[alexcstark]

Sending a message with an attachment could leave the conversation empty until the upload finished. The conversation now shows the local message immediately while attachment preparation continues.

Failed sends remove the unsent bubble and restore the draft. If newer input exists, it is preserved and a persistent Restore message action keeps the failed snapshot available.

CI exposed existing fork blockers that also needed repair:

• Pass runtime objects from both composers and use compiler-compatible notification/settings code. Track both mobile thread-group callbacks in the memoized handler.
• Recognize newer Git's invalid-checkout diagnostic while retaining the worktree and still reporting unexpected errors.
• Fix the thread-group label layout, release-link repository fixtures, native test isolation, browser selectors, focus/hover timing, the usage fixture's clock, and the prompt-editor test's circular import.

Validation includes seven message-preparation browser regressions and 147 focused unit tests, mobile/web typechecks, 11 release-link tests, compiler checks, and focused browser suites. The latest repair additionally passes five server cases, seven mobile route cases, one compiler case, and 19 browser cases. The prompt-editor suite now completes. The original right-panel CI failure did not reproduce locally; its fixed-frame wait was replaced with deterministic timer draining. CI for the repaired revision runs the full suites.

T4 version .26 builds from the repaired revision, and its local signature, embedded runtime, bundle identity, and profile settings verify. Packaging uses exact patched-dependency pins only in a local build checkout; dependency manifests in this PR are unchanged.

Full-app conversation verification remains unverified because the required T3 Browser panel tools were unavailable. The gate records an inconclusive live-test exception; no full-app screenshots or video were captured.

Model: GPT-6; harness: Codex. Reviewed with Opus in Claude Code and the no-mistakes gate.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ The exact PR base did not have a successful artifact. Baseline uses the latest successful main measurement shown below.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	+7 B (+0.1%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	+7 B (+0.6%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.9 KiB	20.9 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: 99c96b0 · PR result: 04d1059 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.