Repository navigation
fix: backport 0.12.1 fast batch v1b-5 (19 fixes, docs/tests conflicts resolved) - #32936
Conversation
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 diff changes several visible dashboard surfaces, but the PR attaches no image, so the blind read did not run. Evidence gapsThe blind-read report says: "BLIND READ NOT PERFORMED: the PR description carries no image attachment". The screenshot list and the recording list are both empty. Each surface below needs a still attached to the PR description:
Notes
[UX-REVIEWED] 4de6fbf Description read[DESCRIPTION-READ] aa2497c8faa1c5d6fe5d116723724e2b85c3280b3e34b33b55fc325aee8b45cc 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 This change only backports 19 fixes already merged on What this change shipsInventory (10 items) — 10 justifiedIntent: give 0.12.0 users the fixes already merged on
[FIRST-PRINCIPLES-REVIEWED] 4de6fbf Description read[DESCRIPTION-READ] 557f711f3592b8336df3465b36137151247272841ffb2dcf322649808709c846 That sha256 names the PR description this verdict read. Recompute it with |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: these are the merged Notes
[DESIGN-REVIEWED] 4de6fbf Description read[DESCRIPTION-READ] 14c82470b8930953cf58328ed1c4c32966daf6cdb90d579490264bbd1e1cfd9e 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 |
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: |
868ebc3 to
cdea497
Compare
…0189) The harness prompt says to put independent calls in one block, and models judge "independent" by syntax. So they send a verdict with the eval that produces it, or a done report with the poll that proves it. Define "dependent call" once (_DEPENDENT_CALLS_RULE) and carry it in the goal conductor (as a shared ledger rule, so a crewmate's conductor gets it too), the pipeline and security conductors, and the worker. Say the verdict record is a later step in the conductor prompt and the goal-conductor skill. Add one sentence at the point of use to the work_report and work_ledger_record descriptions. Closes #30124 Co-author on main: Joe Guo (cherry picked from commit 6a13aeb) Co-authored-by: Joe Guo <12429733+iamwhatever@users.noreply.github.com>
…30084) An unpinned or "Default" claude session serves claude-agent-acp's `default` pseudo-model (the head of the advertised list) or reports no model id. The model registry recognises neither, so supports_effort answered False: the dashboard hid the effort control and a configured level was dropped at start. Add the opt-in set ACP_BACKENDS_DEFAULT_PSEUDO_MODEL (claude). While a member serves no concrete id, the effort select the adapter built for the served model answers whether a level applies, and the start-time resolve and change_effort follow that answer. Concrete ids keep the registry answer. The pre-session capability read answers null rather than false for an auto or empty model on a member. Closes #30070 (cherry picked from commit 072817a) Co-author on main: Bolin Chen
…or (#30110) A Claude session runs on its own AcpClient, which has no reset that starts a fresh conversation on the live process; only AcpSessionProvider has one. AcpProvider.new_conversation reached the missing method as an AttributeError behind a type ignore, the pool logged the fallback at DEBUG, and three docstrings claimed claude already had the cheap reset. new_conversation now raises NotImplementedError naming the backend, the workflow pool logs the resulting cold start at INFO, and the docstrings and the workflows and claude-code-provider specs say a claude worker is reset by kill and respawn. The cheap claude reset itself stays open on #30099. (cherry picked from commit 942ab8b) Co-author on main: Bolin Chen
…#30497) PATCH /api/crons/{id} normalized agent, agent_id, channel and timezone with (value or "").strip(), so a truthy JSON number, list or object raised AttributeError and the request answered 500. Refuse it with a 400 naming the field before anything is written. A falsy value still clears the field, as before. (cherry picked from commit dcf58af)
#30511) Three messages on the Issue Radar tagging view were built by gluing translated fragments around the repo name: "Couldn't load" + repo + "'s labels:", repo + "defines no labels yet...", and "Every open issue in" + repo + "carries at least one label". Languages that put the repo elsewhere read out of order: Japanese, Korean, Chinese and Bengali reversed the sentence, and German showed "o/rs Labels:". Give each message one key with a {{repo}} placeholder and a whole translation, and drop the five fragment keys. (cherry picked from commit 73b40a9) Co-authored-by: Rohan Khanderia <51836404+RohanK6@users.noreply.github.com>
POST /api/crons/{id}/ack passed summary to the store, which keeps
summary[:500]. A JSON null, number, boolean or object has no slice, so
the request answered 500. Refuse those with 400 invalid_summary before
anything is written. A string, a list and an absent summary save as
before.
