PR 25255 verification
main 8b74a69019
PR c64ea208275e721a9010f8d12d21b61705944a23
Chrome /usr/bin/google-chrome via repository Playwright
Both: app-created project /home/user/question-history-verification; actual Codex gpt-6.1-sol low, same 60-git-facts prompt.
CDP conditional breakpoints count actual projector calls and row.item reads; no app source changes, mocks, synthetic rows, or imported projector calls.
main: {"calls":39,"emptyCalls":39,"rows":114,"visits":114,"nonemptyVisits":0,"maxRows":4,"restores":0}
PR: {"calls":36,"emptyCalls":36,"rows":105,"visits":0,"nonemptyVisits":0,"maxRows":4,"restores":0}
Counts differ with provider streaming; the claim is eliminated empty-request row scans, not latency or CPU.
Question path: plan mode -> Codex request_user_input -> choose Green -> provider acknowledges Green -> reload retains answer and acknowledgement.
Repeat: Square selection auto-submits -> provider acknowledges Square with no banner. Initial driver redundantly clicked Submit after auto-submit, producing an already-resolved request banner; corrected flow passed.
Default mode: provider correctly reports native request_user_input unavailable; enabling legacy Plan mode makes it available.
Blacksmith: vp test run packages/client-runtime/src/state/threadRequests.test.ts packages/client-runtime/src/state/threadDetail.test.ts -> 30 passed.
CI: 33 successful, 0 failing, 0 pending on c64ea208275e721a9010f8d12d21b61705944a23.
Not verified: native desktop/mobile shells, remote/relay/tunnel, other providers, large loaded-history timing/CPU/energy.