Repository navigation
fix: backport 0.12.1 batch 60 (24 wide-bar fixes) - #32860
Conversation
_approval_wait_phrase always wrote a plural suffix and never used seconds, so a sub-minute window read 'waited 1 minutes' and a one-hour window read '1 hours'. Pick seconds below a minute and the singular unit for a count of one, matching the minute/minutes choice format_background_turn_abandoned_card already makes. Closes #29735 Co-authored-by: kirocrew-agent <agent@kirocrew.local> (cherry picked from commit 68c7f65) Original author: Raymond Chen (#29925).
… dropped (#30505) The /search handler set truncated whenever the result list reached MAX_SEARCH_RESULTS. Both search collectors stop at that cap, so a folder with exactly 500 matches looked the same as one with more. The panel then showed "500 results (capped)" even though every match was listed. The ripgrep and Python collectors now keep one result past the cap. The handler sets truncated only when that extra result is present, and trims the reply back to MAX_SEARCH_RESULTS. Closes #30503 (cherry picked from commit 3adab5e) Original author: rohnkum (#30505).
Opening an artifact whose files the store refuses or cannot read returned aiohttp's plain-text 500, which the detail page cannot show. The detail, version and event-log reads answer that failure with the same JSON error body the list route already returns. (cherry picked from commit ded740c)
A Range header aiohttp cannot parse (a reversed range, several ranges, another unit) on GET /feature-videos/<release>/<file> was answered 416. RFC 9110 section 14.2 says such a header is ignored and the whole file sent with 200, as the file-stream and outbox routes already do. Only a well-formed range that starts past the end of the file stays 416. Closes #28959 (cherry picked from commit 305c6d2)
…rd (#31127) The issue-radar reference hover card closed on every scroll event in the page, captured from any element. A scroll that does not move the hovered link (another pane, a nested list, a scroll that ends where it began) therefore dismissed a card the pointer was still on. The card now compares the link's position with the one captured when it opened and closes on scroll only when the link moved. A resize still closes it. This is the same rule the chat link-preview layer follows since #28498. (cherry picked from commit ccc2167)
…1480) With KIROCREW_HOME set to a symlink, a subagent's result path was handed out under the symlink's target, because data_home() resolves the override. #4047 kept the declared spelling for a symlinked $HOME but not for this override, so a reader keyed on the configured path did not recognize the file it was told to read. agent_dir_for_display now rebuilds the run directory on the absolute KIROCREW_HOME value when that override is the active home. Validation and every file operation still use the resolved _agent_dir, and data_home() is unchanged. (cherry picked from commit a31eccd)
When the graceful shutdown bound cancels the gateway teardown, the wait for admitted AutoNudge writes that follows it never ran, so the warning naming a write left in flight at the deadline was skipped. Run that wait in a finally around the teardown, so a cancelled or failed teardown still waits out the graceful budget and logs what is left. (cherry picked from commit 7878189)
The four log level buttons showed the chosen level by colour alone. A screen reader heard four plain buttons and was never told which level is on. Wrap them in the shared RadioChipGroup, named by the Log Level label, and mark each with radioChip so the chosen level reads as checked. (cherry picked from commit c77ef4b) Original author: Rohan Khanderia (#31876).
An "Empty trash" job could finish reporting more bytes freed than its own total. The delete counts the files a batch's manifest lists, the manifest itself and the batch's staged crew logs, while the job's total counted only the files the manifest lists. The total now counts the same three things, the manifest's size read through the descriptor the approval already pins. The comments and the spec that describe the total say so. Closes #29749 (cherry picked from commit 6385b99)
The DAG / Phased switcher on a project page showed the chosen layout by colour alone. Wrap the pair in the shared RadioChipGroup with a translated group name, and mark each button with radioChip so the chosen layout reads as checked. (cherry picked from commit ecd76bf) Original author: Rohan Khanderia (#32070).
The share dialog's pre-share warning lists every sensitive kind it found, such as an AWS key and a local path. It joined those translated labels with a hard-coded ", ", so German read "A, B, C" instead of "A, B und C" and Chinese got an English comma instead of "A、B和C". Both the card and the post-text warnings now go through fmtList, so the list follows the active language. Fixes #32037 (cherry picked from commit 69bdcd6) Original author: Rohan Khanderia (#32057).
#32269) The Load more button under a remote provider's artifacts used a native disabled while the next page loaded. A disabled button leaves the tab order, so a keyboard press dropped focus to the page for the whole load. Mark the busy button aria-disabled and ignore a second press, the way the Ops Mission Control refresh does. (cherry picked from commit ecddebb) Original author: Rohan Khanderia (#32269).
