f53e4)
The project board's three column headings ("To do", "In progress",
"Done") were English literals in a module-level table, so they stayed
English in every language while the Failed, Skipped and Cancelled
groups under them were translated.

Each heading now reads a catalog key through a getter, so it follows
the active language and re-resolves after a switch. The three keys
are added to all twelve catalogs and the pseudolocale.

Fixes #32440

(cherry picked from commit 5c5c677)
#32436)

Retry on a failed app page swaps the view for the loading line, which unmounts the focused button. When the retry failed too, focus was left on the page body. Return it to the new Retry button; a successful retry still shows the app as before.

(cherry picked from commit 7a2738a)
…#32508)

Co-authored-by: w <w@k.local>
(cherry picked from commit 64fade4)
@buluoray
buluoray requested review from a team and KiroCrewCaptain as code owners November 8, 2026 03:05
@buluoray
buluoray requested review from tim-finnigan and removed request for a team November 8, 2026 03:05
@buluoray
buluoray merged commit 8e72744 into release/0.12.0 Nov 8, 2026
11 of 23 checks passed
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.

8 participants