Repository navigation
Conversation
|
bot: Live proof and the design-gate comparison for this pull request, run on the emulator at this head. Design-gate contact sheet for the five picker targets (base vs head cells side by side; the header row gains the leading Exercised on the real New session composer, signed in with a fake seeded account:
The remaining unreadable pixel text on the sheet headers is the rule reporting on the diagnostic fixture, not on the product fix; the live dump already proves the placement. |
Code Review SummaryStatus: No Issues Found | Recommendation: Merge Executive SummaryThe incremental change only tidies the two mounted tests flagged previously: the mode-picker Cancel test now unmounts and asserts the Back listener is released, and the folder-picker expired-arm test restores the Assumptions: none. Files Reviewed (2 files)
Previous Review Summary (commit e14566f)Current summary above is authoritative. Previous snapshots are kept for context only. Previous review (commit e14566f)Status: 2 Issues Found | Recommendation: Address before merge Executive SummaryThe production change is a small, consistent one: every composer picker shell now wires Assumptions: Overview
Issue Details (click to expand)SUGGESTION
Files Reviewed (12 files)
Reviewed by deepseek-flash-latest · Input: 39.2K · Output: 2.9K · Cached: 170.4K Review guidance: REVIEW.md from base branch |
The mode-picker cancel test now unmounts like its neighbours, and the folder-picker cancel test restores the flag it flips. Behavior of the fix is unchanged.
|
bot: Final-head proof, at |










Evidence
Evidence: demonstrated — on the Android emulator at head
e14566f37the four composer picker sheets that the New session flow opens renderCancelat the leading edge andDoneat the trailing edge of the sheet header, and tappingCancelcloses the sheet while the row keeps the value it had; the repository picker rendered onlyDoneon the base revision of the prepared pair.What changed
Cancelnow calls the same dismissDonecalls on every composer picker, soCancelcloses the sheet and leaves the current selection untouched. A picker commits a choice on the row tap and reports it through the picker bridge; the header controls only pop the route. Six files, 37 mounted tests, all passing.repo-picker.tsx: the live list arm and the dead-bridge "Options expired" shell.branch-picker.tsx: the dead-bridge shell; the live list arm already hadCancel.mode-picker.tsx,instance-picker.tsx: every arm, through one shared close function.profile-picker.tsx: the expired shell.folder-picker.tsx: the expired shell. Its live arm keeps the leadingBackframe on purpose: there the leading control drills out of a nested folder, it does not abandon the pick.One mounted test per screen asserts that
Cancelpops the route once and reports no pick; the mode and profile tests also assert that the picker slot is not cleared.Live walkthrough at this head
Signed in on the Android emulator with the fake seeded account and drove the New session composer; every control placement below is a
uiautomatornode with real bounds.CancelCancel[37,174][188,300]Done[918,174][1044,300]GitHub · picker-lab/private-reposelectedCancel[37,1237][188,1363]Done[918,1237][1044,1363]Mode: CodeunchangedCancel[37,1237][188,1363]Done[918,1237][1044,1363]Run on: Cloud AgentunchangedCancel[37,1237][188,1363]Done[918,1237][1044,1363]Pick a profile: Default environmentunchangedSelecting
GitHub picker-lab/private-repo, reopening the sheet, then tappingCancelkeeps the row's value exactly (Repository: GitHub · picker-lab/private-repobefore and after).Design gate
Prepared pair, five screen-file diagnostic targets (
repo-picker,branch-picker,mode-picker,instance-picker,profile-picker), basecb91de0faand heade14566f37, fourstandardcells per revision, 5/5 targets captured on both revisions.comparereportsstatus: comparable; it lists 20 finding groups as introduced, all one rule on the diagnostic sheet header, and all explained by the fixture rather than the change:safe-areaon the diagnostic sheet header (View[0]measured against the top inset, and the title's paragraph): a diagnostic route renders the sheet bare under(auth), not inside a presented form sheet, so the header row reads as reaching into the top system inset. In the real presented sheet, the walkthrough's own dumps show the same header starting lower on screen, inside the sheet.Cancelcontrol joins that row's group and raises its count. The control that existed before (Done) keeps a byte-identicalrecton both revisions, so nothing moved in the row.Design gate: 4 cells, 40 errors, 0 warnings
One screen in this flow cannot be a diagnostic target at all:
folder-picker.tsxcalls a hook that only the signed-in shell provides, and a diagnostic route renders signed out. That screen's expired-armCancelis covered by its mounted test.Branch sheet coverage
The live branch list could not be opened against the seeded fixture: the fixture's placeholder installation mints no credentials, so the branches query fails (
listRepositoryBranchesreturns an internal error), the row shows the failure and a retry that does not recover it, and the sheet's trigger never renders on screen. The branch picker's live arm already showedCancelbefore this change; its expired arm — the arm this change adds the control to — is proven by the gate pair'sbranch-pickertarget and the mounted test.The folder picker trigger does not exist in this state either: it is only rendered for a connected CLI instance target.
Production prerequisites
None.
A prerequisite pull request for the design-gate tooling made the prepared pair in this pull request possible; it landed before this branch's gate run (#11215,
cb91de0fa).Proxy answers
The owner's proxy gave these answers in the owner's place.
Answer: All four composer pickers
Confidence: 0.60 · Precedent: Fix the root cause and the class of the issue on every affected surface: owner wrote "fix this for all shared services model pickers on all surfaces" (session 75f8833fe8c1)
Answer: Add it to branch-picker too
Confidence: 0.75 · Precedent: Fix the root cause and the class of the issue on every affected surface: session 00eb2a1d3e4b, "if anything repeats the same mistake over and over again, fix the class"
Answer: Add Cancel to its expired arm
Confidence: 0.50 · Precedent: Fix the class on every affected surface: session 75f8833fe8c1, "fix this for all shared services model pickers on all surfaces"; balanced against rejecting changes to deliberate behavior (PR fix(core): apply global permission rules after repository rules kilocode#15356)
Answer: Separate prerequisite PR
Confidence: 0.55 · Precedent: Git and pull requests: 'Put each separate concern in its own PR, including a fix for a CI blocker that many PRs share' (session a3cb05d9ad53, owner chose 'Fix it separately'). The route-file gap blocks every route-only mobile change, not just this one. 'Keep each PR diff to files that the feature needs' (session 12a6ef34f658, owner dropped out-of-scope CI files).