Rebase fork onto upstream 89c5a19 and refresh FORK.md

[jmclaren7]

Important

Do not click Merge. GitHub reports this PR as conflicting (mergeable_state: dirty) and that is expected: main still holds the pre-rebase fork commits (ead4077), and this branch holds the same changes replayed onto a new upstream base with new SHAs. There is no sane merge between the two.

This PR exists for review only. Once approved, the intended integration is a force-push of claude/dazzling-clarke-vxb3p2 onto main, rewriting main's history to the new upstream base — the same procedure used for the 2026-07-23 reset. The pre-rebase tip ead4077 and the older lineage (pre-reset-main-2026-07-23 / 8744b86) both remain on the remote as recovery points.

Because of this, the "197 changed files / +16,725 −1,824" figure above is main-vs-branch and is mostly upstream's own 36 commits. The fork-authored diff is the 37 files listed below.

What Changed

Rebases this fork's five commits from base b41e89e onto pingdotgg/t3code main at 89c5a19 (fix(dev): skip browser-blocked ports (pingdotgg#4608)), taking in 36 upstream commits. Replayed with git rebase --onto; no textual conflicts.

Fork commits after the replay:

6829ccb	Make CI fork-runnable (standard runners, drop mobile job)
f186a73	Forward Ctrl+<letter> chords to the embedded terminal
bef9432	Add fork identity docs (FORK.md, README banner, AGENTS.md policy)
646dcbe	FORK.md: record 2026-07-23 reset outcomes
2fa5c7d	Add Copilot and Gemini CLI providers (ported from aadit)
529dd99	FORK.md: record 2026-07-27 rebase (this PR's only new work)

Net fork diff vs upstream is unchanged in shape: 37 files, ~7.1k added lines.

FORK.md — added a fork-purpose paragraph and a last-rebase marker, restructured Dropped / now upstream into Superseded changes with a table pairing each dropped change to the upstream change that replaced it, and reconciled a contradiction the file had been carrying (entries 1, 8, 10 were listed as Active with "keep" verdicts while also appearing under Dropped). Entry numbers are stable IDs tied to the original commits, so the Active list now documents its gaps rather than renumbering.

README.md — the fork banner had gone stale, advertising two superseded changes (Windows path-with-spaces builds, hover timestamps). Rewritten to match, and to separate what is on main from what is still on the unmerged pr/* branches.

Why

A clean replay is not the same as a correct one, so every active FORK.md entry was re-checked against the incoming range:

• Entry 9 (always-visible new-thread button) is now partially superseded. Upstream ships the button with the same data-testid="new-thread-button" and the environment-label tooltip this entry added — but still hover/focus-gates it on desktop (opacity-0 → group-hover/project-header:opacity-100), always-visible only under max-sm. The intent survives; the tooltip half is upstream's now. Flagged to re-derive from upstream's current Sidebar.tsx instead of re-applying the stale hunk.
• Entry 11's modify/delete conflict on TODO.md is moot — TODO.md is no longer in upstream's tree at all.
• Entry 1 is superseded by edb1240 (fix(cli): publish nightly branded favicons pingdotgg/t3code#4372), which reworked the cli.ts build spawn to hardcode shell: false — independently arriving at exactly this fork's fix. Entry 8 is superseded by formatChatTimestampTooltip.
• Entries 2, 3, 4, 6, 7 re-verified as not superseded, with evidence recorded against 89c5a19 rather than the stale ada410b marker.

Verification at 89c5a19 (node 24.18.0, pnpm 11.10.0):

• pnpm install --frozen-lockfile — clean. This is the load-bearing check: upstream rewrote ~145 lines of pnpm-lock.yaml and reshuffled the pnpm-workspace.yaml catalogs in this range (Clerk upgrade, Upgrade Clerk packages and Expo integration pingdotgg/t3code#4440) while the fork adds its own @github/copilot* importer entries, and a lockfile that replayed badly is the most likely silent breakage in a rebase that reports no conflicts.
• pnpm run typecheck — 0 errors across all 15 packages (suggestion-level Effect hints only, several pre-existing upstream).
• pnpm run test — 294/295.

The one failing test is pre-existing upstream and environmental, not a rebase regression

packages/shared/src/Net.test.ts → "findAvailablePort returns preferred when it is free".

Upstream's isPortAvailableOnLoopback requires a port to be bindable on both 127.0.0.1 and ::1, and canListenOnHost only treats EADDRNOTAVAIL as "available". On an IPv6-less host ::1 returns EAFNOSUPPORT, so the preferred port is judged unavailable and findAvailablePort falls back to a different port, failing the assertion.

The verification container has no IPv6 (confirmed directly: binding ::1 returns EAFNOSUPPORT). This is upstream code the fork does not touch — git diff upstream/main HEAD -- packages/shared/ is empty — and it arrived with pingdotgg#4608. Recorded in FORK.md's verification note so it is not chased again on the next rebase.

Checklist

• This PR is small and focused — one new commit; the rest is an upstream replay
• I explained what changed and why
• I included before/after screenshots for any UI changes — n/a, no fork-side UI changes
• I included a video for animation/interaction changes — n/a