Repository navigation
fix: last 0.12.1 batch (24 fixes merged 11-08 16:05-18:56Z) - #33364
Conversation
…start on it (#33222) KAS reads no cli.json, and carrying chat.defaultModel safely needs a pin mark on the session handle (#33023). Until then an unpinned KAS chat keeps main's model behaviour, and the session gets one notice that names the kiro-cli default model and says how to pick it in Crew. Original author: Joe Guo (cherry picked from commit 04fbd9e)
The log viewer kept an accepted level change in its own state. The dashboard's query client never refetches a cached read, so after the Logs page remounted, or in a second viewer (the Activity panel's Logs tab, the Developer panel), the selector showed the old level and the lines were filtered by it while the backend logged at the new one. The accepted level now goes into the shared level query, so every viewer and every remount shows and filters by the level in force. Original author: Rohan Khanderia (cherry picked from commit ec9d3f9)
…#32935) The nine non-Slack dispatchers took agent.default_agent from the config handed in at channel start. The field carries no restart mark and Slack and the dashboard follow it live, so a dashboard change of the default reached Discord, Teams, Telegram, Webex, Feishu, iMessage, WeCom, Weixin and WhatsApp only after a gateway restart, and the default_template fallback for a removed spec never applied there. Each dispatcher reads the default at the point of use through agent_switch_command.LiveDefaultAgent, which goes through default_template and keeps the last value while the agent section is discarded, as Slack's default-agent applier does. A change reaches a conversation when its next conversation starts, as a dashboard default reaches the next new chat. LiveDefaultAgent.for_route pins the default a route's conversation first reads in ConversationState.pinned, and an open conversation keeps it until /new, an idle or daily reset, or an /agent switch drops the pin. Without the pin a default changed while a message waited for admission ran the new agent under the old agent's key, and one changed while a reply ran moved the key /stop and the busy check derive away from the running turn. Telegram also reads the agent with the key before admission and hands it to _run_turn, so a /new during the wait cannot change it. An /agent reset to the default switches to the live default, compared against what the current conversation runs. The agent is part of the session key, so a new conversation can change bucket. ConversationState takes a bucket_fn (the agent a conversation runs) and drops the pin and reseeds the counter from the new bucket before a rotation applies, so a /new lands past all of the new agent's saved generations. A route with an /agent pick keeps its agent and its counter. Resetting an /agent pick to a default that names the same agent now drops the pick instead of answering "Already using" and keeping it. The per-channel kirocrew fallback constants were unreachable behind default_template and are removed; the Discord and Telegram test helpers prime the agent section beside their own. Original author: Bolin Chen (cherry picked from commit b2b235e)
On WeCom and Teams a configured person's DM target is available only after they have written to the bot. The owner rule counted only available targets, so with two configured people the one who had not written yet was skipped and the other was named the owner. The rule now counts every configured direct target, available or not, and names the owner only when there is exactly one and it is available. That is what the helper, the session-control spec and the cross-channel owner notice already state. On Teams, an owner DM with two or more allow-listed people now falls back to the dashboard notification. The decision is recorded under docs/decisions/. Original author: Bolin Chen (cherry picked from commit 7c3dd18)
…33144) The stored-context source note and the test-request result were each built from two catalog pieces placed around a <code> element, so no language could order the sentence its own way: Hindi and Bengali read verb first, Chinese lost its verb, and French, Portuguese and Russian got a space before a period and a participle that does not agree with "session". Each sentence is now one catalog message rendered with <Trans>, carrying the code element as markup, written in every shipped catalog. English reads the same. Closes #32766 Original author: Bolin Chen (cherry picked from commit e3f7215)
…3240) The row's Test button was disabled while a test was in flight, so the keyboard press that started it lost focus to the page body. It now uses aria-disabled and aria-busy, as the Rescan button on the same page does; handleTest's in-flight claim already keeps a second press inert. Original author: Rohan Khanderia (cherry picked from commit 95f25ed)