(cherry picked from commit 08357c7)
validate_cron_expr handed its input to croniter, which calls .lower()
on it. PATCH /api/crons/{id} passes the body's cron through unchecked,
so a JSON number, list or object raised AttributeError and answered
500. Read a non-string as invalid, so both store writers raise their
ValueError and the route answers its 400.
(cherry picked from commit 4b5eee0)
On the wide notifications page the note detail sits beside the feed. Its Close, Approve, Reject, cron ack or action link swapped the panel for the empty state and focus fell to <body>, so a keyboard user had to Tab back from the top of the page. The shared detail-close focus hook only saw the narrow layout's detail; it now sees the selected note in either layout, so focus returns to the row that opened the detail, else the first row. With that hook live beside the feed, the feed's clear-refocus no longer waits on <body> once the clicked control has unmounted (an expanded stack's header goes when one note is left): it hands focus to the next row at once, so the page hook does not send it to the first row. Closes #29175 (cherry picked from commit e7e3284) Co-author on main: Bolin Chen
…31461) On macOS every audio decoder except a fixed system one runs from a snapshot staged under the data home's run/ root, which every agent sandbox denies. Run from an agent's shell, kirocrew doctor therefore failed to stage an authentic decoder, logged it as a digest mismatch, reported "ffmpeg: not found" and told the user to reinstall. An OSError, which only reading or staging the decoder raises, now propagates out of _authenticated_ffmpeg, and _open_authenticated_in logs it as a decoder that could not be read or staged, naming the errno's system text and never a path. A size, digest or signature refusal keeps the digest message, and the set of decoders that authenticate is unchanged. From a confined shell on macOS, doctor prints "not checkable from this shell" for a decoder it did not find, adds no issue, and points at the user's own terminal. Linux stages into a memfd, so a miss there keeps its verdict. Closes #27030 (cherry picked from commit d6fc73c) Co-author on main: Bolin Chen
…ls (#31459) The top-bar credit surface was keyed on the kiro-cli backend alone, so on KAS, the default harness and one that draws on the same Kiro account, a failed usage read hid the segment and left no way into the account modal. Decide the surface by the Kiro harness set (kiro-cli and KAS) instead, so the failed read keeps its "unavailable" dash and modal there, while other harnesses still hide it. Closes #29795 (cherry picked from commit 963d468) Co-author on main: Bolin Chen
…31440) A theme pack's welcome image (branding/welcome.*) rendered as a plain <img>, so an animated GIF, WebP or SVG kept playing when the user asked for reduced motion: the global rule stops the page's own animations, not an image's frames. The welcome screen now renders it through PackLoaderImage, which already paints loader art once onto a canvas under reduced motion. The welcome hero is hidden on a short viewport, so a still painted while hidden now takes its size from the canvas's CSS box instead of the 14px carousel fallback and stays sharp when the hero shows. The themes spec, theming contract and feature map say so. Closes #23841 (cherry picked from commit 1674b1f) Co-author on main: Bolin Chen
) The trust menu put the pending command and its program name into <Trans> as interpolation values. <Trans> parses its translated string as markup, so a command holding tag- or placeholder-shaped text rendered as something other than itself on the consent surface: `echo <i>x</i>` showed an italic x. The command and the program name are now the <mono/> chip's own child, as the approval card's confirmation line already does, and every catalog carries a <mono/> slot in those ten strings. The guide's catalog matcher counts a self-closing slot as a run-time part, so a translated trust row still maps back to its trust key. Part of #22834 (cherry picked from commit f3dfda0) Co-author on main: Bolin Chen
#32135) The board lane's hidden-folder notice says "Show" and its click opens the sort-and-filter menu with the Folders section unshelved. Folders sit last in that menu, so when the menu is taller than the room below the funnel the section the word promised opened out of sight, and a keyboard reader landed on the menu's first row. The notice now opens the menu for its folder list. The Folders section focuses its heading and keeps itself in view while the menu settles its height, until the reader scrolls, presses or types. Opening the menu from the funnel is unchanged. The keep-in-view helper is the one the side panel's List all tabs menu already ran, moved to a shared module so both menus use the same code. Part of #14087 (cherry picked from commit 5b599b7) Co-author on main: Bolin Chen
