Repository navigation
fix: backport 0.12.1 fast batch v1b-1 (23 user-facing fixes) - #32840
Conversation
…is (#30152) The cron store documents ValueError for every refused field, and every_secs already maps int(float('inf'))'s OverflowError to it. The timeout and timeout_secs checks on add (validate_create_timeouts) and update (apply_job_update) caught ValueError and TypeError only, and the app cron SDK converted both with int() before the store saw them, so an infinite value raised OverflowError instead. One helper, fields.whole_seconds, now reads both fields on all three paths and refuses an unreadable value with ValueError. Nothing is written either way. (cherry picked from commit 66d2f06)
…0069) The Search Everywhere palette is aria-modal, but keys pressed in it that its own capture-phase listeners do not handle bubbled to the page's document shortcut listeners, so Alt+, or Ctrl+1 acted behind it. The panel now stops every key except the quick-search gesture, whose window listener toggles the palette shut, as the Command Bar does since #29718. The toggle test the Command Bar used moves to lib/quickSearchShortcut so both overlays share one copy. Fixes #29998 ## Pattern harvest Rule candidate: review-prompt Pattern: an aria-modal overlay that lets non-Escape keys bubble to document-level shortcut listeners Co-author on main: Matt Briones (cherry picked from commit 3736af2)
…exits before reading stdin (#30436) When a worker with a broken harness install writes its {"id": null, "error": "import failed ..."} reply and exits without reading stdin, the supervisor's drain() of the request can lose the race and raise ConnectionResetError. The except branch then returned "helper worker is gone" without ever reading the reply the worker already wrote. Read stdout after a write failure and surface that reply; fall back to "helper worker is gone" only when stdout has nothing. Closes #30253. Co-authored-by: kirocrew-agent <agent@kirocrew.local> (cherry picked from commit 360372b)
(cherry picked from commit 34e8a26)
UX Review (Opus 5.5) — ✅ PASS (a BLOCK with no Trigger and no Outcome does not block)UX-level review of UX-Verdict: PASS cannot evaluate: missing screenshots for every changed control. The blind read was not performed because this revision attaches and commits no image. Pointing to the source PRs on Evidence gaps
[UX-REVIEWED] 49a8cfa Description read[DESCRIPTION-READ] 514c285a6340f80dc2bca6614398cf5028903cbc1a8dca72ad589cfa5e1a9310 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 |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Every item is a pick of a fix already merged on What this change shipsInventory (10 items) — 10 justifiedIntent: give 0.12.1 users the user-facing fixes
More than 10 items: also #30069, #30436, #30615, #31166, #31297 and #31728, all listed in the description. [FIRST-PRINCIPLES-REVIEWED] 49a8cfa Description read[DESCRIPTION-READ] 36fca9b037411bad2b005a2292a224629291c5c681c4899618afe26261a393b6 That sha256 names the PR description this verdict read. Recompute it with |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: these are clean cherry-picks of fixes already reviewed on main. None of them adds a surface or a persisted-data one-way door, and the only removal drops config keys nothing reads. Notes
[DESIGN-REVIEWED] 49a8cfa Description read[DESCRIPTION-READ] b9646975056e594ceab2cfe14dddcce71eb7dca102a8c70ad3a8df469140916d 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 |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
A process that took <home>/gateway.lock without being a gateway (for example `flock -x gateway.lock sleep 3000`) was reported as "another gateway already owns <home>; stop it first (kirocrew stop)". kirocrew stop cannot stop it, so the hint pointed the operator the wrong way. When the acquirer does not hold the dashboard port and its command line was read and names no Kiro Crew program at all (no `kirocrew` token and no `-m` of a gateway module root), the refusal now says the holder is not a Kiro Crew gateway and offers `ps -f -p <pid>`. Any Kiro Crew launch, flags before the subcommand included (`kirocrew -v gateway`), and an unreadable command line keep the old wording. The exit status is unchanged. Fixes #30407 ## Pattern harvest Not generalizable: the wording depends on this refusal's own facts (port ownership plus command line), not on a repeatable code shape. Co-author on main: Matt Briones (cherry picked from commit 47386d6)
* fix: drop retired browser keys on dashboard config saves * test: state current behavior in retired browser key test docstring --------- Co-author on main: Jacob Largent (cherry picked from commit d3549e5)