…ng parent (#21467) * fix(chat): tell same-named @ file results apart by their distinguishing parent The @ file menu's path sub-line cut at the END of the path, so when one result set held several same-named files (six README.md under a shared prefix), the cut landed inside their common parent chain and the rows read identical down to size and age — a user could pick the wrong file. Since #21152 ChatEmbed mounts the same menu. disambiguatedPaths now gives every row of a same-named group the shortest suffix of parent directories that tells it apart from the group, head-anchored with an ellipsis (…/core/src/README.md), so the distinguishing segment is the first thing after the cut and survives the narrowest menu width; separator-aware for native Windows rows. Unique-named rows keep the full path, and the row tooltip keeps it everywhere. Closes #21162 * fix(chat): swap path separators with replaceAll, not a regex Review follow-up (bolichen97, 2026-11-05): disambiguatedPaths now uses exact-string replaceAll for its two single-character swaps. The screenshot-only commit is dropped from the branch so no PNGs land on main. --------- Co-authored-by: iamwhatever <iamwhatever@users.noreply.github.com> Original author: akushonkamen (cherry picked from commit 5c1585f)
On the Discover Updates list, Update All and the pressed row's Update took native disabled while their update ran. Update All stays on screen as the batch's progress label, and a row whose update fails stays on the list, so disabling the pressed button blurred it to the body and a keyboard user lost their place. Hold the pressed button with aria-disabled and refuse a second press in its handler; the rows nobody pressed stay disabled while anything runs, and Update All is disabled while a row's own update runs, so a batch cannot post that app a second time. Fixes #33236 Original author: Rohan Khanderia (cherry picked from commit 3f6fad4)
…3247) The Artifacts page's kind filter labelled each option with an English literal, `kind: ${k}`, while the tag filter beside it reads its "tag:" prefix from the catalog. Under German the two dropdowns read "kind: markdown" next to "Tag: ops". The option label is now one catalog string, "kind: {{kind}}", with the word translated in all 12 locales. The kind id itself stays as written, as the tag names do. Original author: Rohan Khanderia (cherry picked from commit c72e584)
An expanded sub-agent node in the Session Breakdown tree labelled each
turn with a hard-coded English "t1", "t2". Every locale showed the
English letter, the number skipped locale formatting, and a screen
reader read a bare "t one". Show a per-locale short mark from the
catalog, which still fits the narrow column, and give assistive tech
the whole word ("Turn 1", in each locale's wording).
Original author: Rohan Khanderia
(cherry picked from commit b306a0b)
On KAS, an agent spec that lists the shell (or `code`) in `allowedTools` and also sets a boolean `autoAllowReadonly` on its shell `toolsSettings` entry lost both grants, so every shell and `code` call prompted. On kiro-cli a shell named in `allowedTools` runs whatever its `deniedCommands` do not deny, and `autoAllowReadonly` only adds read-only commands, so it restricts nothing there. The `allowedTools` reading now leaves a boolean `autoAllowReadonly` out of the authored restrictions, as the `allowedCommands` answer already does. `denyByDefault`, every other key and a non-boolean value still keep the grant at the prompt, and a `permissions` block naming a policy bundle is still refused whole. Fixes #30165 Original author: Bolin Chen (cherry picked from commit 533952a)
…33202) An app installed from a folder (`kirocrew app install <dir>` or the dashboard's install from path) was written with the record default origin "registry", so its details, `GET /api/apps` and the trust dialog all claimed it came from the Kiro Crew registry. install_app records "registry" only when a registry install scoped a repository, and "local" otherwise. update_app records "registry" when a registry update replaces a record no built-in owns, so an app first installed from a folder does not keep "local" once registry code runs in it. The trust dialog shows an installed local app as "Source: Local install", its review note no longer says Kiro Crew lists the app, and a same-named registry listing no longer lends a local install its registry name there. The dialog that switches a folder install to its repository still names the listing, since it trusts the repository. The new note ships in every catalog. Builds on #29933. Original author: Bolin Chen (cherry picked from commit de33aea)
On Linux the sweep's pin scan of /proc now runs in a fresh interpreter (python -P -m kiro_crew.sandbox_pin_scan_child keyed|legacy) that writes the pinned names and the coverage verdict as one JSON line. In the gateway every /proc syscall of the scan had to take the GIL back, so on a busy host the scan slowed itself and every other gateway thread. A child that cannot start, runs past 300 s, crashes, exits non-zero or replies malformed counts as an incomplete scan with coverage unproven: the pass reclaims no directory, as an incomplete in-process scan does. The scan also stops reading another real user's tasks; they cannot bind a source this uid staged, and they now lower only the host-wide flag. Other platforms and any other proc root keep the in-process scan. Closes #33171 Co-authored-by: Raymond Chen <chenmingwei23@users.noreply.github.com> Original author: Raymond Chen (cherry picked from commit 2da9ec8)
