Repository navigation
fix: backport 0.12.1 wide batch 5 (15 user-facing fixes) - #32948
Conversation
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 of the Remote Crew launch form. The PR changes visible controls and copy there and attaches no image, so the blind read was not performed. Evidence gaps
Notes
[UX-REVIEWED] 4c4f217 Description read[DESCRIPTION-READ] b4176c0a218264c4dcd4fc3ee6c8a14d406638aa3c8e09f0d85e0ad8c189be04 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) — ✅ PASS (a BLOCK with no Trigger and no Outcome does not block)Design-level review of Design-Verdict: PASS cannot evaluate: missing a screenshot of the Remote Crew launch form as it renders on BlockersThe Remote Crew launch controls change with no rendered evidence of this revision Notes
[DESIGN-REVIEWED] 4c4f217 Description read[DESCRIPTION-READ] de7855604cefe21bbfec8a445ccbc9b98ab1943a5c8d188003f4dd89161cf4f7 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 13 fixes that are already merged on Notes
What this change shipsInventory (10 items) — 10 justifiedIntent: give 0.12.x users fixes already merged on
[FIRST-PRINCIPLES-REVIEWED] 4c4f217 Description read[DESCRIPTION-READ] f1d477cfe7b3ec6e94aa6491054d8fe0f880e8a124c95c50f6c51174544ab83d That sha256 names the PR description this verdict read. Recompute it with |
Security Scope Review -- ✅ PASSWhich legitimate operations does Verdict basis: adjudicated, zero confirmed regressions. Security scope review -- adjudicated candidates
No candidate flipped from allowed to refused. Scope-Verdict: PASS Reason: The only shell-gate tightening here is in the ssh-to-self check. Once the own-address cache is published, an IP literal in host position is now also checked by asking the kernel whether that address belongs to this machine. That check returns true only when the kernel chooses the destination itself as the source address, so remote addresses should keep passing. No regression is confirmed, and I am not claiming per-host verdicts. One part can't be classified: the new dashboard-side frame relay guard. It is named under UNADJUDICATED below. TIGHTENING: ADJUDICATED: 19 candidates, 0 confirmed newly-refused PLATFORM COVERAGE: posix (14 rows) and windows (5 rows); no
UNADJUDICATED: The dashboard-side app-frame relay (
The ACP [SCOPE-REVIEWED] 4c4f217 |
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: |
…30522) comments.json is decoded verbatim, so a hand edit can leave an anchor_quote that is a number, a list or an object. The rescan that runs on every content write tested it with "quote not in content", which raised TypeError on that row. The caller swallows the error, so the write still landed, but the rescan stopped there: no other thread was ever flagged anchor_orphaned when its quoted text was removed. The comment list route raised on the same row too, while redacting the anchor text, so the Library showed no threads at all. A non-string quote is now skipped like an empty one, and the rest of the threads are rescanned. The list route sends that comment with no anchor, and a non-string prefix or suffix as null. The stored value is left as it is. (cherry picked from commit 82d0033)
…#30584) A registry-sourced app update re-registered the app and restarted its backend but never promoted the rewritten app-crons.json into the gateway's running CronService. Only the local-source update branch did, so a cron field a new registry version changed kept its old value until the next gateway start. Former built-ins now update from their own registry repositories, so this path is the one they take. Call the existing _resync_app_crons_after_update after re-registering in the registry branch of POST /api/apps/{name}/update, and in POST /api/apps/registry/install and /install-stream when they install over an app that is already there (the detail page's Update button). A fresh install still promotes nothing: it leaves the app disabled and the enable registers its crons. (cherry picked from commit e3c7b2b)
A Tool Search deferred call on KAS reaches Crew as the envelope
{tool_id, arguments}, with the tool's own arguments (and the agent's
purpose line) nested inside. extract_tool_purpose read only top-level
keys, so the tool pill and the approval card lost the purpose for
every such call.
Add one strict envelope reader, session_directive.deferred_call_arguments,
and read the purpose through it. The session-directive digest and the
workspace-call gate each carried their own copy of the same strict
rule; both now call the shared reader and keep their own tool_id
identity check, so what they accept and refuse is unchanged.
