Repository navigation
feat(lts): ship the LTS edition from lts-v* tags on its own lts channel - #33804
Conversation
UX Review (Opus 5.5) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing blocks. The blind reader understood the dialog's main control, "I stopped it — Retry", rated it "sure", and would use it. The screenshots cover all three dialog states. Notes
[UX-REVIEWED] ce6d75f Description read[DESCRIPTION-READ] fa58bea62d3b639fa2f2c8324bbc760ffb45adbb286149b38ee2e990a619c908 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 |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Nothing blocks. The accepted RFC crew-lts-edition (D11) already covers the lts lane, and the removed guard said itself that it should be dropped once lts publishes. What this change shipsInventory (10 items) — 10 justifiedIntent: let LTS users install and auto-update a signed LTS build from its own
[FIRST-PRINCIPLES-REVIEWED] ce6d75f Description read[DESCRIPTION-READ] 350ee9b5ebbb94cd820bda5c80df97c81ab92543661e0acad9885eb448256a16 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: the Notes
[DESIGN-REVIEWED] ce6d75f Description read[DESCRIPTION-READ] fbb115f6e65d648b58f6ce61f4b243e7179d8d0325cf06579292f169f7223c9c 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: |
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: |
5111c9f to
bb9a638
Compare
bb9a638 to
b26f161
Compare
b26f161 to
5acaefb
Compare
5acaefb to
8ad61ec
Compare
daf1157 to
339a78a
Compare
|
UX Review on daf1157 (BLOCK): fixed on head 339a78a, and the screenshots in the body were re-captured from it.
|
339a78a to
1e3d44c
Compare
|
Opus 5.5 Review on 339a78a (BLOCK,
|
|
This follows up my earlier review of 8ad61ec. Here is where each item stands on head 1e3d44c:
Nothing from that review still stands, and this review found no blocking defect on this head. The wheel stem and the channel agree at every writer and reader (build-wheel, publish-cli, I cannot approve yet. PR Readiness failed on this head. 10 check runs failed, including every AI review lane, PR Policy and the readiness signal, and none of them has re-run. They look like the shared GitHub API rate limit, not a code problem, so they need a re-run. A maintainer also needs to review this PR by hand. It adds a tag-triggered workflow that signs and publishes under Optional:
|
Withdrawn: every item in this review is fixed or answered on the current head (see the follow-up comment). Other holds on this PR are listed there.
1e3d44c to
889543e
Compare
release-lts.yml runs on lts-vX.Y.Z tags only (release.yml stays v*-only) and calls the same build, sign and publish workflows with edition lts and channel lts. It writes only feed/lts/, cli/lts/ and desktop/lts/, under the LTS file stems (KiroCrew-LTS, kirocrew_lts). The desktop app is a side-by-side KiroCrew LTS app on the nightly pattern with the shared mac appId; the wheel is the kirocrew-lts distribution; no Docker image and no GitHub release. Every new workflow input defaults to oss, so v* releases are unchanged. The client now knows the lts lane (desktop, cli.sh, wheel engine, signed manifest), so the temporary 'lts has no builds yet' guard from the D11 change is removed.
