Repository navigation
fix(release): backport twenty-nine user-facing fixes to 0.12.0 line - #33529
Conversation
… 400 (#31372) PUT /api/skills/{name} checked only that content was truthy, so a number, list, object or true reached the SKILL.md writer and failed there with a 500. Answer 400 invalid_field_type before any write, as the create route already does. Every string saves as before. Fixes #31370 (cherry picked from commit 2d620b8)
) Leaving the chat page for Settings, Apps or Crew and returning by Back mounted the chat again on a POP, and its Back/Forward reader stayed disarmed until the page made a navigation of its own. Every further Back onto an earlier session entry was ignored, and URL sync then pushed the session on screen over it, so the arrows stopped retracing the sessions visited before the trip. The reader now skips only the entry the page mounted on, which the mount deep-link path owns, and honours a POP onto any other entry. Closes #21042 (cherry picked from commit b993f2a)
…pp_mention event (#26770) * fix(slack): dedup observe-channel history pushes by msg_ts A Slack @mention is delivered as both a `message` and an `app_mention` event sharing one msg_ts, and socket mode can redeliver an envelope. Observe mode records channel history above the msg_ts turn-dedup, so the same message was written to history twice (in memory and on disk) while the turn ran once. Guard the observe push with its own `observe:` seen key so each message is recorded once, without consuming the bare msg_ts token the paired app_mention turn needs. Closes #26769 * test(slack): sort the observe-dedup test imports --------- Co-authored-by: iamwhatever <12429733+iamwhatever@users.noreply.github.com> (cherry picked from commit 45955da)
#31403) On macOS the main dashboard window had no keyboard way to close it. Cmd+W is "close session" in the renderer on every platform, and the stock macOS Window menu has no Close entry, so nothing closed the window from the keyboard. Write the macOS Window menu out as the stock items plus Close Window on Cmd+Shift+W, matching Ctrl+Shift+W on Windows/Linux. The menu keeps the windowMenu role, so AppKit still lists the open windows. Close is the native role: it fires the same close event as the red button, so the main window hides to the tray (leaving fullscreen first) or quits when set to, and a second window just closes. Cmd+W still reaches the renderer. Closes #11393 (cherry picked from commit c318773)
…es (#31933) The Connections gallery hides vendor-blocked and launch-gated providers, but POST /api/connections/mint and /api/connections/test accepted any registry slug, so a stale tab or a hand-built request for figma started a real kiro-cli sign-in that cannot succeed. Name the gallery rule once as registry.is_visible and refuse a hidden provider on Mint and Test with the existing 400 unknown_provider. Cancel and Disconnect still reach every registry provider, so a grant stored before its provider was hidden stays revocable. Closes #14199 (cherry picked from commit f27dfcb)
While the gateway is offline, menus and the tag popover show a standing reason row under their dimmed actions. The row never wrapped, so it became the widest row and the host grew the moment the connection dropped: the Move to folder submenu, the session menu that hosts it, its Turn time limit and Send a copy submenus, the sidebar folder and New chat menus, the artifact folder menu and the tag popover all resized. The row now takes no width of its own and fills its host, so the sentence wraps inside the width the host already had. (cherry picked from commit 648cebf)
…31548) A workspace create with --copy-from stages the copy, then renames it into place. Both the CLI and the dashboard caught every OSError from that rename and said the destination already exists. On Windows a file held open inside the staged copy refuses the rename with "Access is denied", so the user was sent looking for a directory that is not there. An existing destination (FileExistsError) keeps the same message and the dashboard's workspace_dir_occupied code. Any other rename failure now names its cause; the dashboard answers 409 with the new code workspace_publish_failed. Closes #27822 (cherry picked from commit d0583bc)
…31395) AcpProvider.cleanup_session and AcpSessionHandle.destroy resolved and unlinked kiro-cli's <sid>.json + .jsonl on the gateway event loop, so a slow home mount stalled every session and could trip the loop watchdog. Both now hand the delete to a worker thread through drained_to_thread, which keeps a cancelled caller waiting until the delete has finished: teardown is where cancellations land, and a dropped delete leaves a transcript nothing else removes. The shared transcript helper now logs under a caller-neutral "transcript cleanup:" prefix. Builds on #24003 by @chenmingwei23. (cherry picked from commit 3fb2a88)
