Repository navigation
Conversation
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Check that the prerequisite gate always fills its cached query before the modal opens; [DESIGN-REVIEWED] 8ae6ca5 Description read[DESCRIPTION-READ] 322597afa8505b9edc297988db69b5517ca53698fc14d5017b455ffcc266b0d7 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 Check by hand that a real What this change shipsInventory (8 items) — 8 justifiedIntent: a user with both a personal and an enterprise Kiro subscription can find out, from the Kiro Account modal, how to move Kiro Crew to the other account (#4184, #13868). This is an ADDITION: the capability already exists and nobody could find it.
[FIRST-PRINCIPLES-REVIEWED] 8ae6ca5 Description read[DESCRIPTION-READ] 0013cd8f91bdacfee8443efab905f142cfb86252934db03533c05c5b845f4aaa That sha256 names the PR description this verdict read. Recompute it with |
GPT 6.1 Review — ✅ no blocking findings (all downgraded on adjudication)GPT 6.1 flagged blocking issues on This comment is updated in place on each push. Review detailsBLOCKING -- website/src/components/KiroAccountModal.tsx:700 -- Cached settings suppress config-refetch failures Adjudication (Opus 5.5) — is blocking on each finding proportionate?F1 (LOW): The failure needs two things at once: a successful earlier read cached under 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: |
UX Review (Opus 5.5) — 🔴 BLOCKUX-level review of UX-Verdict: BLOCK cannot evaluate: missing the config-unreadable notice, and the screenshots predate HEAD. They show wrapped commands and the reassurance line last; the code puts it before the commands and uses Watch
Evidence gaps
[UX-REVIEWED] 8ae6ca5 Description read[DESCRIPTION-READ] cbc2fa738141dd503613f1f99556a8defd5f53a0a26c05138cc01c8ed4a8bbc2 That sha256 names everything this verdict read: the PR description, plus the 8 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 |
be441b1 to
d909e99
Compare
|
Dispositions for the review round on
|
d909e99 to
b10261d
Compare
|
Dispositions for the review round on
|
b10261d to
b10ccb7
Compare
|
UX round on
|
b10ccb7 to
98db74a
Compare
|
Rebased onto main (
|
98db74a to
2b2af43
Compare
|
|
|
|
|
|
|
|
|
|
bolichen97
left a comment
There was a problem hiding this comment.
Needs a maintainer design call before merge. The code itself looks correct and small.
Head reviewed: 2b2af43. Bots read: Design PASS, First Principles PASS, UX CONCERNS, Opus no blocking findings, GPT changes requested (blocking). Readiness was still running at review time.
What I checked:
_command_wrapperkeepslogin_commands_foroutput the same.logout_command_foruses the same prefix and suffix, so bundled, off-PATH and PATH installs get a matching sign-out command.logout_commandis set at all three probe sites that setlogin_command. The redacted non-owner payload does not serve it.- The modal block is positive-gated on the kiro-cli harness, the owner flag, a reading, and all three served commands. It never falls back to a bare name. Each guard has a test that reads as a real negative case.
- The branch merges cleanly with current main. The
merge conflictlabel looks stale for this head.
Why this needs a maintainer:
- The PR refers to an issue that has the T3 tier. The PR declines the multi-account store that the issue asks for and ships only an in-product hint. That scope choice is a design decision, so it needs a design sign-off from a maintainer. I do not approve it as a reviewer.
- The GPT lane blocks this head on
errors-use-error-notice(see the inline comment). Either the small fix below or a maintainer decision on that lane is needed before readiness can pass.
Overlap: an open PR, #17784, also changes KiroAccountModal.tsx and its test file. The two changes look independent. The PR that merges second will need a rebase.
🔴 0 must-fix. 🟡 1 should-fix.
| // on kiro-cli: the identity sweep recycles kiro-backed children only. Another | ||
| // harness can still hold a reading, so the harness is checked POSITIVELY, the | ||
| // same way the credit pill does (`kiroUsageReadout.tsx`). | ||
| const { data: kirocrewCfg } = useQuery<AcpBackendConfig>({ |
There was a problem hiding this comment.
🟡 A failed backend config read hides the switch steps with no notice, and the GPT lane blocks on it
If the backend config read (api.kirocrewConfig()) fails, kirocrewCfg stays undefined. Then isKiroBackend(undefined) is false and the whole block returns null. The user gets no switch steps and no sign that a read failed. Every other read in this modal goes through ErrorNotice (for example the forecast read in CreditUsage).
const { data: kirocrewCfg } = useQuery<AcpBackendConfig>({
queryKey: ['kirocrewConfig'],
queryFn: () => api.kirocrewConfig(),
})
if (!isKiroBackend(kirocrewCfg)) return nullWhy it matters: the actual harm is small because the block is an optional hint. But the GPT lane maps this to the blocking errors-use-error-notice rule on this head, so readiness cannot pass until the code changes or a maintainer decides on that lane.
