Repository navigation
fix(release): backport twenty-one user-facing fixes to 0.12.0 line (s3b3) - #33549
Conversation
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: the batch backports fixes main already ships, every pick is accounted for, and the platform-specific picks (deb deps, Electron fullscreen) are byte-identical to their verified main PRs. Notes
[DESIGN-REVIEWED] 7563c2c Description read[DESCRIPTION-READ] a56340d678d1675a6ebb15893dbd4ef29fd22690f6a29150c35b3a2c6022737e 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 Every item is a cherry-pick of a fix that already merged on main. Each one carries the main PR's own provenance and adds nothing beyond what that PR shipped. What this change shipsInventory (10 items) — 10 justifiedIntent: 0.12 users get fixes that are already on main in 0.12.1, without waiting for 0.13. This is a FIX (a backport batch). There are 21 picks; the 10 most noticeable are listed here.
[FIRST-PRINCIPLES-REVIEWED] 7563c2c Description read[DESCRIPTION-READ] fccae642824f8aa47d9b6f8a937a8e2ca7827a58e1a38dc907f6a2fe1e55b34c 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 of every changed surface except the app pop-out. The three supplied images cover only #32865. The Evidence gaps
Notes
[UX-REVIEWED] 7563c2c Description read[DESCRIPTION-READ] acbdb9b576b5ef77a17e18a1b07af6091831148b1a08abb5277d9ac70959cab6 That sha256 names everything this verdict read: the PR description, plus the 6 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: |
…31572) In Skills > Pending review, a collapsed row's Approve was natively disabled with nothing saying why; on a queue with no flagged candidate no hint rendered at all. A natively disabled button gets no hover or focus, so it could not carry one. The collapsed row now renders Approve aria-disabled and wraps it in the shared HoverTip, which shows the existing "Review the candidate to enable Approve." sentence on hover and keyboard focus and exposes it as the button's description. The click handler refuses while disabled, so the cached detail of a row that was opened and collapsed again cannot be approved from the collapsed state. Other disabled reasons keep native disabled. Closes #27255 (cherry picked from commit 2a7a21b)
…32448) On Ubuntu 24.04 libasound2 is only a virtual package, and OSS4's liboss4-salsa-asound2 provides it too. apt picks that stand-in, which lacks ALSA symbols the desktop binary links against, so the app installed from the .deb exits with "undefined symbol: snd_device_name_get_hint" before it opens a window. The deb now depends on "libasound2t64 | libasound2 (>= 1.0.17)". Naming the real t64 library first fixes a clean install; the version keeps an already-installed stand-in, whose Provides carries no version, from satisfying the dependency. Releases without libasound2t64, such as 22.04, fall through to their real libasound2 as before. The deb smoke step also checks that no ALSA symbol of the installed binary is unresolved, and the packaging test pins the entry. Builds on #28115. Closes #27964 (cherry picked from commit 9f2bf27)
The collapsed tool-call preview sent any command containing a newline to the verbatim fallback, so a heredoc or a script showed as plain text while a one-line command in the same session kept its colours. Lay each physical line out as its own row. A heredoc body shows as typed text up to its delimiter, blank lines kept at full height; a line ending in a backslash goes on in a row of its own as the same command; the operator a line ends on leads the next command row; a # comment is plain text. A quote or substitution left open across a newline keeps the verbatim fallback. Builds on #25782. Closes #25769 Co-authored-by: Lê Bảo Duy <lebaoduy1993@gmail.com> (cherry picked from commit 7fdaa45)
The Connections page probe spawned stdio MCP servers with a fixed 1 MiB asyncio read limit, so a server whose tools/list reply is one longer line probed as an error with "Separator is not found, and chunk exceed the limit", although the MCP gateway reads the same server with its configured ceiling (64 MiB by default). The probe now takes its limit from resolve_read_buffer_limit(), the gateway's own resolver (KIROCREW_MCP_READ_LIMIT, then mcp_gateway.read_buffer_limit_bytes, then the default), on every probe. A test drives a real stdio child with a 2 MiB tools/list line at the default ceiling and at a 1 MiB env ceiling, and the MCP architecture doc says the probe follows the same ceiling. Closes #30309 (cherry picked from commit 0a2010c)
