Repository navigation
fix: backport 0.12.1 fast batch v1b-3 (23 user-facing fixes) - #32894
Conversation
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Clean backport of 23 fixes already merged on Notes
[DESIGN-REVIEWED] 0d963aa Description read[DESCRIPTION-READ] 8b2264d1914a13f289b5a8483cd092fc1196758992f7a5d7d637e78f3abbb734 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 |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS This PR only cherry-picks fixes that are already merged on What this change shipsInventory (10 of 23 items shown; list capped) — 10 justifiedIntent: give 0.12.0 users twenty-three fixes already merged on
[FIRST-PRINCIPLES-REVIEWED] 0d963aa Description read[DESCRIPTION-READ] 15fe1eda0aa7b4fd6a077bf256b23cd7496629dbaffb1ada84e57fbd24a8284e That sha256 names the PR description this verdict read. Recompute it with |
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 PR attaches no image and commits none, so no blind read was performed on the controls and copy this backport changes. Evidence gaps
The stated reason, "the main PRs show it", doesn't count as evidence for this revision. Only images attached to this PR's description, or committed under [UX-REVIEWED] 0d963aa Description read[DESCRIPTION-READ] 6150429258c2a2ac5801828cf581f3edffe032db1598515b9368593a3df10756 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 |
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: |
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: |
f3b2b99 to
99b50da
Compare
…loses (#30182) The Add from artifacts dialog is modal, but dismissing it (Escape, backdrop, close button) left focus on the page body instead of the control that opened it. The shared focus trap records document.activeElement when its open effect runs, and by then focus has already entered the dialog, so its restore target is a control that unmounts on close. Capture the opener at the open click and restore to it after the dialog unmounts, on every dismiss path. The trap keeps moving focus into the dialog on open but no longer owns the return (restoreFocus: false). The shared hook is untouched. Also handle the empty-state opener: adding an artifact refetches the listing and unmounts the empty-state button that opened the dialog, so the captured opener is detached by close time. When that happens, focus falls back to the header Add button, which renders once the listing is non-empty. Adds tests for focus-in on open, focus restore on Escape and the close button, and the empty-state fallback. Co-authored-by: kirocrew-agent <agent@kirocrew.local> (cherry picked from commit 7b5c444)
…ncy (#30085) Harness-parity row H16 counts a harness-native child at its parent: a resume naming the child is refused with native_child_not_resumable naming the parent, and the children are reported as uncharged residency. AcpClient already keeps the Claude Agent-tool child ids, but it had neither method, so a spawn_continue on a Claude child id missed as conversation_gone and session health reported no children. Bind both on AcpClient from client_native_children. The refusal delegates to the session-handle function over the same child-id set. The report counts the turn's children plus the cap's overflow, as the handle does: the client's id set outlives the turn, so it is not the count. The kiro path is unchanged: it serves through AcpSessionHandle, which already had both. Closes #30059 (cherry picked from commit 6780d83)
…d tab (#29935) A Markdown link whose destination is a local path the probe reports missing rendered as an ordinary accent link that opened a new browser tab at the dashboard origin plus that path, a URL that names nothing. The same path in backticks is a neutral copy chip, so nothing told the reader the link would go nowhere. A path-candidate link whose probe answers missing and whose destination reads as a file (relative, home-rooted, or rooted with a file extension) now swallows the plain click, drops its new-tab target and renders muted, matching the inline copy chip and the muted not-open session link. A rooted, extension-less route link keeps navigating. Co-authored-by: kirocrew-agent <agent@kirocrew.local> (cherry picked from commit 31ff5cb)
…30594) The side panel's find router returned early whenever the file was in edit mode, trusting the editor to claim Ctrl+F. In an edit-mode diff the editor only claims it while its content element has focus, so a click on a deleted row left the chord to chat find, which replaced the panel. Let the chord reach the editor, and stop a chord it did not claim on the panel root before chat find's document listener. Co-author on main: Zezhen Xu (cherry picked from commit 679b205)