Suggestion: read error from the query. Before the harness check, render <ErrorNotice message={...} askAgent onHandoff={...} /> when the read failed, as the forecast read does. Add one test with a rejected kirocrewConfig query. If a maintainer decides that an optional hint may stay silent, I will withdraw this.
Rule: website/AUTOSDE.yaml errors-use-error-notice
2b2af43 to
6f407ec
Compare
|
|
|
|
|
|
… modal The Kiro Account modal now shows a Switch Kiro account block: the gateway-served kiro-cli logout command and both sign-in commands (personal and organization SSO), as click-to-copy code. The gateway serves logout_command composed with the same wrapper as the sign-in commands, so a bundled or off-PATH kiro-cli gets its absolute path. The block shows only beside a reading, only to the owner, and only on the kiro-cli harness. Refs #4184
|
|
6f407ec to
8ae6ca5
Compare
|
|
bolichen97
left a comment
There was a problem hiding this comment.
Follow-up on the new head 8ae6ca5. The should-fix from my last review is fixed. This still needs a maintainer design call before merge.
What changed since 2b2af43 (I read the incremental diff of KiroAccountModal.tsx and its test; the rest of the range is the rebase onto main):
- A failed harness config read now renders
ErrorNoticewith the agent hand-off instead of silently dropping the block. It uses the sharedfailedWithNoDatahelper, and it sits after the owner check, so a non-owner still sees nothing. - The new key
switch_account_config_unreadableis present in the locale catalogs I spot-checked (en, en-XA, zh-CN, ja). - A new test rejects the config read and checks the notice, the absent steps, the still-rendered account reading, and the hand-off closing the modal.
- The two absence tests now wait for the config query before asserting, so they fail for the right reason.
- The reassurance line now sits above the commands.
- The GPT lane is green on this head.
What still needs a maintainer:
- The PR refers to an issue with the T3 tier and declines the multi-account store that the issue asks for, shipping only an in-product hint. That scope choice needs a design sign-off from a maintainer. I do not approve it as a reviewer.
- The UX lane could not evaluate this head because the body screenshots predate the one-line commands and the moved line. Fresh shots of the modal, or a maintainer decision on that lane, are needed.
- The author asks a maintainer to run one real sign-out and sign-in switch, or to accept the tested session sweep as evidence.
- Some backend test shards were still running when I read the checks.
Overlap: #17784 also changes KiroAccountModal.tsx and its test file. The changes look independent; whichever merges second will need a rebase.
🔴 0 must-fix. 🟡 0 should-fix.
|
🤖 Kiro Crew Auto-Pipeline [operator: buluoray#ce473fca]: closing this PR. The operator is pausing work on T3/T4 issues for now, so this stays unmerged. The branch is kept: reopen the PR or reuse the branch if the issue is picked up again. |
Problem / Motivation
Goal: A user with more than one Kiro subscription can see, from the Kiro Account modal, how to switch Kiro Crew to another account.
Users with both a personal and an enterprise Kiro subscription have no in-product hint about how to move Kiro Crew from one to the other (#4184, re-sighted as #13868). The reporter's own workaround is
kiro-cli logoutthenkiro-cli login. They pointed at the signed-in-account area of the dashboard as the place to look. That area is the Kiro Account modal, and today it shows the account but says nothing about changing it.Why it matters
Without the hint, a user has to guess whether switching accounts means restarting the gateway, editing files, or reinstalling. On main,
kiro-cli logout+kiro-cli loginalready works end to end:KiroPrerequisiteServicefingerprints the identity store, and_retire_sessions_on_identity_changeinchat_runner.pyrecycles children started under the old account at their next turn. The capability exists but nobody can find it.Not a goal
kiro_prerequisite.pysays Kiro Crew never authenticates on the user's behalf for the kiro-cli backend, and this PR does not reopen that.What changed (motivation → approach → change)
Gap: the account can already be switched, but the dashboard never says how. Approach: show the two kiro-cli commands in the one place the user looks for their account, with no new credential surface. Each command is taken verbatim from the gateway, so it is correct on every install shape. On the desktop app's bundled kiro-cli the bare name is not on the user's PATH, so the gateway serves its quoted absolute path, the same way the setup gate already does for
login_command. Change:src/kiro_crew/kiro_prerequisite_status.py: newKIRO_CLI_LOGOUT_COMMANDandlogout_command_for(binary, environ). They are composed from the same(prefix, suffix)wrapperlogin_commands_foruses, now factored into_command_wrapperso the two cannot drift.PrerequisiteStatuscarrieslogout_command, set inkiro_prerequisite_probe.pyat all three sites that setlogin_command;src/kiro_crew/kiro_prerequisite.pylists the new names in its facade.website/src/components/KiroAccountModal.tsx: below Manage account, a Switch Kiro account block with a visible heading. It has one sentence ("In a terminal, sign out of Kiro CLI, then sign in with your other account."), thenlogout_command, then both sign-in commands under the setup gate's own labels: personal (login_command) and organization SSO (sso_login_command), because a personal-to-enterprise switch needs the second one. All three are click-to-copyCopyCommandblocks, each kept on one line (whitespace-nowrap) so a long bundled path never wraps inside a flag. Right after it comes "Your chats stay. Your next message in each chat uses the new account.", before the commands, so "logout" does not read as risky. If the harness setting cannot be read, the block is replaced by anErrorNoticewith the agent hand-off (new keyswitch_account_config_unreadable). It shows only beside a reading, only to the owner (setup_allowed), only on the kiro-cli harness (isKiroBackend, the check the credit pill uses), and only when the status serves all three commands; it never falls back to a bare name. The status is read from the prerequisite gate's cached query throughusePrerequisiteStatus()inhooks/useGatewayPlatform.ts, now exported, so no kiro-cli spawn is added.website/src/api/client/onboarding.tstypeslogout_commandas optional.switch_account_label,switch_account_steps,switch_account_effectandswitch_account_config_unreadableinen.manual.jsonand all 11 translated catalogs, withen-XA.jsonregenerated. The commands are not in the catalog.docs/system-specs/modules/learn-cron-dashboard.md(logout_command).Backwards compatibility
Compatible: the status payload gains one field (
logout_command), andlogout_commandis optional on the client, so an older gateway simply shows no switch steps.Removes nothing: the diff adds a status field, a helper and a rendered block, and removes or tightens nothing.