The Your crews / Set up a new one row in Settings > Remote Crew marked the open tab only with a border colour, so a screen reader heard two plain buttons and no selection. Wrap the row in the shared RadioChipGroup and give each option radioChip state, with a named group in all 12 catalogs. Tests and screenshot capture scripts that found the options by the button role now find them as radios. (cherry picked from commit 1c2a0bc) Co-authored-by: Rohan Khanderia <51836404+RohanK6@users.noreply.github.com>
…#32163) The Refresh buttons on the Skills and Steering tabs set disabled while the list reloads. A disabled button leaves the tab order, so a keyboard user who pressed it lost focus to the page for the whole refresh and had to tab back from the top. Both now mark themselves aria-disabled and ignore a click while busy, the way the App Store and file browser refresh buttons already do. The busy name still says the list is refreshing. Neither tab read its list query's error, so a refresh that failed looked like one that found no changes. Each tab now shows that error in an ErrorNotice above the list, without the agent hand-off, since an open edit may still hold an unsaved draft. (cherry picked from commit f057395) Co-authored-by: Rohan Khanderia <51836404+RohanK6@users.noreply.github.com>
_logout and _update_approve in cli_server_requests.py printed every failure line (gateway not running, busy, ownership refused, HTTP error, not-ok body, no armed request) to stdout before exiting 1, so a stdout redirect swallowed the reason and a script reading stdout parsed the error as a result. The sibling _token handler already sends refusals to stderr; route these two to match it, leaving the success and informational lines on stdout. No wording, exit code or other behaviour changes. (cherry picked from commit c5dfd26) Co-authored-by: Rohan Khanderia <51836404+RohanK6@users.noreply.github.com>
The ACP provider's plugin registry name was a fixed English literal, "Packages", interpolated into translated copy: the Prompts tab description on Agent Capabilities, the Prompts card tip and the Skills tab's read-only package group. Every other language showed the English word mid-sentence. The label is now a getter that reads a catalog key at use time, so it follows a language switch, with the word translated in all 12 locales. The Prompts card tip no longer lower-cases it: case-folding a translated name is wrong in languages that capitalize nouns. (cherry picked from commit 0f18737) Co-authored-by: Rohan Khanderia <51836404+RohanK6@users.noreply.github.com>
…32521) A global mcp.json that exists but cannot be read (permission denied, a directory) was refused by the MCP writer routes with "cannot parse global mcp.json" / config_corrupt, the same reply as bad JSON, which sends the user looking for a syntax error that is not there. Rebuild (POST /api/mcp/sync), PUT /api/mcp/servers/{name} and POST /api/mcp/apply now answer 500 config_unreadable ("cannot read global mcp.json") for a read failure and keep config_corrupt for a file that does not decode or parse. Closes #30355 (cherry picked from commit 22a9144) Co-author on main: Bolin Chen
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: |
cdea497 to
4de6fbf
Compare
Problem / Motivation
Goal: Bring nineteen 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. Their code applies to
release/0.12.0cleanly; where a pick conflicted, the conflict was only in a doc, a test, translator notes or a generated index.Not a goal
main; nothing is changed only here.CHANGELOG.mdchange.website/src/i18n/pluralKeys.jsonand is left out.What changed (motivation → approach → change)
Each main squash commit was cherry-picked with
-xontorelease/0.12.0@193db41120, in merge order onmain, one commit each. Three applied cleanly and keep their original author. Sixteen conflicted only outside product code and were resolved by hand:A hand-resolved commit is authored by the batch owner, because the commit guard refuses any other author identity, and names the original author in a
Co-authored-by:trailer (or a plain "Co-author on main" line where their address is internal).6a13aeba449eef012685072817ac5f2f012bb24a942ab8b155ad75c743820895e024f4176defccdfdcf58af19b8e2fa32b9c73b40a9f7849f3edcf1408357c75c361e08c905a4b5eee0c18ea9c7ae422e7e328420b499c8a0026d6fc73cb3cfadb6ff5c5963d468a589a3afce1391674b1fb390bb3b0dce8f3dfda0bd79f9aac36245b599b7f74ff14bcec631c2a0bce466bd2106d5bf057395fd466fa57bb46c5dfd2659557dc3603240f1873779f94f34a3f2222a91444d4868ebc316eBackwards 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. The 18 Python test files these picks touch gave 625 passed, 2 skipped and 10 failed. The 10 failures areTestDoctorcases intest_cli.pyandTestDoctorSttcases intest_cli_stt_memory_doctor.py; the same 10 fail on a cleanrelease/0.12.0on this host.tsc -p tsconfig.app.json --noEmitpasses; the 21 vitest files these picks touch gave 661 passed;gen-ui-index --checkandlint:i18npass.mypy src/kiro_crewandscripts/docs-lint.shpass.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 #30189, #30084, #30110, #30461, #30497, #30511, #30717, #30868, #31210, #31461, #31459, #31440, #31759, #32135, #32042, #32163, #30020, #32438, #32521.
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.