An installed app that records its source updates through the pinned branch of _resolve_install_entry. That branch swallowed the official catalog's CatalogUnavailable into an empty candidate set, so during an outage the update said "no registry entry matches its recorded source" and sent the user looking for a re-homed app. Every adopted former built-in carries sourceUrl, so all of them took that branch. _registry_app_candidates now lets CatalogUnavailable propagate (any other lookup error still yields no candidate), and _resolve_install_entry turns it into fixed outage or no-install-source text, as the fresh install branch already does. The pinned lookup also runs inside install_attempt(), so an update click makes a real fetch instead of refusing from a listing page's failure memory for FAILURE_TTL after the catalog is back. Both refusals still reach no external cache. Part of #26967 (cherry picked from commit 3d961ec)
* fix: refuse unknown fields on post /api/autonudge * test: state current behaviour in the unknown-field test docstring --------- Co-author on main: Jacob Largent (cherry picked from commit 53690a0)
…1383) The source-links stop note, the browser permission explainer, the usage chart slider label and the crew memory-binding notice each joined two translated sentences in code with an English space or full stop. In Japanese and Simplified Chinese that put ASCII punctuation between full-width sentences. Each is now one catalog message in all shipped catalogs. English reads the same. (cherry picked from commit 7cc1a84)
…31059) The literal-value swap that keeps stored text away from i18next lived inside i18nT(), so the t from useTranslation() still handed values to i18next. i18next fills the first copy of a placeholder it finds, so a value holding another placeholder of its string (or {{productName}}) was filled in and the real placeholder showed raw. i18nT() and the useTranslation() t both end in i18next.t, so initI18n() now wraps that one function with the same swap, and i18nT() calls it like the hook does. The i18n authoring doc says so, and states that <Trans> re-parses its string, so stored text there is a component's child rather than a value. Part of #22834 (cherry picked from commit 960b69c)
…0832) kirocrew learn add wrote through write_lesson directly and never ran the LEARN_ADD_SCHEMA length check that POST /api/lessons and the learn_add MCP tool run, so it saved a rule longer than the 500-char cap the other two surfaces refuse. Validate rule and negative through the same schema before the store is opened, refusing with the same message, a non-zero exit, and nothing written. A rule exactly at the cap still saves. Closes #30340 Co-authored-by: kirocrew-agent <agent@kirocrew.local> (cherry picked from commit d831b16)
…31458) A missing home under a read-only parent, a read-only mount, or a home made read-only with `chmod -R a-w` stopped `kirocrew gateway` with a raw traceback: project-dir detection, the live-target read, the home's setup and the crash log all touch the home before the persistence preflight in the async boot ran. The gateway branch now runs that probe as its first step and exits 1 with the "Cannot persist state" line; the async copy is removed, so the boot does the same work once. Project-dir detection reads a home it cannot create as holding no saved path. A home whose own directory denies its owner write is now refused instead of being reset to owner-only. (cherry picked from commit 038776a)
(cherry picked from commit 4e4d350)
#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)
A steer the running turn did not take runs as its own turn: the turn ended before it read the steer (requeued, then superseded), or the steer path was unavailable and the send fell through to the queue. Its decision receipt still read 'Jev chose: interrupt this turn', which on the row below the Stop card reads as a decision about the new turn. UserMessage now passes ranAsOwnTurn to SteerDecisionLine for a row with no meta.steer or a requeued or superseded steer row. A steer choice on such a row shows 'Jev chose: interrupt the turn, but this runs as its own message' and sets data-ran-as-own-turn. A queue choice is unchanged. (cherry picked from commit 2bc83d0)