login_commands_forreturns the same values it did; its wrapping moved into_command_wrapperunchanged, and its existing tests pass as they are.Tests
test/test_kiro_prerequisite.py,test/test_kiro_prerequisite_handlers.py,test/test_kiro_prerequisite_probe.py:test_logout_command_shares_the_login_prefix_on_a_bundled_install: on a bundled copy, logout is the quoted absolute path, with the same prefix as login.test_windows_bundled_logout_command_names_powershell: on Windows it gets the same PowerShell wrapper as login.test_off_path_override_logout_serves_its_absolute_path,test_system_resolution_keeps_the_bare_logout_command: these cover an off-PATH override and an ordinary PATH install.test_update_cli_refuses_the_bundled_copy_without_spawningnow also asserts the servedlogout_commandon a real bundled probe.test_non_owner_dashboard_user_gets_only_redacted_readinessasserts the redacted branch serves nologout_command.website/src/test/KiroAccountModal.test.tsx:tells a signed-in user how to switch account with the gateway-served commands: with a bundled-path status, the block has a visible heading, its threecodeelements are exactly the served logout, personal login and SSO login commands in that order, and both tier labels are shown.leaves the switch steps out when the gateway serves no logout command.omits the switch-account steps for a non-owner, who gets only the bare stand-in commands(setup_allowed: false).omits the switch-account steps on the claude / kas harness, where chats do not run on the kiro-cli account.says why the switch steps are missing when the harness setting cannot be read: the notice shows, the reading stays, and the hand-off closes the modal.omits the switch-account steps when there is no signed-in reading(api-key,signin-required,none,null).Results:
pytest -n0on the three prerequisite test files: 158 passed.vitest run src/test/KiroAccountModal.test.tsx: 49 passed.platform_compat.IS_WINDOWSto False, so its POSIX assertion holds on the Windows runner. Mutations: removing the wrapper fromlogout_command_forreddens the 4 non-bare logout tests. A bare-name fallback in the modal reddens the no-logout test. Dropping the reading guard reddens the 4 no-reading cases. Dropping the owner guard reddens the non-owner case. Dropping the harness guard reddens both non-Kiro cases. Dropping the config-failure notice reddens its case.tsc --noEmit, eslint on the touched files, black/isort/flake8/mypy on the touched Python files, andI18N_BASE_REF=origin/main node scripts/i18n-check.mjs(19 checks PASS).Manual verification
Built the SPA and opened the modal through the shared
serveDist+stubDashboardApiharness. The stubbed usage reading and/api/kiro-prerequisitecovered a PATH install (light, dark,zh-CN) and a bundled macOS path (desktop and 390 px). The capture asserted that the dialog'scodeelements equal the served commands before each shot. The script was scratch-only and is not committed. Not run on this host: a realkiro-cli logout/loginround trip, and the Windows PowerShell form, which is covered only by the unit test. The "moves them at the next message" sentence rests on the existing identity sweep inchat_runner.pyand its tests, not on a fresh run.Screenshots / video
PATH install, light:
Desktop app's bundled kiro-cli (absolute path served by the gateway):
Dark, zh-CN, and 390 px phone width
Related Issues
Refs #4184
Decision for the owner
#4184 asks whether Kiro Crew should store several Kiro credentials as named accounts, with an in-dashboard switcher and per-session pinning. This draft declines that store, because Kiro Crew deliberately never signs in to kiro-cli for the user and the logout/login identity sweep already switches every chat, and it ships only the missing in-product instruction (rejected alternative: a Crew-owned multi-credential vault, which would widen the credential trust boundary for a workflow that already works).
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.