889543e to
ce6d75f
Compare
|
Required changes for the LTS version publish. Merge first |
Problem / Motivation
Goal: An
lts-vX.Y.Ztag builds, signs and publishes the Kiro Crew LTS edition on its ownltschannel, under its own names, with no change to thev*release.Part of #32858. #32882 gave LTS builds the
ltsupdate channel (RFC crew-lts-edition D11), but nothing publishes to it. Today the only release path isrelease.ymlonv*tags, which builds the oss edition and writes the oss feeds. An LTS user has no published build to install or update to.Why it matters
Without release legs, LTS exists only as hand-run local builds. Users cannot install it from the CDN, the desktop app cannot auto-update, and
cli.sh --channel ltshas nothing to fetch. Shipping LTS through hand-built artifacts would also bypass the signing and notarization the oss release goes through.Not a goal
lts-v*tag is a separate, owner-run step (see the pre-release checklist below).release.yml(v*) builds or publishes.What changed (motivation → approach → change)
Approach: reuse the build, sign and publish workflows
release.ymlalready calls. One neweditioninput goes through the build workflows, and every LTS-only name is keyed onchannel == 'lts'. A separaterelease-lts.yml(owner decision) keepsv*untouched. A test pins thatrelease.ymlpasses no edition and never publisheslts.Release
.github/workflows/release-lts.yml: runs only onlts-v*and mapslts-vX.Y.ZtoX.Y.Z(other shapes refused). It runs the release-candidate tests, builds withedition: lts, and publishes 11 legs (CLI, six Linux, Windows, three macOS) onlts. It shares therelease-publishqueue withrelease.yml.build-wheel.yml,build-desktop.yml,build-windows.yml:editioninput (defaultoss) andKIROCREW_BUILD_EDITIONenv. The wheel is renamedkirocrew-lts, including the self-referencingvoiceextra.publish-cli.yml,publish-linux.yml,publish-windows.yml,sign-and-notarize.yml: onlts, the stems arekirocrew_lts,KiroCrew-LTSandKiroCrew-LTS-Setup, written only underfeed/lts/,cli/lts/anddesktop/lts/.packaging/build-desktop.sh: LTS is a side-by-side "KiroCrew LTS" app with its own Windows app ID and NSIS GUID. It refuses a nightly version. The Windows app ID is derived, not written out:website/electron/package.jsonbuild.appIdplus.lts, which today iscom.amazon.kiro.crew.lts.main.jsclaims the same ID at run time (production ID plus.lts). The NSIS GUIDfec201a5-e96c-5955-8d08-750be2d446a1is UUID v5 of that ID, andpackaging.test.jspins all three.scripts/check_feed_advance.py,scripts/ci/runner_watchdog.py,packaging/signing/cli-manifest.py,cli.sh: knowltsandkirocrew_lts.smoke-macos-install.shandsmoke-windows-install.ps1therefore assert the refusal (LTS boot refused, no/api/health). The macbounded()keeps the child's exit status inBOUNDED_STATUS.Client
/api/health(loopback only) returnsedition, andinstance-guard.jsgains anltsfamily. An LTS shell no longer adopts an oss gateway on the shared port, and the reverse prompts. When the other edition's gateway has no app window, the app never adopts or kills it. A new native dialog asks the user to stop that gateway: "Run this command in Terminal:" thenkirocrew stop --port <port>(family-takeover.js). On macOS a gateway whose app has quit (an orphan, main'sorphanedHolderPid) is offered main's orphan stop instead, and only a launchd job gets the note that a service may restart it.feed-lane.jspasses the required channel intoresolveChannel, so an LTS app does not auto-download an olderltsrelease.src/kiro_crew/release_channel.pydistribution_name()iskirocrew-ltsonly when that distribution is installed, elsekirocrew(the desktop backend installs the unrenamed tree).src/kiro_crew/skills_discovery.pyreads it.Docs:
docs/build/release.mdgains "LTS edition releases" with the checklist below.Pre-release checklist and known gaps
Owner-only repo settings, needed before the first
lts-v*tag. The order matters:release-tags-admin-only(id 19310912): addrefs/tags/lts-v*. Today it covers onlyrefs/tags/v*.proddeployment policy: add taglts-v*. Today it allows only branchmainand tagv*.The other order leaves a window where
prodadmitslts-v*while anyone with push access can still create that tag.Known gaps (accepted, low risk):