mint_remote_token killed and reaped its ssh child only when its wait timed out. A cancelled mint (an instance edit stopping the refresh loop, a gateway shutdown) left ssh running to mint a token nothing would store. The cancel arm now runs the same kill_and_reap and re-raises. Builds on #25162 by @leonlaiyc. (cherry picked from commit 0db7919)
The risk badge's thumbs carried the label "Jev", repeating the word the line opens with, and nothing visible said a vote is only feedback: that meaning lived in the buttons' title and aria-label, which a sighted reader on touch or keyboard never sees. The label now reads "Feedback only". The key is renamed from rate_jev to rate_label because its meaning changed, and all twelve catalogs, the pseudolocale and the translator context carry the new value. The badge row wraps, so on a phone-width pane the note and the thumbs take their own lines instead of the note being squeezed to a narrow column. The decisions spec states the label. Closes #12298 (cherry picked from commit 84be3f2)
…32494) "Draft a prompt in chat" on an artifact with no chat yet opened the new chat with an empty message box. The gateway lists a new session in its slots frame before it answers the create request, so the panel mounted the embedded chat and activated the session from that frame. The prompt is staged only once the create returns, and the chat reads a staged prompt only when it mounts or switches session, so it was never read. The panel now waits for the create to return before showing the new session, which is the order the flow already assumed. Closes #30338 (cherry picked from commit bcc5647)
…32511) The session MCP servers menu built its "loaded this session" set only from kiro-cli's tool_search result JSON. The default backend (KAS) reaches deferred tools through its own tool_load / tool_call pair and never sends that output, so every tool stayed "deferred" and each server read 0/N even after its tools ran. A tool message that records an MCP call now counts as loaded too, named by its mcp_server / tool_name meta, else by its @server/tool title. (cherry picked from commit 113ff88)
…32506) The composer passes its own trigger to the goal popover, and the popover put the "Next cycle in ..." title and its hover refresh only on its built-in button, so hovering the composer goal button showed the cycle without the countdown. A supplied trigger now gets the same title and refresh while a goal loop is active, and its accessible name stays countdown-free. A loop held for an approval or waiting on a person shows the cycle only, matching the popover's status line. Closes #30329 (cherry picked from commit f5deb9c)
…2480) Sending at the bottom of a chat whose last reply is long briefly opened a gap below the last row, the height of a whole estimated row (about 2,300px under a long reply), until the prompt's row mounted. The transcript reaches the virtualizer through a deferred value, so React can render the longer list and then throw that render away. The height tree is synced to the row count in render, so the thrown-away render left it holding the new row; the next render, on the list still on screen, reused the memo and drew a bottom spacer for a row that was not committed. Nothing pinned across it because no row was appended in that commit. The height owner now re-syncs the tree whenever it holds a different row count than the render reading it. Closes #29719 (cherry picked from commit c48df82)
A document comment box takes the same one-mic lock as the chat composer but has no chat behind it, so a blocked chat fell back to "Microphone in use in another chat" for a holder that is not a chat. The fallback key already says "elsewhere" in its name; its text now reads "Microphone in use elsewhere" in English and all translations, which is true for every holder the chat cannot name. A chat the session list knows is still named, with its jump button. Closes #30383 (cherry picked from commit d9768f8)
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 changes user-visible copy and controls on several surfaces but attaches no image, so the blind read was not performed. Evidence gaps
Notes
[UX-REVIEWED] cd33703 Description read[DESCRIPTION-READ] 3f6ea641f85b61b044a80d9999b9852440c28e07a1df00234d9e8a499d033749 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 |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: this carries 24 fixes already merged on Notes
[DESIGN-REVIEWED] cd33703 Description read[DESCRIPTION-READ] ea24ea163629cb6512f7497372741e55933cb1708aca98c4a794c49977e39e6c 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 backports 24 fixes that already merged on What this change shipsInventory (10 items) — 10 justifiedIntent: give 0.12.x users the user-visible fixes that already shipped on
[FIRST-PRINCIPLES-REVIEWED] cd33703 Description read[DESCRIPTION-READ] 7d0a0052c6d9300aff027d742db51b79bcdadf3c966a197e31f61ee15214a948 That sha256 names the PR description this verdict read. 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: |
Problem / Motivation
Goal: Bring 24 small fixes that merged on
mainafterv0.12.0ontorelease/0.12.0, one commit each, in a PR of their own.Why it matters
Each row is a user-visible fix from the 0.12.1 wide-bar list (Ray, 11-07). Every pick applied cleanly onto release.
Not a goal
main; nothing is changed only here.CHANGELOG.mdchange.cfgname that release'scli_doctor.pydoes not define (flake8 F821 at line 654; 83 doctor tests fail), so it needs a prerequisite and is left for a later round.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, with no prerequisite and no hand-resolved hunk; every commit has added and removed lines identical to its main commit.ffdfa4aa7dab0e7e6edd68c7f65d24b49e4744bfd10a48b0fa4e88c6ba13d5e25625bddac869bb5a3adab5e0c4acce790ee1ded740cacb077c8590eb305c6d26c146bac6c4adde9ed9b2335746935c18b83ae4fae5fa8cb41548ccc2167cc19fd13339dba31eccd7b560abbdd479787818992ef92ede2bd7c77ef4ba2497681d79436385b9954006a7ce0883ecd76bf1d3477b11783f69bdcd6e9a7907978530ecddebbe005c305532a50db79190aeb29f72c6af84be3f274d1fdbc078d4bcc564704939f6a81831113ff88c51266e940b5ef5deb9c2fb80909c68d9c48df8254e44234913ced9768f8adccd3370363f11 commits carried a non-public author email or an internal co-author trailer, which the push guard refuses, so they are authored with the public identity and name their original author in the message; internal trailer lines are removed. The other 13 keep their original author.
builder-security checked: each change is the one already merged on
main; no bind, listener, new route, auth mechanism or storage grant is added.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
release/*, so the direct test modules were run here: every touched test file plus every test file that imports a touched module, 22 files,-n 4 --dist loadgroup. 1143 passed, 3 skipped, 0 failed.tsc -p tsconfig.app.json --noEmitpasses.vitest runon every touched test file and every touched component name: 60 files, 997 passed, 3 failed. The 3 are insrc/pages/SchedulePage.secrets.test.tsxand fail the same way on a cleanrelease/0.12.0@aa7dfb4aae.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: each frontend pick carries the same rendered change as its main PR, whose screenshots show it.
Related Issues
Backport to the 0.12 line for 0.12.1 of #30142, #29925, #30139, #30207, #30505, #31068, #31105, #31056, #31157, #31127, #31480, #31508, #31876, #31992, #32070, #32057, #32269, #32357, #32077, #32494, #32511, #32506, #32480, #32529.
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.