[adampeterhiggins]

What's in this sync

Adds 458 upstream commits from November 12–13.

New features include keeping the computer awake while agents work, Codex cyber access program selection, managed Antigravity goals and CLI updates, OpenCode server usage limits, spreadsheet row/column resizing, folder drops to add projects, secret validation and mobile secret visibility.

Fixes cover provider prompts and recovery, thread navigation and queue handling, preview rendering, GitHub watch events, and remote connections. Performance work reduces history reads for long threads, speeds terminal attachment over slow connections and reduces mobile startup data.

New features

• f419698 feat(web): drop folders on the sidebar to add them as projects (feat(web): drop folders on the sidebar to add them as projects pingdotgg/t3code#35202)
• 1d218d1 feat(web): screen reader window title leads with the thread's status and name (feat(web): screen reader window title leads with the thread's status and name pingdotgg/t3code#34697)
• 315b06c feat(providers): managed Antigravity and Codex installs can update past their pin (feat(providers): managed Antigravity and Codex installs can update past their pin pingdotgg/t3code#21249)
• 63d5c6d feat(opencode): read usage limits from the connected server (feat(opencode): read usage limits from the connected server pingdotgg/t3code#34146)
• 5cff012 feat(web): resize rows and columns in spreadsheet previews (feat(web): resize rows and columns in spreadsheet previews pingdotgg/t3code#35548)
• 4099e62 feat(antigravity): support native goals and managed CLI updates (feat(antigravity): support native goals and managed CLI updates pingdotgg/t3code#35409)
• 359daaf feat(codex): choose a cyber access program for Daybreak-eligible models (feat(codex): choose a cyber access program for Daybreak-eligible models pingdotgg/t3code#27928)
• ee9e36c feat(secrets): check a secret reference before saving it (feat(secrets): check a secret reference before saving it pingdotgg/t3code#31723)
• e0d6aad feat(server): watched GitHub pull requests react when a check suite completes (feat(server): watched GitHub pull requests react when a check suite completes pingdotgg/t3code#35709)
• 587deab feat(server): keep the computer awake while agents work (feat(server): keep the computer awake while agents work pingdotgg/t3code#34128)
• 7fb54eb feat(mobile): secret field gets an eye toggle to show what you typed (feat(mobile): secret field gets an eye toggle to show what you typed pingdotgg/t3code#35783)

Notable performance changes

• a4d0a06 perf(server): sending to a very long thread reads about 5x less of its history (perf(server): sending to a very long thread reads about 5x less of its history pingdotgg/t3code#35130)
• 8b2763f perf(mobile): each environment sends about 85% less data at launch (perf(mobile): each environment sends about 85% less data at launch pingdotgg/t3code#35072)
• 7d3635e perf(server): oversized transcript exports fail fast and thread opens skip unused handoff bodies (perf(server): oversized transcript exports fail fast and thread opens skip unused handoff bodies pingdotgg/t3code#34547)
• 3e03072 perf(server): thread list reads no longer scan a long thread's runs and items (perf(server): thread list reads no longer scan a long thread's runs and items pingdotgg/t3code#35167)
• 5e2438c perf(server): sending to a thread with thousands of finished subagents reads about half as much (perf(server): sending to a thread with thousands of finished subagents reads about half as much pingdotgg/t3code#35178)
• 5911884 perf(server): sending to a long compacted thread waits about 3x less on its history read (perf(server): sending to a long compacted thread waits about 3x less on its history read pingdotgg/t3code#35180)
• 156dcd8 perf(server): long threads open 2 to 6x faster from a cold cache (perf(server): long threads open 2 to 6x faster from a cold cache pingdotgg/t3code#35286)
• 2482b03 perf(client): threads load while the socket connects, with fewer round trips (perf(client): threads load while the socket connects, with fewer round trips pingdotgg/t3code#35071)
• 4fd40e1 perf(server): Load earlier turns takes about half the time on a running long thread (perf(server): Load earlier turns takes about half the time on a running long thread pingdotgg/t3code#35224)
• a909ea4 perf(server): older history pages load without indexing the whole thread (perf(server): older history pages load without indexing the whole thread pingdotgg/t3code#34573)
• 7fab4ef perf(usage): the Usage page shows cost and tokens at once instead of after a full scan (perf(usage): the Usage page shows cost and tokens at once instead of after a full scan pingdotgg/t3code#33993)
• a1e0e50 perf(server): status refresh on a fork's default branch is 2.5x faster (median 7,826 to 3,168 ms) (perf(server): status refresh on a fork's default branch is 2.5x faster (median 7,826 to 3,168 ms) pingdotgg/t3code#33954)
• fc533a6 perf(server): avoid loading provider drivers for terminal imports (perf(server): avoid loading provider drivers for terminal imports pingdotgg/t3code#33898)
• c9f0cac perf(server,client): attaching to a terminal with a large history over a slow link drops from 5,555 ms to 751 ms median (7.4x) (perf(server,client): attaching to a terminal with a large history over a slow link drops from 5,555 ms to 751 ms median (7.4x) pingdotgg/t3code#33942)
• 719707c perf(server): save large agent recordings faster (perf(server): save large agent recordings faster pingdotgg/t3code#33930)
• a03f718 perf(web): PDF Find jumps to the first match in long PDFs 3.1x faster (median) (perf(web): PDF Find jumps to the first match in long PDFs 3.1x faster (median) pingdotgg/t3code#33956)
• a3cde98 perf(ci): run server test files in parallel on each shard (perf(ci): run server test files in parallel on each shard pingdotgg/t3code#35618)
• 8bfb723 perf(server): stop serializing terminal chunks just to measure them (perf(server): stop serializing terminal chunks just to measure them pingdotgg/t3code#24179)
• a0844ea perf(server): merged-PR threads no longer re-read git state every minute (perf(server): merged-PR threads no longer re-read git state every minute pingdotgg/t3code#24099)
• e253b88 perf(server): uploading a large document with seven notebooks finishes 1.8x faster (median 823 ms -> 461 ms) (perf(server): uploading a large document with seven notebooks finishes 1.8x faster (median 823 ms -> 461 ms) pingdotgg/t3code#33946)
• bb9658d perf(shared): command output failure scan rejects clean output in one pass (perf(shared): command output failure scan rejects clean output in one pass pingdotgg/t3code#26935)
• b7a224c perf(server): sending, editing and steering queued messages no longer read a long thread's whole history (perf(server): sending, editing and steering queued messages no longer read a long thread's whole history pingdotgg/t3code#29521)
• 0b0a61a perf(server): reduce allocation while reading usage history (perf(server): reduce allocation while reading usage history pingdotgg/t3code#36009)
• 4596872 perf(server): linked-thread counts update when links change instead of polling every 10s (perf(server): linked-thread counts update when links change instead of polling every 10s pingdotgg/t3code#35865)
• 2d8a843 perf(server): shell reads of a worktree or family no longer parse every thread payload (perf(server): shell reads of a worktree or family no longer parse every thread payload pingdotgg/t3code#36146)