…2053) The machine-identity strip on the System page's Performance plane glued an English colon and space after each translated label (Hostname, OS, Python, Working directory) in code, so Chinese and Japanese showed an ASCII colon where they write a full-width one, and French lost the narrow space before it. Each row is now one catalog message ending in a <value/> tag, rendered through <Trans> with the value as that tag's own child, the pattern the i18n catalog guide prescribes for stored text. Every catalog places its own colon, and a host string such as a directory named a<br>b is still shown exactly as written. (cherry picked from commit 75e43c8)
The Memory card on Settings > Overview built its line from five catalog fragments glued around two raw numbers. Every language kept English word order (Japanese read the idle hours after the verb, Korean put them at the end) and English said "1 days of history". Use two whole sentences, one with a plural day count, in all 12 catalogs. (cherry picked from commit 322081e)
The Agent Worlds scene picker showed each scene's name and description in English in every language: the SCENES list held them as module-level English literals, and the i18n codemod skips src/pages/scenes/. Each scene's label and desc is now a getter that resolves a catalog key under pages.scenes.config at read time, so the page and its popout show the active language and follow a language switch. The keys carry English plus the eleven shipped translations; the pseudolocale is regenerated. Closes #28898 (cherry picked from commit feef219)
…32036) Closing the Schedule page's job dialog mid-save cannot cancel the request, yet the dialog still offered Cancel with no word that the schedule might be created anyway. Its late answer then ran the closed dialog's success path, which closed whatever dialog the user had opened since, and a late failure vanished with the unmounted form. While a save is in flight the dialog's Cancel reads Close and a strip above the footer says the schedule may still be created (or the changes saved). A late answer only refreshes the list; a late failure is reported on the page, one dismissable notice per save, worded by whether the server refused it or never confirmed the outcome. The Logs tab is locked mid-save and Save is off on Logs, so the form that reports a result stays mounted, and a page leave or browser Back is refused while the dialog holds typed work, an edited alert threshold or time zone included. (cherry picked from commit f3ca014)
…#31208) The About panel joined a translated update-error label to its reason with an English colon, and the Remote Crew sign-in note joined two translated sentences with an English space. In Japanese and Simplified Chinese that put a half-width colon or a stray space between full-width text. Each is now one catalog message, with the reason as a placeholder, in all shipped catalogs. English reads the same. (cherry picked from commit 03016b6)
99b50da to
0d963aa
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@8e72744cbb, in merge order onmain, one commit each. Every pick applied cleanly, and every commit keeps its original author.7b5c4440730ec7ce370a5bbd6ddb8e51474d86e26780d836ebe03b3ae3f736dee37f402d5d7c78a631ff5cbb2b03178046e1259187cd18bf282b6a642ed83e2a20ba89cc1906679b2050616d4ea5a9573d961ec8438c85b3947553690a07d21f5e58634f60999cbb20aee45ca58e7cc1a8471a106fdb9dcd960b69c28f9e663d6effd831b16af490558ac601038776a989faaa4f3f784e4d3506732a7a7f25b3e68ca0abb9edbf03e35e2bc83d0a523179f1e10c75e43c830bee4701358b322081ede5b539df60d7feef2198948609b178a3f3ca014476fdc8781fb903016b64c899b50dafaePicks whose co-author trailers carried internal addresses have each trailer rewritten as a plain "Co-author on main: " line, because the commit guard refuses the address.
#30084, the next row in merge order, now conflicts in
docs/system-specs/modules/harness-parity.mdonly, so it goes to the docs/tests batch. #30839 (*.localhost preview refusal) was dropped: UX review found the refused card's only action shows a raw config key instead of a label.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. #31087 makes the autonudge API refuse unknown fields it silently ignored, 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 20 Python test files these picks touch gave 1,368 passed and 2 skipped.tsc -p tsconfig.app.json --noEmitpasses; the 24 vitest files these picks touch gave 485 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 #30182, #30215, #30085, #30227, #29935, #30519, #30473, #30594, #30597, #31087, #31310, #31383, #31059, #30832, #31458, #31431, #31698, #31055, #32053, #32439, #32443, #32036, #31208.
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.