…#32485) A split-view pane whose one-time history page arrives after the live chunks of the turn it shows kept two copies of the reply: the page's streaming row and the one the live chunks built. When the slot runs and the page's row reaches at least the slot's chunk floor, the pane's live row now takes the page's text in place of a second copy, keeping its place under any reasoning, and the floor rises to the page as the other snapshot readers raise it. Closes #29707 (cherry picked from commit d25b0b2)
With any app disabled (most built-ins ship disabled), the first catalog pass of a new process asked each flat skill for its app: line. No walk had vetted the stored rows yet, so every one of those reads ran the full unvetted-path check, and both owner roots were resolved again per skill. At 1,500 skills that added seconds to the first reply. Until a walk vets the stored rows, the gate now caches a flat row's saved metadata as its frontmatter while the file's stat fingerprint still matches, the same check and cache list_skills already applies, so only files the index holds no current row for are read. The apps and builtins roots are resolved once per catalog TTL instead of once per skill. The saved-metadata read now treats a cell nested past the JSON decoder's depth like any other undecodable cell, since the gate reads it too. Closes #29801 (cherry picked from commit dd16b79)
…32444) The Accounts and Overview Refresh buttons refetched GET /accounts, which the backend answers from a five-minute probe snapshot, so a key fixed after the page loaded kept its failure. Refresh now asks for ?refresh=1 and writes the answer into the list's one cache entry. That re-probe also skips the identity probe's 30-second cache, so a key fixed seconds before Refresh is checked again; ordinary reads keep both caches. A failing key showed only a warning badge and Reconnect, never the probe's own error. Reconnect now leads with that error (already redacted server-side) through AwsErrorNotice, so the cause is on screen wherever Reconnect is offered. Closes #29229 (cherry picked from commit 8373101)
…32381) A cancel landing inside the spawn await of the git auto-update's reset, kiro-cli update or core-dep pip install made asyncio SIGKILL the leader pid alone, so the child's process group kept running against the install. These three spawns now start through update_provider's _start_installer, which keeps the spawn running past the cancel, waits for it under the update stop's bound, then stops the child. A new stop= keyword lets these children use _kill_and_reap, since they have no TERM trap to honour. Closes #26795 (cherry picked from commit 136c4cc)
…32270) A desktop window opened on a configured Remote Crew (client mode, local gateway off) still exposed the crash-report, WSL, file-open, folder, terminal, first-run, pane-cache and update-governance bridges, so the remote dashboard called channels the shell then refused and logged on every load. The window now passes --kc-remote-gateway to its preload, decided by the same per-window predicate the main-process gate reads, and the preload withholds exactly the channels that gate refuses. The gate is unchanged and stays the authority. Builds on #16566. (cherry picked from commit 31a47a0)
On a Linux host with a glibc older than 2.27 (Amazon Linux 2: 2.26), the locked pillow==12.3.0, which browser-harness pins exactly, has no wheel, so pip builds its sdist. Without a C compiler, Python.h, zlib.h or jpeglib.h that build fails, and the install error keeps only pip's tail, "Failed to build pillow", which names no cause. Check for them first, from the filesystem, and refuse at the preparing stage, before the marker or the venv is touched, naming what is missing and the yum or apt-get command that installs it. A venv that already holds the locked pillow is not checked. Closes #32241 (cherry picked from commit d3aff60)
…2519) The retired knowledge folder-discovery keys were dropped only by save(). The Settings PATCH and PUT and `kirocrew config set` edit config.json directly, and the raw-write drop helper covered just the two retired keys that warn at load, so the five knowledge keys stayed in the file forever. The drop map is split into its RENAMED and RETIRED halves, the helper drops every RETIRED key (a renamed spelling is still read, so it stays), and `config set` calls it too. Closes #30337 (cherry picked from commit ba92caf)