…#31700) A phone-access QR minted from a session pinned to one device carries that pin into its child, so only the device already showing the code can redeem it. The mint answered 200 anyway, and every other device that scanned the code was refused with no explanation on the dashboard. The QR mint now refuses such a caller with 403 caller_device_bound and a message that says where a working code comes from. A login-scoped pin names a person rather than a device, so it still mints. The device bound itself is unchanged. Closes #22952 (cherry picked from commit d4e70da)
…#31427) A chat that never ran a dynamic workflow still listed Workflows in the side panel's + menu, and picking it opened a panel that could only say there were no runs. The row now waits for the chat's first run, listed by the workflow-runs API or live in the chat's run rows, the way Summary waits for session summaries. It fails open while the list is unread, and an open Workflows tab and the run cards' panel link are unchanged. Builds on #21102 by @adlio. (cherry picked from commit 9929750)
) The desktop header shows the session menu chevron before the session title, but the phone's single top bar put it after the title, so the control switched sides between viewports. The phone's titled trigger now renders the chevron first, matching desktop. (cherry picked from commit 277b277)
…31472) Settings > Notifications > Sources showed what the section is and what muting costs as one paragraph under the heading. The section intro now sits in the heading's info tip, and only the consequence line (muted notifications keep their history but never badge, sound or banner) stays visible, the same split the other Settings rows use. Closes #29729 (cherry picked from commit 83b7a6d)
…1617) When the move into a just-created folder fails, Retry in the session menus' New folder dialog re-read the whole folder list first, to re-create a folder deleted in the meantime and refresh the names the failure notice shows. That covers a gap nobody has reported and is larger than the dialog needs. Retry now moves straight into the folder the first try created. When the move answers that the folder no longer exists, the dialog forgets it, so the next submit creates the folder again instead of retrying a move that can never succeed. Any other move error keeps the folder and the names from the create. The create branch loses its retry-only values, and the three tests that pinned the re-read go; one test pins the deleted-folder retry. Closes #24179 (cherry picked from commit 6cbe425)
On macOS, saving a file from the dashboard (the file editor, a steering document or a skill) replaced it with a new copy that kept only the permissions its folder hands down. An access entry the owner had set on the file itself was lost by a single save. The new copy now takes the replaced file's own access entries before any content is written to it. A save that cannot keep them is refused, the original file is left as it was, and no temporary file is left behind. Cleaning up that temporary file never opens its content to an account the entries kept out. (cherry picked from commit 6f9aaf6)
…1425) When the port answers /api/ready with 200 at the readiness deadline but the run-marker identity check cannot tie that answer to the replacement (its best-effort marker write failed), restart printed that the replacement is "not serving port P" and invited a re-run that stops a gateway that may be serving. The wait now returns a separate unconfirmed verdict for that case. Restart says it could not confirm the replacement is the one serving, warns that a re-run would stop it, names the run directory to check, and audits replacement_unconfirmed_within. It still exits 1 and never reports success for an unproven identity. Closes #21247 (cherry picked from commit 14c66ba)
…31058) When a steer is requeued and the queue drains, the queue_pop reducer rebuilds its text as a new user row. The reducer now stamps the client-only meta.landedFromSteer on that row when the nearest earlier user row with the same text is a requeued or superseded steer. UserMessage plays the existing msg-highlight accent ring once for 1.2 s on it, none under reduced motion. A module-level guard keyed on the row identity keeps a virtualized remount from replaying it. (cherry picked from commit 6add8f6)
A chat link whose destination the link policy refuses rendered as plain prose, indistinguishable from the sentence around it. Both chat renderers (MdAnchor and Mochi's ChatPanelMarkdown) now give the inert label a muted dotted underline from existing tokens, plus a generic translated reason as the hover title and as a visually-hidden suffix. Allowed links are unchanged and the label is still never an anchor. Fixes #9945 (cherry picked from commit 0975050)
