PR 24468 runtime verification, 2026-11-01, Linux Boat VM

Baseline: github/main ce9e016406 (separate /home/user/main worktree)
vp i: passed
vp run dev: server 16733, web 8693, isolated .t3, automatic seeding disabled
Playwright 1.60.0, /usr/bin/google-chrome, actual composer UI, shared browser profile
Created project for /home/user/t3code through Add project > Local folder.
Selected Codex GPT-6.1-Sol and submitted: Reply exactly BASELINE_READY. Do not use tools.
Actual assistant response: BASELINE_READY
Tab A stashed: Reply exactly STASH_A_VERIFIED. Do not use tools.
Tab B stashed: Reply exactly STASH_B_VERIFIED. Do not use tools.
Reloaded A. Storage and stash menu held only STASH_B_VERIFIED. STASH_A_VERIFIED was lost.
Stopped baseline by captured process group PID before starting PR.

Candidate: 3bee846df581dfc1ad1a26a378d19307ce6ff260, exact contributor head
vp i: passed
vp run dev --home-dir /home/user/t3code/.t3: server 13773, web 5733
Copied only task-created main state with both servers stopped. No mocks, seeded fixtures or demo data.
Repeated the identical two-tab UI flow with the same prompt text.
A's live badge became 2 after B stashed.
After A reloaded, menu and persisted entries contained STASH_B_VERIFIED and STASH_A_VERIFIED.
Restored B via its menu row, submitted through composer, actual Codex GPT-6.1-Sol response: STASH_B_VERIFIED.
Deleted A in B. A's badge disappeared. Persisted entries: [].
Stashed a neighboring prompt in A. B observed only that prompt, with no deleted entry resurrected.
Ctrl+S in B's empty composer restored the sole entry.
Attached the actual baseline screenshot through the composer file input, stashed it, and waited for real image encoding.
Reloaded B, restored the image prompt from the stash, and submitted through the composer.
Actual Codex response: the screenshot shows a t3 code thread with a "BASELINE_READY" response and one stashed prompt requesting "STASH_B_VERIFIED" above the empty composer.

Blacksmith: vp i && vp test run apps/web/src/promptStashStore.test.ts
Result: 1 test file passed, 17 tests passed. Expected quota-rejection stderr came from existing tests.
GitCafe reported 29 successful checks, 0 failing, 0 pending at the exact PR head.
No source edits, commits or pushes needed.

Not exercised in the real app: native desktop/mobile, Firefox/Safari, relay/tunnel, storage-quota rejection, suspended/crashed image encoders and the 15-minute fallback without Web Locks.