- Honour includeMcpJson when modelling an agent's MCP servers. An agent whose spec opts in to the user's mcp.json now has those servers in the set a cron waits for, the unresolved @server ref guard, a sub-agent's first prompt and /api/mcp/active, so a slow mcp.json server is waited for and no longer reported as declared but not configured. - Keep an edition MCP server's own command, args and env when a global mcp.json entry has the same name. A narrower or args-less global entry no longer replaces the edition's tool selection in rebuilt specs, so chats keep the tools the edition enables. - Recreate the gateway's default MCP backend work dir before a spawn when it has vanished. Pooled servers no longer fail every cold spawn with ENOENT and fall back to per-session exec until a restart; any other missing cwd still fails the spawn. - Re-attempt a timed-out KAS managed MCP server (kirocrew-core, kirocrew-cron) once at session start through _kiro/mcp/resetServer. A busy host that misses the initialize deadline no longer fails every chat; a second failure or any other failure still fails the start. Fixes #32971 Refs #30915 Original author: Bolin Chen (cherry picked from commit 57ecf6f)
…on (#33320) The Projects page planning banner and the run detail's planning overlay append dots to "Generating execution plan" from a 500 ms interval. That is JS-driven text, so the global reduced-motion CSS cut never reaches it, and the line kept ticking for a user who asked for no motion. Both now read the live preference through `useReducedMotion` and hold the line still while it is set; the static subtitle below still says the plan is being built. Original author: Rohan Khanderia (cherry picked from commit 5dae400)
A goal loop the service stood down on consecutive_failures (several cycles in a row failed) or model_refused (the model declined a cycle) showed only the bare "Paused" line in the warn tone, because the goal popover's status lookup knew neither reason. Both reasons now have their own line in every locale and render through ErrorNotice, with no hand-off beside the unsaved goal fields, like the other failed-cycle reasons. Closes #32530 Original author: Bolin Chen (cherry picked from commit 0bca0bc)
…33256) The Personal Shopper Sites tab Remove button was disabled while the replace-all save ran. A disabled button leaves the tab order, so a keyboard user who pressed it lost focus to the page, and a failed removal left them nowhere. Mark it aria-disabled while busy; removeSite already refuses the press. Original author: Rohan Khanderia (cherry picked from commit 7088d8a)
…33326) The import preview's origin line glued two catalog fragments around the manifest's raw UTC timestamp, so every language showed an ISO string like 2026-07-30T15:04:00Z. It is now one key with {date} and {origin} placeholders, the date formatted with fmtDateTime for the reader's locale, and the two fragment keys are gone. Original author: Rohan Khanderia (cherry picked from commit 6bdfdfa)
…#33158) On Windows os.path.commonpath raises ValueError for paths on two drives. The crewmate import's file step compared the resolved skills folder and the imported-scripts folder with it, so a junction to another drive escaped as an unhandled error (a 500) instead of the workspace_link / scripts_dir_link 409 the route promises. Both checks now go through one local containment test that treats a path on another drive as outside, the shape the other commonpath sites in the tree already use. The refusal rule, codes and messages are unchanged. Closes #32693 Original author: Bolin Chen (cherry picked from commit 768a29a)
Sensitive Change ReviewSensitive files changed:
How to clear this check:
Only the author's comment counts this way. A reviewer can instead |
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 images, yet it changes several dashboard surfaces users see. Evidence gaps
[UX-REVIEWED] 88f2cd0 Description read[DESCRIPTION-READ] 1af03be981522bf69999abdfa8063d676448af4cc4d32ebc67f55cc9f451f8cd 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 backports 24 already-merged main fixes one commit each, with every hand-resolved hunk written down. The removed or re-signed helpers ( Notes
[DESIGN-REVIEWED] 88f2cd0 Description read[DESCRIPTION-READ] 35c40b7aa3d078f306d0289de56b7012cb485793333efca6cd2f4043d11b803b 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 is a backport of 24 fixes already merged on Notes
What this change shipsInventory (10 items) — 9 justifiedIntent: ship 24 fixes already merged on
[FIRST-PRINCIPLES-REVIEWED] 88f2cd0 Description read[DESCRIPTION-READ] a0c02ee4d9e8925bb565061c133b5e2e64eb5b6c5a4f1d5de7679da03aae4f5c 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 fixes merged on
