[yaacovcorcos]

What Changed

Follow-ups to #861's untracked backoff and Codex kill deadline, which T3 has since merged as pingdotgg/t3code#34460 and #34461. All changes sit in the existing fork-marked blocks and .scient.test.ts files:

• One listing for overlapping reads: a status read that arrives while the same git ls-files --others listing is running waits for it instead of starting another walk of the untracked tree. This uses an Effect Cache with zero time-to-live, so only an in-flight listing is shared, and a shared listing keeps running until the last read waiting on it leaves. Each read still builds its own temporary review index.
• Original error kept: the branch-totals error keeps the listing's GitCommandError as cause, for a fresh failure and for the remembered one.
• Codex test checks "ready": the app-server kill-deadline test checks that the stand-in printed ready before stopping it, so an early exit can no longer pass without testing the force-kill.
• UPSTREAM.md: records that #34460 and #34461 are merged in T3 and links the two new T3 PRs.

Why

CodeRabbit raised these three points on the T3 PRs, and they were merged without being addressed. A status read restarted by a file change during a slow listing (up to the 30 s timeout) started another full walk of the tree before the backoff could apply. The same changes are offered upstream as pingdotgg/t3code#34574 (listing and cause) and #34575 (test). When they merge, the next alignment takes T3's version and drops these blocks.

Verification

• New test share one listing between overlapping status reads (GitVcsDriverCore.scient.test.ts): it holds the first read's listing until the second read's last Git command before its own listing has returned.
  • On main: fails with 2 listings.
  • With this change: passes with 1 listing.
  • In the T3 worktree, the same test failed 3 of 3 runs on T3 main and passed 5 of 5 with the change.
• Ready check: with the stand-in replaced by exit 0, the old test passes and the new one fails.
• The cause change has no test of its own: status reports a totals failure as missing totals, so the cause shows only in the readBranchChangeTotals span.
• Suites run: GitVcsDriverCore.scient.test.ts and CodexAdapterV2.scient.test.ts (86 tests), server typecheck, vp lint, vp fmt --check, seam check and provenance all pass on macOS.
• Review: Codex (gpt-6.1-sol) reviewed all three changes and found nothing blocking. It noted the test's count of branch checks would need updating if the status read's command sequence changes.

Platforms: no platform-specific code. The shared listing behaves the same on macOS, Linux and Windows. The Codex test stays skipped on Windows, as before. Run on macOS only; CI covers Linux and Windows.

Documentation Impact

Documentation impact: Updated — UPSTREAM.md

UI Changes

None.

Made with Claude Opus 5.5 in Claude Code; reviewed by Codex (gpt-6.1-sol).

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.5 KiB	5.5 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	4.2 KiB	4.2 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.3 KiB	1.3 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	22.3 KiB	22.3 KiB	0 B (0.0%)	24.4 KiB	✅
Codex	Live turn messages	1	1	0 (0.0%)	8	✅
Claude	Total thread wire	5.5 KiB	5.5 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	4.2 KiB	4.2 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.3 KiB	1.3 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	22.7 KiB	22.7 KiB	0 B (0.0%)	24.4 KiB	✅
Claude	Live turn messages	1	1	0 (0.0%)	8	✅

Baseline: 9706640 · PR result: 9fefdd6 · 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.
Bounded compact HTTP snapshot with live cursor.

• Codex decoded thread snapshot: 88.2 KiB
• Claude decoded thread snapshot: 88.4 KiB

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