Closes #16721
(cherry picked from commit f0ca734)
…an authenticating reverse proxy (#29975) A sandboxed app frame has an opaque origin, so its own requests carry no cookie. A reverse proxy in front of the dashboard that signs every request in by cookie answers them with its login page: the frame document never loads where the iframe navigation is cookieless, and elsewhere the module import and every ctx.fetch get an HTML 403. The card then fails with "Couldn't load this preview." The dashboard now does the frame's network for it: - A first mint also returns the frame document (srcdoc variant, after the mint's own checks). The dashboard sets it as the iframe's srcdoc. Sandbox stays allow-scripts, credentialless stays. The policy is a <meta> copy of frame_csp ('self' spelled as the dashboard's origin; sandbox and frame-ancestors cannot be in a <meta>), so nothing is widened. - ctx.fetch, and the module source when its import is refused, go through appFrame/frameRelay.ts: only from the frame's own window with its per-frame nonce, only to the frame's own app routes (/app-frame-fetch/<app>/, /apps/<app>/ui/ GET), only with headers the frame could have sent, and the answer exposes only what a CORS read would have. The gateway's capability and grant checks run on the relayed request unchanged. - The bootstrap imports the module by URL first, so the direct path is unchanged. When that is refused, the source runs as a nonce'd inline module that hands over its entry with globalThis.kirocrew.register(activate). A module that only exports activate keeps the URL path. Point hosts skip relay messages before their own rate and size bounds. (cherry picked from commit e9f4fad)
Disable, trust withdrawal and uninstall probed the backend's port once, right after stop_app_backend returned. That stop signals the whole process group but waits only on the root it spawned, and a sandboxed backend's root is the sandbox wrapper, which exits before the server inside it closes its listening socket. The single probe caught the socket still bound and reported "backend still running ... not ours to stop" for a backend that was already exiting; on trust withdrawal that turned into a 409 refusal. stop_backend_and_observe now re-probes a still-listening port for a bounded grace (2 s, 50 ms polls) and returns a port only when it is still listening at the end. A listener that outlives the grace is reported exactly where it was, so withdrawn trust still refuses. Part of #26967 (cherry picked from commit 7443c94)
…30921) After /clear or Reset conversation the replay was suppressed, but the fresh session's first prompt still carried a "Recent Session Context" block quoting the same session's turns from before the reset, so the agent could recall them. The dashboard turn now tells the builder it is the first turn after a reset, and that block is left out. Fixes #29721 (cherry picked from commit 5e431e9)
…#30203) STT is enabled by default, but a CLI or source install on Linux or macOS does not pull in the optional voice extra or a system ffmpeg. Doctor marked both gaps with an error and exited 1, so every fresh CLI install failed `kirocrew doctor` and broke `doctor && gateway`. Report them as a warning with the install hint instead. Only a macOS/Linux desktop release whose payload ships both parts still treats a missing one as an issue, since that is a damaged install. Windows keeps its non-fatal report in every case. (cherry picked from commit 94579a2)
…ling (#30870) The 5-minute RSS tick built its candidate set only from the session map, so a runtime no session owns -- the warm-pool spare pre-spawned at boot -- was never measured. A child it spawned could grow past the ceiling and run unrecycled. Add a second pass that snapshots the warm providers, measures each runtime tree with the same per-tick reader the session pass uses, and tears down any over the ceiling. A spare over the ceiling is first taken out of the pool by a synchronous removal that carries its own membership check, so a spare a session claimed across the measurement await is not found and is left to the session pass; the removed runtime is then torn down through the pool's own discard path, its sweep shield dropped per pid as it goes, and one refill is scheduled through the same helper the health sweep uses so the pool is not left a spare short. The ceiling value and the action on a hit are unchanged, and a spare is claimed by nobody, so no sub-agent, injection or background-launch guard applies. The discard already offloads its kill off the loop. Closes #30406 Co-authored-by: kirocrew-agent <agent@kirocrew.local> (cherry picked from commit 302b4fb)