mainon 2026-11-08 (between 16:05 and 18:56 UTC) ontorelease/0.12.0for 0.12.1, one commit each, in a PR of their own.Why it matters
Each fix changes behaviour a 0.12.0 user can hit today. KAS chats name kiro-cli's
chat.defaultModelwhen they do not start on it, keep a shell grant beside a booleanautoAllowReadonly, and re-attempt a timed-out managed MCP server once instead of failing the chat. Agents that opt in tomcp.jsonget those servers waited for and modelled. Non-Slack channels read the default agent live, and a channel names its owner only from one configured DM target. A folder-installed app is no longer labelled as a registry app. A cross-drive Bot import folder gets the 409 refusal instead of a server error. The rest are focus, reduced-motion, log-level and translation fixes on dashboard pages.Not a goal
main; nothing is changed only here.CHANGELOG.mdchange.What changed (motivation → approach → change)
Each main squash commit was cherry-picked onto
release/0.12.0@2196761088, one commit each inmainmerge order, with the(cherry picked from commit …)trailer. The repository's leak guard requires the public identity, so every commit is authored as Ray Xu and names its original author on anOriginal author:line.04fbd9e1092746a29cc83568061617972970802bec9d3f9aa4f557bc8efdf7eca77d80ffe596e7afb2b235ee76bbcd82a863903b71d4f0ce9e052aa57c3dd188e8cc38ba3cd5e3f7215702e4afb589eff3bcdeb27fcb3d767c8295f25ed3df9e66332cea5c1585fd4b3d05500b0b3f6fad4c7783a8f3a5a0c72e5849b4ffb664565cb306a0b0224b947d53ef533952af9fc37104ad14de33aea51133292188a42da9ec838d16db7005cc57ecf6f27e07575a9a235dae400e679802fc89d90bca0bcd5a538752cf4d7088d8a645ffb61e57328b8f5e62abe1ae2a21846bdfdfabaa03f29d3035768a29a3f888f2cd0d59Hand-resolved hunks:
src/kiro_crew/feishu/transport_dispatch.py:134: adds onlyself._default_agent = LiveDefaultAgent().main's adjacentdashboard_statefield comes from another change and is not brought in.src/kiro_crew/mcp_tools/messaging.py:50-56(comment): takesmain's wording and keeps 0.12's Feishu clause, because 0.12's Feishu still marks its targets unavailable.docs/decisions/README.md:139: adds only this PR's decision row.docs/system-specs/modules/file-search.md:429(docs): keeps 0.12's "ChatEmbedopts in" and adds the same-name disambiguation clause.website/src/i18n/en.context.json:4: adds onlypages.artifactsPage.kind_option.website/src/pages/appDetailPageLocalSwitch.tsand oneAppDetailPageAutoAction.test.tsxblock: the conflicts with fix(apps): let Update move a local-only grant to the app's repository #30979's "Trust the repository and update" switch dialog were left on 0.12's side, because 0.12 does not have that dialog, so this PR's one test for it is not brought in. The origin fix is kept: on 0.12apps/manager_install.pyrecordsorigin="local"only for a marketplace source and otherwise falls back toInstalledApp.origin, whose default is"registry", so a local-folder install is recorded as a registry one. fix(apps): record a local-folder app install as local, not registry #33202'smanager_install.py,manager_records.py,TrustAppModalanduseAppActionschanges fix that, andtest/test_app_manager.py,TrustAppModalRegistryTier.test.tsxandappstoreCategories.test.tscover it.src/kiro_crew/acp/runtime_session_create.py:520: keeps 0.12's plainreturn await self._finish_create_session(...)and addsmcp_json=mcp_json,.main'stry/exceptaround it uses_unclaimed_new_sessionsfrom another change.website/src/pages/overview/PortabilityTab.tsx:9: import line resolved tofmtBytes, fmtDateTime(0.12 does not importfmtListthere).One file under
.github/changes: #33074 removessrc/kiro_crew/mcp_gateway/backend.pyfrom.github/black-baseline.txt, as onmain;black --checkpasses on that file here.Left out of this batch:
ReportProblemModal.tsxon keys added by #31712 (diagnostics plural counts) and #30156 (Security rule count), neither of which is on 0.12. That is a prerequisite chain under a 28-file rewrite, so it waits.Backwards compatibility
None: each fix keeps its main PR's compatibility notes; nothing here changes stored data or config beyond what those PRs already shipped on
main.Tests
Proof: not provable -- other: a backport of main commits whose proofs ran on their own PRs
release/*, so the related set ran here: 75 Python test files in three batches, one pytest process at a time with no xdist, withTMPDIRoutside~/.kiro/crew. These are the changed test files plus the related atomic-write, owner-only files, MCP gateway backend, MCP probe, MCP session report, spawn audit, owner DM, channel transport, ssh tunnel, sandbox mount sweep and Bot import route suites: 3,498 passed, 2 skipped, 0 failed on this head (88f2cd0d59).check-i18n-strings.mjsandi18n-check.mjspass.vitest run src/i18n/: 1,193 passed, andnavLabels.test.tsxtimed out once under load, then passed 8/8 alone; no pick touches it.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 #33222, #32942, #33262, #32522, #32935, #33049, #33273, #33144, #33238, #33240, #21467, #33243, #33247, #33258, #33274, #33202, #33195, #33074, #33320, #33190, #33256, #33317, #33326, #33158.
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.