Verification evidence for issue 12131

Verification for pingdotgg#12131

Synthetic fixture in an isolated local development server and two isolated Chromium sessions, Linux. Twelve user turns, eight agents in the oldest turn and nine in the latest. Each agent has one million synthetic tokens. No provider runs or live application data used.

Before: fresh client shows nine idle agents and 9.0M tokens. Loading earlier chat turns in the same browser restores 17 idle agents and 17.0M tokens.

After: another fresh browser shows 17 idle agents and 17.0M tokens while earlier chat turns remain unloaded. Loading earlier turns and reloading keep the same count and total.

Server regression covers agent lifecycle outside both turn and 500-activity windows, status rows without classification markers, tool heartbeats, ordering, deduplication, and exclusion of old background tasks. The test fails against unchanged source and passes with the patch.

Not tested on native macOS desktop or through a real relay/tunnel; both use the same server snapshot path.

Verification artifacts for issue 12100

Verification artifacts for pingdotgg#12100.

Before: upstream 0b83045, real Codex in an isolated Chromium session and worktree-local T3 server. Two follow-ups queued during sequential shell tools. First dispatched at 17:50:39.989Z, 168 ms after tool.completed, while original turn was running. Both follow-ups became input to that original turn.

After: both remain queued through tool completion; oldest starts a new turn only after original completes; second waits for first follow-up completion. Captured before rebase; repeated successfully on final commit b752e31 after rebasing onto f4ef515 (pingdotgg#12075). final-turns.json contains three completed provider turn IDs and proves non-overlapping start/end timestamps. timeline.json contains selected application events from before and after.

Send now and Ctrl+Shift+Enter steer one queued message while preserving remaining queue/draft. Stop restores remaining queue into existing draft. Focused checks: 338 tests, web typecheck, targeted lint (existing warnings), formatting. Claude Opus 5 review identified a late-checkpoint acknowledgement race. One condition now ignores timestamp-only updates on the previous turn; the focused regression failed before the guard and passes after it. Claude re-review found no issues. Final browser verification was repeated after this guard.

Videos play at 3x speed, with no frames intentionally removed. This is live Codex verification, not a mock provider. Native desktop and other provider runtimes were not exercised. Web and desktop share the changed queue implementation; mobile uses a separate queue.

Verification evidence for issue 12041

Isolated local web reproduction and verification of Files tree expansion persistence. Screenshots show the tree after reload before and after the fix. Recording shows switching to Diff and back, then reloading.

Issue 11930 verification evidence

Isolated browser verification for pingdotgg#11930 using real OpenSSH and a local SSH agent fixture. Before: 30-second generic fetch timeout. After: 10-second SSH key authorization error, with the prompt retained in the draft. These are PR evidence assets, not an application release.

PR evidence: issue #11903

Before and after isolated Chromium screenshots and expansion recording for issue pingdotgg#11903. Disposable fixture data; web client evidence, not native iOS verification.

Verification artifacts for issue 11899

Isolated browser captures of Settings > Connections using controlled local cloud-state fixtures. No production relay changes.

Verification artifacts for issue 11892

Verification artifacts for pingdotgg#11892. Captured in the real web composer using isolated development state copied from a read-only database snapshot.

Before: bf3be75. After: 73cbd7a.

home-before.png and home-after.png show the caret after pressing Home in the same leading-skill fixture. The videos then type BEFORE to make the insertion position visible.

The browser checks use the mounted composer, its registered node types, and actual keyboard events. check.js contains 62 selection checks; wrap-check.js contains three soft-wrap scenarios. Run each function with Playwright CLI run-code against an authenticated isolated T3 Code tab containing a composer. macOS shortcut checks override navigator.platform to exercise the application mapping in Linux Chromium; native macOS and Windows were not run.

Verification artifacts only. No application release.

Verification artifacts for issue 11810

Before/after screenshots and a short file-opening video for pingdotgg#11810. Captured from an isolated local web client with synthetic issue fixtures. Verification artifacts only.

Verification artifacts for issue 11799

Before/after and browser recovery evidence for pingdotgg#11799. Captured in an isolated local browser against event-backed fixtures. No production state or provider permission was changed.

Verification artifacts for issue 11777

Before/after screenshots, a short inspection video, and restart evidence for pingdotgg#11777. Captured with real Codex children in an isolated local web client. Before and after use separate runs of the same request; restart evidence resumes the patched child after restarting the server. Verification artifacts only.