…30947) Restart, tag deletion and web-app teardown use native confirm() dialogs. On Windows, closing one can leave the app window ignoring the keyboard until it is reactivated (#15976). Ask through the existing in-app useConfirm dialog instead. Keep the full restart reassurance in the body and use the existing short action label for its title. Restart intentionally uses the neutral style because its question is nondestructive. Translate the tag question with a locale-quoted name, and render teardown resources as list items between the existing warning paragraphs. The page stays live during an in-app question. Capture the tag id and name before waiting, then re-check its cached row and shared pending deletions after approval. Skip a missing or degraded tag or a deletion already in flight. Keep the column-filter question above its popover, and let the modal own its outside presses and keys. Capture the teardown slug and callback before waiting. Cancel unanswered consent when the card switches artifacts, and re-check expiry and pending state after approval. Invalidate the captured artifact on success without notifying or changing a departed or replacement card. Keep the dialog mounted across all card returns and ignore repeated question presses. Add three locale keys across the owning English catalog, translated catalogs and pseudolocale. Convert the behavior tests to answer the in-app dialogs and remove the three migrated paths from the native-dialog allowlist. Leave the theme editor and DisplayPanel migration for a later PR using the nested Radix dialog. Part of #15976 (cherry picked from commit 5718e01)
#31681) The trust banner joined its host list with an English comma and space. Japanese and Simplified Chinese read an ASCII comma between full-width text, and German, Spanish, Italian, Hindi and Korean lost their word for 'and'. The list now goes through fmtList with the narrow style, so English still reads 'a, b'. (cherry picked from commit 8f1aa7c)
On a `kirocrew service install` host Dev Fleet cannot drive the unit, so Restart Gateway and Make Live fell to the foreground restart. Its detached `kirocrew restart` runs inside the unit's cgroup and is stopped with the gateway it stops; an older `Restart=on-failure` unit never relaunches that clean exit, leaving the host with no gateway. The foreground eligibility check now says no for a gateway that a generated service definition launched, so the restart is never attempted there: Restart Gateway answers with the service reason and the manual command, and Make Live stages the cutover for the operator. The no_user_unit message no longer calls every such unit a root-owned system unit, since a user-scope install reaches it too. Closes #31647 (cherry picked from commit de4a8b0)
…32289) A background title call that failed (a provider throttle that outlasted the one-liner's retries, a timeout) left the session showing "New Session..." with no fallback, and a one-message session never got another attempt. The failure now shows the truncated first-message fallback the SKIP reply already gets, left unlocked so the next user turn asks the model again within the existing five-attempt budget. Closes #31854 (cherry picked from commit 7a6e729)
A sub-agent on its parent's shared runtime never enters the session store, so the runs it spawned read no approval policy and were treated as untrusted: under a trusted chat, a depth-two spawn and its tool calls asked on a key no tab shows. The spawn gate and the tool policy now read the parent run's own grant for such a key, while that run is live and its lineage starts at the root session and is not revoked. A sub-agent's own grant is not passed on. Part of #28364 (cherry picked from commit f828840)
(cherry picked from commit 70df3f7)
(cherry picked from commit 7caab2f)
The Strands model Save and Cancel buttons went natively disabled while the save ran, which drops a keyboard user's focus to the page, and a refused save left it there. Both now use aria-disabled with a guard so the pressed button keeps focus, and when the draft is saved or dropped and the buttons leave, focus returns to the model field. (cherry picked from commit 2da3750)
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. The blind read was not performed because the PR attaches no image, and this backport changes several user-visible controls and strings. Evidence gapsEach item below needs a screenshot in the PR description, then a re-run of this workflow:
Notes
[UX-REVIEWED] f74d213 Description read[DESCRIPTION-READ] d699cedd3ef39c67ee4a9bbc3072831d82d34014a5db943f62d4a1bbdc9508c0 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 The description and the diff disagree about #31427. Everything else is a clean replay of fixes main already ships. Notes
What this change shipsInventory (10 items, list capped) — 9 justifiedIntent: 0.12 users get fixes main already merged, without waiting for 0.13. FIX.