…31314) The ops-mission-control incident chat panel leaves through a full page load. Inside an app pop-out that load ran in the pop-out itself, which turned the window into a second dashboard and dropped its pop-out registration. The panel now leaves through hardNavigate, which uses the hard exit an app pop-out installs and keeps the full page load everywhere else. The pop-out's hard exit takes the destination, so it hands the panel's path and query to the main window the way the frame's other exits do. Closes #26430 (cherry picked from commit 57030af)
) A file tab opened on a path that is already gone holds the localized "file not found" sentence. The panel told that placeholder from a real last copy by comparing the buffer with the sentence in the current language, so after a language switch the old-language sentence read as a real copy: the banner offered it for download and the close guard asked to keep it. The batch close preflight never excluded it at all. Carry the verdict on the tab as PanelTab.placeholder, set where the sentence is seeded (chip open, working-tree diff open, cold-tab and restored-tab hydration) and cleared by an edit or any disk refresh, and read that flag instead of the sentence. Part of #14449 (cherry picked from commit c2dc4e4)
A refused rename of the session title said only "Couldn't rename the session" while the header reverted to the previous title, so a reader could not tell which name was rejected or whether it stuck. The notice now names the refused title, as the header showed it, and says the previous title stays, in the main header, the split-view pane header and the mobile top bar. The regenerate button's tooltip and accessible name in de, bn and hi now hold the button's visible label, as the other locales already do. In hi the quoted Undo label and the Undo button use the informal form the catalog's register gate requires. Closes #10152 (cherry picked from commit f949d8c)
…31840) MCP Management sits below the servers list on the MCP Servers sub-tab, and its own view switcher labelled the per-server share switches "Servers". Two near-identical labels on one screen let the inner tab read as a way back to the outer list. The tab now reads "Per-server sharing" in every shipped language, and the assessment view's lede and "Open ..." button name it the same way. The view's `servers` key is unchanged. (cherry picked from commit 1f819b8)
…31295) Pressing Stop a second time hard-kills the turn on purpose, but the stop card settled as "[Stop Failed, Session Reset]" because every hard outcome mapped to stop_failed_reset. The card the user escalated now settles as force_stopped with its own status label, whichever hard callback lands first; an unasked-for kill after the cooperative cancel times out keeps the failed label. A force-stopped card still gives the model its "user stopped the previous turn" note. Closes #29706 (cherry picked from commit 58c02b0)
…anes (#31961) The pull request pane and the crews pane each show a placeholder built as "Select a" + noun + "to see its details.", three pieces glued in code. Languages that place the noun elsewhere read it out of order: Japanese showed "を選択 Pull Request その詳細を表示するには", and Hindi and Korean broke the same way. Give both one sentence key with a {{subject}} placeholder and a whole translation, and drop the two fragment keys. (cherry picked from commit a5664dd)
f3507b2 to
1e4f653
Compare
…32091) The Schedule page folder toggle was named by a translated verb phrase with the folder name glued on after a space, so verb-final languages read the name after the verb (German "Ordner aufklappen Nightly"). The name is now a placeholder in the catalog sentence, so each language places it; English reads exactly as before. (cherry picked from commit 5f464a7)
The campaign detail's View report button opens and closes the report section in place, but it carried no aria-expanded, so a screen reader heard only a label change with no open or closed state. Give it aria-expanded and aria-controls pointing at the section. (cherry picked from commit 8544127)
The comment submit bar built its label from two translated parts joined
with an English space: "{count} comments" and "to send to this chat".
Every language then got English word order, so Japanese read
"3 コメント このチャットに送信" and Korean, Chinese, Hindi and Bengali
were glued the same way.
The label is now one plural key, comments_to_send, with the count
inside each translation. The two keys only this label used are gone,
and the plural registry names the new key.