(cherry picked from commit 5226eaa)
* fix(usage): count KAS sessions on the Usage page The Session Activity card and Daily History table read only kiro-cli's sessions/cli transcripts. The KAS harness writes each session to sessions/<id>/sess_<uuid>/messages.jsonl in another row format, so an install on the KAS backend showed almost no sessions and no messages or tool calls, while the same days showed real credit spend. The parse now also lists the KAS records beside the cli directory and counts their user rows, Say assistant rows and tool_call rows. A KAS record without a messages file is skipped, not reported as a failed load. Each transcript's counts are kept keyed on its mtime and size, so a refresh reads only the transcripts that changed. * fix(usage): list Windows cloud-files and dedup session folders Pin the KAS root and sessions/cli with links_only, as the upload path does, so a cloud-files or dedup folder is listed instead of read as missing or unreadable. KAS record directories opened through the root pin inherit links_only, so they are listed too. Links and junctions are still refused. Co-authored-by: Ray Xu <buluoray@gmail.com> --------- Co-authored-by: Ray Xu <buluoray@gmail.com> (cherry picked from commit 4182398)
(cherry picked from commit 91bc37b) # Conflicts: # docs/decisions/README.md
(cherry picked from commit ab349e1) # Conflicts: # website/src/i18n/locales/bn.json # website/src/i18n/locales/de.json # website/src/i18n/locales/en-XA.json # website/src/i18n/locales/en.json # website/src/i18n/locales/es.json # website/src/i18n/locales/fr.json # website/src/i18n/locales/hi.json # website/src/i18n/locales/it.json # website/src/i18n/locales/ja.json # website/src/i18n/locales/ko.json # website/src/i18n/locales/pt.json # website/src/i18n/locales/ru.json # website/src/i18n/locales/zh-CN.json
The ACP provider's hooksSection ("ACP Agent Hooks") and sessionProcess
("ACP subprocess") labels were fixed English literals shown in or
inside translated copy: the Hooks page section heading, its aria-label
and its load-failure and empty-state titles, and the Agent Config tab
tip. Every other language showed the English words.
Both labels are now getters that read a catalog key at use time, as
pluginRegistryName already does, translated in all 12 locales. The
Hooks page no longer lower-cases the section name, which turned the
ACP acronym into "acp" and is wrong for translated nouns.
(cherry picked from commit c3b50d3)
# Conflicts:
# website/src/i18n/en.context.json
Follow-ups from the review of the app pop-out window. - The sidebar's pop-out action no longer stops keydown, so the shell's document-level shortcuts fire while it has focus; the rail has no keyboard drag sensor that needed the stop. - An Ask the agent in an app pop-out hands its prompt to the main window without committing /chat in the pop-out, so the app is not remounted and no /chat entry is left for Forward to replay. - A POP onto a location that is not the pinned app (a deep link such as /apps/foo/extra, a differently cased route, a Forward) settles the window on the app without handing anything to main. - Code Review Sage's "Open shared copy" is a router Link, so in a pop-out it opens in the main window instead of loading the dashboard in the pop-out. - The action's tooltip and accessible name are one string per state that names the app; the title bar's Return says what it does (go back to the main window and close this one), with the artifact pop-out's wording and translations. The strings move to their components' namespaces. - The routing, sidebar, error hand-off and pop-out applier specs describe the app pop-out. (cherry picked from commit 3535135) # Conflicts: # website/src/i18n/locales/en.json
Settings search offered no row that opens the Feishu channel. The Feishu panel is drawn by the shared bot-channel panel, but the registry extractor fanned that panel's rows out only to Discord, Telegram and WeCom, and the Feishu spec's App ID / App Secret labels were never indexed. Fan the shared rows out to Feishu, index the two credential fields, and drop the coverage waiver that claimed the fan-out already happened. (cherry picked from commit 3c1e70c) # Conflicts: # src/kiro_crew/docs/ui-index.generated.json
f69df06 to
7563c2c
Compare
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: 0.12 users get twenty-two user-facing fixes that are on main but not on the 0.12.0 line, in 0.12.1.