An address this machine gained since the last periodic refresh is recognized as its own in the same check, instead of only after the next refresh. (cherry picked from commit a67c9dd)
d1a86e7 to
4c4f217
Compare
|
Dropped #30664 and #30861 from this batch. GPT 6.1's two blocking findings on d1a86e7 hold against the main code too: #30664's legacy-route lookup in website/src/apps/BuiltinAppRoute.tsx:147 keeps only |
Problem / Motivation
Goal: Bring fifteen 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.dm_scope: unified, a second person's Telegram DM is folded into the first person's turn and never answered./clearthe agent can still recall the cleared conversation.kirocrew doctorfails when the optional voice extra or ffmpeg is missing.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@989d318f81, in merge order onmain, one commit each. Every commit keeps its original author.#30664 and #30861 are not carried here. GPT 6.1 found a defect in each that is also on
main. #30664's legacy-route lookup (apps/BuiltinAppRoute.tsx:147) drops the query error, against the blockingerrors-use-error-noticerule. #30861 shows a browser card in Bot chat, which no decision record allows over the messages-only Bot chat decision. Both get settled onmainfirst. Their rows under "Why it matters" stay as written when the PR opened.Hand-resolved hunks, none of them in product code:
docs/system-specs/modules/cli.md, the doctor Speech-to-Text paragraph. Release gained fix(doctor): stop a sandboxed shell reporting the decoder as missing #31461's macOS sentences after this branch was cut. They were kept, and the closing "Windows preserves..." sentence was replaced by fix(doctor): stop failing on a missing optional voice extra or ffmpeg #30203's text, the only sentence fix(doctor): stop failing on a missing optional voice extra or ffmpeg #30203 changes.website/src/test/RemoteCrewPanelLaunchForm.test.tsx, which fix(remote-crew): show an empty region as missing and stop a repeat launch #30779 adds. Release now has fix(a11y): announce which remote crew settings tab is open #32042, which made "Set up a new one" a radio tab. fix(a11y): announce which remote crew settings tab is open #32042's one-line update to this file (buttontoradio, main1c2a0bce46) was folded in.docs/system-specs/modules/security.md. Its one inserted sentence about_kernel_says_own_addresswas added to the release paragraph.5e8741f525583f9ef4be82d0033ea1fae17d8528e3c7b2b212765114783ff0ca73453defa82e692ae9f4fad62ea3498914171548e9f0cc4204a91bfe2b6ad8c0e71ca8c419f87443c94b6a94c0ee03945e431e9ed0f16d29cd7f94579a2e361725129a25302b4fb92645ed7369e292cb8bdce51917084db1a67c9ddf004c4f2170b3#30779's co-author trailer carried an internal host address and is rewritten as a plain "Co-author on main" line, because the commit guard refuses it.
#30771 (refuse
ddto a raw disk device) is the next clean row but is a command-deny tightening, so it waits for its own batch.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. #29975 changes how app frames load: they go through the dashboard's own origin instead of the sandboxed frame's, which is what lets the cookie reach an authenticating proxy.
Tests
Proof: not provable -- other: a backport of main commits whose proofs ran on their own PRs
release/*, so the related set was run here. Every touched test file plus the test files that import a touched source module, 156 files, gave 7,262 passed.test_doctor_stt_transcribe_providerfail on a cleanrelease/0.12.0@989d318f81.94579a2e36, because this host's doctor run also reports an outdated service definition and low tmpfs space.tsc -p tsconfig.app.json --noEmitpasses; vitest on the touched test files plusRemoteCrewPanel.tabState, 67 passed;gen-ui-index --checkandcheck-i18n-stringspass.mypy src/kiro_crewandcheck_harness_parity.pypass.check_file_line_capfails only ondashboard/chat_api/resume.py, which is already over the cap on release and is not touched here.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 (#29975, #30555, #30556, #30779) 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 #30306, #30522, #30584, #30540, #29975, #30555, #30556, #30595, #30921, #30203, #30870, #30779 and #30815. Source issues: #29771, #30521, #16721, #29974, #30554, #30466, #29721, #30406, #15102.
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.