[FIRST-PRINCIPLES-REVIEWED] f74d213 Description read[DESCRIPTION-READ] 9171508f0745d4097bb5067a450106d5507040ac066e968251a98168b365fb63 That sha256 names the PR description this verdict read. Recompute it with |
Design Review (Opus 5.5) — ✅ PASS (a BLOCK with no Trigger and no Outcome does not block)Design-level review of Design-Verdict: PASS cannot evaluate: missing screenshots of the reshaped dashboard surfaces. Also, the description says #31427 ships, but this head reverts it. BlockersThe changed UI surfaces have no rendered evidence Notes
[DESIGN-REVIEWED] f74d213 Description read[DESCRIPTION-READ] d465fe817f87b982d8363f08e4e931899df222e931db01821a944cba97487add 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: |
Problem / Motivation
Goal: 0.12 users get 29 small user-facing fixes that already merged on main, picked cleanly onto the 0.12.0 release line for 0.12.1.
These fixes merged to main after the release branch was cut. Each one changes something a 0.12 user hits today: lost focus, untranslated labels, a stuck dialog, a misleading error, a lost permission bit, a restart that fights its service manager.
Why it matters
Without a pick, 0.12.x users keep hitting these until 0.13. Every pick here applied cleanly, with no conflict resolution by hand.
Not a goal
What changed (motivation → approach → change)
Each pick is
git cherry-pick -xof its main merge commit, in main merge order, on top ofrelease/0.12.0at7a89cc18d60f581d994b18afd6a3d0c8ef0b0e23. Each commit keeps its(cherry picked from commit …)line.Dropped:
website/src/test/ChatSidebar.boardColumnTagFilter.test.tsx"keeps focus in the tag question when the popover repositions behind it" fails on release. Reverted in this batch; the file's release version passes again.errors-use-error-notice): a failed workflow-runs read in the side panel shows no error notice. Reverted in this batch rather than fixed on release.Backwards compatibility
Compatible: each pick is the same change main already ships; no key, endpoint or file format is narrowed beyond what main does.
Removes nothing: no pick removes a documented capability.
Tests
Proof: not provable -- other: release backport of already-merged main fixes, each proven on its own PR
Run locally on this head, one process each, no
-n:Results on the head before the #31427 revert:
test/*.pyfile, 975 passed, 6 skipped.website/electron/test/*.test.jsfiles with the package preload, 115 passed.tsc -p tsconfig.app.json --noEmit: clean.After the #31427 revert:
ContextBreakdownPanel.test.tsx36 passed andsidePanelAddMenu.test.tsx30 passed (the two test files #31427 touched);tscclean.Before the #30947 revert, vitest had 1 failure out of 761 (the test named above); nothing else failed.
Manual verification
N/A — each fix was verified on its own main PR; this PR only replays them.
Screenshots / video
Screenshots come from each pick's own main PR; the code here is the same change, picked unchanged. Picks whose main PR carried no screenshot are not shown.
#31403 (from its main PR; the picked change is identical)
#31702 (from its main PR; the picked change is identical)
#31110 (from its main PR; the picked change is identical)
#31700 (from its main PR; the picked change is identical)
#31406 (from its main PR; the picked change is identical)
#31472 (from its main PR; the picked change is identical)
#31058 (from its main PR; the picked change is identical)
#28418 (from its main PR; the picked change is identical)
#32378 (from its main PR; the picked change is identical)
Related Issues
Picks for 0.12.1 from the fixes merged to main but not on the release line.
Sibling sweep
Siblings: none -- a backport batch; each pick's own PR carried its sweep
Pattern harvest
Not generalizable: a release backport batch, not a new defect fix.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)