(cherry picked from commit d382c63)
#31630) A Builder ID or IAM Identity Center sign-in stores the SSO-OIDC client that RegisterClient issued, but dropped its clientSecretExpiresAt. Once that registration lapsed (about 90 days), every refresh failed with a bare invalid_client and nothing could say why. The expiry is now kept on the stored token and carried through every refresh, and kirocrew doctor's crew vault line names the date it lapsed with a "cannot renew" verdict. The refresh is still attempted, so the issuer's refusal still records the marker the sign-in card reads. Part of #24640 (cherry picked from commit 4312d22)
…31728) The update-found modal and the notification banner attached their window Escape listener in a passive effect. When either opens from a default-priority update whose render runs past the scheduler's 5 ms slice, React paints it and yields before running passive effects, so an Escape pressed in that gap was dropped and the first press did nothing. Both listeners now attach in a layout effect, the shape #24284 gave the shared Modal and useDialogFocusTrap. Each component gets a test that opens it outside act() with a slow render and presses Escape from the MutationObserver callback that first sees it; both fail before the change and pass after. (cherry picked from commit a254f30)
The chat folder icon registries hold one generation task per folder, and only the task's own done callback removes its entry. A task whose event loop was torn down (a test loop per test, or an embedded gateway restart) never runs that callback, so its entry stayed: - regenerate, both folder delete paths and the folder cleanup cancelled it, which schedules onto the dead loop and raises "Event loop is closed"; - the icon preview answered icon_busy for that folder forever; - the strong-reference task set kept the dead task for good. Each of those entry points now first drops icon tasks bound to a loop other than the running one, the remedy the artifact folder icons, chat_nav and hook_reconcile already use. The app startup-hook and unadmitted-turn registries named alongside get the per-test clear in their owning test modules instead of a production change. A startup task on a dead loop must keep counting: closing a loop does not stop a worker thread the hook started with asyncio.to_thread, which is why a cancelled startup task leaves a residual marker. The unadmitted-turn map is restored by its context manager's finally, so a dead entry only exists when a test closes its loop mid-turn, and cancel_unadmitted has no caller. Closes #28511 (cherry picked from commit a8d25e5)
Twenty-four values outside apps.* in pt.json and es.json were the English text word for word, so Portuguese and Spanish dashboards showed English in the YOLO timer line, the hook form, the Dev Fleet cutover text, the Knowledge Library settings, the cloud deployment notice and the Remote Crew panel. Translate those values in place; keys, placeholders and the other catalogs are unchanged. The wording is carried over from #25761. Part of #25624 Co-authored-by: thalisantunes <6085020+thalisantunes@users.noreply.github.com> (cherry picked from commit 0c2c67f)
…#32428) The import review listed every skipped row as a muted note, so a file that could not be read or parsed looked the same as an item left out on purpose. The scan now marks each diagnostic raised from a caught read, parse or open failure with failed: true, the scan endpoint passes it through, and the review renders those rows as an inline ErrorNotice while deliberate skips stay notes. Closes #29216 (cherry picked from commit 8bf2eca)
…1042) A webhook context's detail pane shows its stored size twice: in the Stored size row as an English template "12,345 chars", and in the summary badge as the number glued to a single-form "chars" key. So one character read "1 chars", and the row stayed English in every language. Use one plural key with the formatted number inside for both, with the forms each language needs, and drop the unused "chars" key. (cherry picked from commit 2f65272)
1e4f653 to
49a8cfa
Compare
Problem / Motivation
Goal: Bring twenty-three fixes for problems a 0.12.0 user can hit, merged on
mainafterv0.12.0, ontorelease/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/0.12.0cleanly, so the code each one replaces is on release.Not a goal
main; nothing is changed only here.CHANGELOG.mdchange.What changed (motivation → approach → change)
Each main squash commit was cherry-picked with
-xontorelease/0.12.0@aa7dfb4aae, in merge order onmain, one commit each. Every pick applied cleanly, and every commit keeps its original author.66d2f0649846372f47fd3736af2d38e495b2125a360372b51c5ed5cbebab34e8a26a033f38ff48175c9a750b6f457a67411247386d66b817871ac7f4d3549e537ad842a49b3757030af8c698b9ab4fb7c2dc4e4ba3eb8df4a84af949d8cbb0d479456ab61f819b880a233c094b5258c02b0f4a56b2c2afb3a5664ddbd6e3ffa82b485f464a7181aff148842b26f71547a45cbf59009485441273f8f8fba88d05d382c63fc85b0737af684312d22e40f7f8c5db1fa254f30cf5070df3a86fa8d25e5f61a780f24a7c0c2c67f8a94e65f896238bf2eca3de4fcce096e12f652722a349a8cfa1e2Three 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.
#30624 (skip the claude-not-found warning when the SDK ships its binary) was dropped: review found its new SDK-binary scan runs filesystem calls on the gateway event loop. It stays on
mainonly until that is fixed there. #31427 (offer workflows in the + menu once the chat has a run) was dropped too: review found a failed workflow-list read in the side panel shows no error.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. #31166 drops browser config keys that 0.12.0 no longer reads, on the next dashboard save, as it does on
main.Tests
Proof: not provable -- other: a backport of main commits whose proofs ran on their own PRs
release/*, so the direct test modules were run here: the 14 Python test files these picks touch gave 560 passed.tsc -p tsconfig.app.json --noEmitpasses; the 26 vitest files these picks touch gave 702 passed;lint:i18npasses at the same count as clean release.mypy src/kiro_crewpass.i18n:checkwithout a base reports the same findings class as a cleanrelease/0.12.0on 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 #30152, #30069, #30436, #30615, #30894, #30657, #31166, #31314, #31353, #31297, #31840, #31295, #31961, #32091, #32051, #32046, #31883, #31630, #31728, #32385, #32389, #32428, #31042.
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
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.