feed/lts/, an LTS update check reports a failed fetch.cli.sh --channel ltsover an oss pipx install leaves the oss venv, and thekirocrewlink moves to LTS.packaging/MIN_VERSION.release-publishgroup can be cancelled by the other edition's tag.check_feed_advance.pyhas nolts-v*tag evidence, so an older LTS publish job re-run within the feed's 300 s cache window could roll the feed back. Re-running the newer job advances it again. Nightly accepts the same exposure.make wheel EDITION=ltsinstall upgraded with plain pip would gainkirocrew-ltsbesidekirocrew. No published artifact reaches this state.Backwards compatibility
Compatible: every new workflow input defaults to the oss value, and
release.ymlis unchanged (pinned bytest_release_lts_workflow.py)./api/healthonly gains a field. A gateway withouteditionreads as oss.Reader: src/kiro_crew/dashboard/handlers/updates_check.py:_check_release_feed -- app -- the lts feed is published by this PR; covered by test/test_dashboard_updates_coverage.py
Reader: src/kiro_crew/cli_server_update.py:server update path -- chat -- lts now resolves a real wheel URL; test/test_update_wheel_dispatch.py
Reader: src/kiro_crew/platform/update_capability.py:remediation -- app -- returns the normal wheel command for lts; test/test_update_capability.py
Reader: src/kiro_crew/platform/update_layout.py:non_managed_pip_upgrade_command -- app -- returns the lts pinned command; test/test_update_capability.py
Reader: src/kiro_crew/dashboard/handlers/updates.py:channel_is_published re-export -- app -- no remaining importer on main or HEAD (git grep)
Reader: website/src/pages/settings/AboutPanel.tsx:unpublishedNotice -- crew page -- lts shows the normal update row; AboutPanel.channelSwitcher.test.tsx
Reader: website/src/pages/settings/AboutPanelGatewayUpdate.tsx:unpublishedNotice prop -- crew page -- prop removed with its only caller; AboutPanel.gatewayChannelRestart.test.tsx
Reader: website/src/i18n/locales/*.json:automatic_updates_unavailable_channel_unpublished -- crew page -- key removed with its only use; i18n gate clean
Tests
test/test_release_lts_workflow.py(new):lts-v*-only trigger, shared queue, tag→version map and refused shapes, every leg onltswithedition: lts, no Docker or GitHub release, tests run before publishing,release.ymluntouched, lts-only names gated on the channel, the wheel rename step run against the realpyproject.toml/setup.cfg(no extra left namingkirocrew), smokes pinned to the realLTS boot refusedlog text,distribution_name()fallback.test/test_api_health.py:editionisossby default andltson an LTS build.website/electron/test/instance-guard.test.js:ltsfamily; LTS shell over an oss gateway prompts, oss over LTS prompts, LTS over LTS reuses, a missing edition reads as oss.website/electron/test/auto-update-policy-channel.test.js: an LTS build at 0.14.0 is not auto-downloaded to 0.13.0. It fails with thefeed-lane.jsfix reverted.test_publish_feed_contract.py,test_windows_signing_contract.py,test_ci_runner_watchdog_workflows.py,packaging.test.js,store-rename-migration.test.js(exactKNOWN_CHANNELSpin now includeslts), and the guard tests removed with the guard.Manual verification
kas statuswas run locally with the edition stamped: exit 1 andLTS boot refused, the text the smokes assert.bounded()was run in a bash harness: exit 1, exit 0 and a timeout (124) all report correctly.smoke-windows-install.ps1has no local PowerShell parse check (nopwshhere). A reallts-v*release has not run and must not run until the checklist above is done.Screenshots / video
The new native dialog in
website/electron/runtime/gateway/family-takeover.js. An oss app shows it when it finds an LTS gateway with no app window on the shared port (and the other way round). Captured on this head: the real module and Electron's real native dialog (Linux GTK, Electron 44.9.1, Xvfb with openbox) inside a container. Only the port probes are stubbed so the port never frees, and the clicks are scripted (Retry, Retry, Retry, OK). The capture script is in a PR comment.Background service, round 1 prompt:
Retry round (the port is still held):
Cancelled-launch error after the last retry:
The other frontend change removes the LTS-only "no published builds yet" notice from Settings > About > Updates. An LTS build cannot open that view here without an organisation sign-in, so there is no shot of it; an oss build renders it unchanged.
Related Issues
Part of #32858.
Closes #32859: an LTS desktop app no longer attaches to an oss gateway on the shared port.
Checklist
Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license.