Remaining commits

• Provider and orchestration fixes cover prompt delivery, interrupted turns, queues, subagent cancellation, and bounded tool responses (fix(cursor): user-invoked skills hidden from the model reach the Cursor prompt pingdotgg/t3code#34999, fix(server): keep thread tool responses small by default pingdotgg/t3code#29268).
• Repository identity follows gh's default remote, including inherited Git environment handling (fix(server): project identity follows gh repo set-default pingdotgg/t3code#35005, fix(server): repository identity honors the gh default remote pingdotgg/t3code#21574, fix(server): gh default remote lookup ignores an inherited GIT_DIR pingdotgg/t3code#35764).
• Desktop browser certificates can be approved locally; web search explains overlong queries; OpenCode HTML renders are shown and cleaned up (fix(desktop): browser panel can approve local HTTPS certificates pingdotgg/t3code#24425, fix(web): overlong content search queries explain the 256-character limit pingdotgg/t3code#35020, fix(shared): html_render pages from OpenCode 2 Code Mode show and are cleaned up pingdotgg/t3code#35022).
• Other commits tighten auth, previews, navigation, parsing, tests and CI, with dependency updates and maintenance.

Discord announcements were not used: the channel returned only October 20 messages, with no matching crossposts or announcement embeds.

Integration

Base: 1f27a682e16a2ab872ca34f57f2bba35e9394af5 (main). Pinned upstream: f313d0a6cb843551fb60e57a454a71e71a3a8c04. Range: 1f27a682e16a2ab872ca34f57f2bba35e9394af5..f313d0a6cb843551fb60e57a454a71e71a3a8c04. No parent PR. Before merging, the fork had 659 fork-only commits and was missing 458 upstream commits.

Current main advanced while verification ran. A second merge commit incorporates the GitHub file-link and Slack image-link changes from #575 and #576; upstream history remains intact.

Conflict resolutions

• Preserved fork web tabs, composer context and integrations, account/custom editor controls, citation-only sends, sidebar pages, filters and groups. Integrated upstream focus restoration and collapsed Stop action. ChatView and Sidebar changes were not file splits in this range; existing extracted fork components retain their callers.
• Kept mobile tabs/groups/filters and multi-host pull-request matching while adopting steady sidebar snapshots and generic PR helpers. Review caught a held-row subscription-limit action overwriting a newer model choice; the regression fails before the fix and passes afterward.
• Combined server orchestration, provider and browser-host changes with fork behavior and tests; kept upstream signatures, gh-default repository resolution, and provider-specific usage rates. Preserved the fork's remote-delegation extension in its documentation.
• Desktop retains quick-message and appearance layers while using upstream's shared SSH layer for updates. README uses upstream layout with the fork note; gitignore keeps both local editor and SSH fixture exclusions.

Verification

• Passed focused package typechecks for contracts, web, mobile, desktop, server, client runtime, ACP, Cursor, OpenCode and source-control definitions. Final changed-file typecheck phase passed all 41 tasks after correcting the eval fixture dependency; scripts typecheck passed separately after an initial resource kill.
• Focused tests pass: server 818 cases across 15 files, pricing/aggregation 33, web 620 across 11 files, mobile/shared 329 across 12 files plus the 47-case mobile action regression pass, desktop 27, and the eval background-wait fixture case.
• Changed-file checks passed format, lint, both Knip phases, GraphQL generation, unused patches, 34 Node tests and 318 React compiler tests. This used a scratch copy of check:changed with browser test execution excluded under the requested restriction. The broad 625-file test phase reached its 900-second timeout; it is not a passing full-suite result. The earlier parallel typecheck failure is superseded by the successful final phase and focused checks.
• Catch-up merge: all 25 changed files pass format/lint; six affected package typechecks pass; GitHubIssues, SlackApi and composerObjectLinks pass 73 focused tests.
• git diff --check is clean for resolutions and non-patch changes; the upstream Pierre dependency patch carries space-before-tab patch context, preserved unchanged.
• MCP model context generation passes. Paid model behavior evals were not run (no eval API-key .env configured). Browser tests, runtime UI automation, merges and releases are excluded per the requested scope.

Earlier runs were settled as superseded: one failed before a checkout was created; the other had a clean checkout, no pinned upstream/merge/unpushed work, and no recovery notes. Their branches/worktrees were left in place.

Merge with Create a merge commit. It keeps upstream history and the merge base for the next sync.

Model: gpt-6.1-sol. Harness: Codex in T3 Code.