Each fix below merged to main and was checked as not on
release/0.12.0(no PR number, nocherry picked fromline, no matching patch-id, and the pick still changes the tree). Each is something a 0.12 user hits today.Why it matters
These are bugs 0.12 users hit now: a deb that will not install on Ubuntu 24.04+, KAS sessions missing from the Usage page, a first reply that stalls after every restart, a cancelled update that leaves its git child running, lost crash dumps, and focus that jumps off buttons while they run. Without a backport they wait for 0.13.
Not a goal
What changed (motivation → approach → change)
Every commit is
git cherry-pick -xof the main merge commit ontorelease/0.12.0at7a89cc18d60f581d994b18afd6a3d0c8ef0b0e23, in main merge order. The commit author is the release line's public identity, as on every other commit on this branch (the repo's leak guard refuses the original author emails); thecherry picked from commitline in each message names the source.Picked cleanly (16):
Picked with conflicts only in docs, tests or translation catalogs (5). Each was resolved by keeping release's existing entries and adding only the entries the PR itself adds:
docs/decisions/README.md: kept release's rows, added this PR's 2026-11-07 row (the other row main has there belongs to a decision not on this line)website/src/i18n/locales/*.jsoncatalogs: kept release's keys, added this PR's threeinvite_email_hint*keyswebsite/src/i18n/en.context.json: kept release's entries, added this PR's two ACP label entrieswebsite/src/i18n/locales/en.json: kept release'sdirect_link_note, added this PR's newshell.nav.appRowPopoutActionblocksrc/kiro_crew/docs/ui-index.generated.json(generated): regenerated withnpm run gen:uion the picked source, which adds only the Feishu settings entriesDropped (not in this PR):
instances.mdchanged differently on each branch; would need hand-merged prose.trusted_channel_command,{{cmd}}to<mono/>), not a new entry.cli.mdstop/restart rows changed differently on each branch; would need hand-merged prose.read_endorsed_config_textinconfig/loader.py, which is not on release; its test fails to import.actwithout the import main's file has; 5 tests fail on release.quick_prompts.user_quick_prompt, which is not on release; 26 tests fail.update_schedule.py), which only a human override clears; main has the same code.Backwards compatibility
Compatible: each commit is the fix main already ships, with no new keys, routes or settings beyond what its PR added.
Removes nothing: the only removals are those the picked PRs make on main (#32865 replaces its own pop-out catalog keys and code together).
Tests
Run on this branch, one process at a time, no xdist:
uv pip install -e . --group dev),PYTHONNOUSERSITE=1, one file at a time,-n0: all pass (run on the 22-pick head; after dropping fix: keep the crash dump of a gateway that died without a clean exit #32205 the slack gateway, update-apply and sub-agent injection files were re-run and pass).src/i18n: 71 files, 1197 tests pass.node --require ./test/_preload.cjs --teston the 4 touched electron test files: 127 pass.npx tsc -p tsconfig.app.json --noEmit: clean.npm run i18n:checkagainst the release head: 19 checks PASS.node scripts/check-i18n-keys.mjs: all references resolve.Proof: not provable -- other: a release backport; each fix was proven by its own PR on main
Manual verification
N/A — each fix is the change main already ships and was verified there; the tests above re-run on the release line.
Screenshots / video
Why no screenshot: every UI change here is a backport of a main PR that carried its own screenshots, and the backport changes nothing beyond it.
Related Issues
Backports #31572, #32295, #32448, #32414, #32501, #32485, #32510, #32444, #32381, #32270, #32251, #32519, #32481, #26951, #25315, #32536, #32332, #32315, #32545, #32865 and #32631 to the 0.12 line. Sibling batches: #33529 and #33530. A rebase may be needed once those land.
Sibling sweep
Siblings: none -- a backport batch; each source PR carried its own sweep on main.
Pattern harvest
Not generalizable: a release backport of fixes already harvested on main.
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.