Skip to content

chore: sync the fork with upstream 16d40942ac - #24

Merged
radys-s merged 72 commits into
mainfrom
tomas/sync-upstream-2026-11-20
Nov 20, 2026
Merged

radys-s merged 72 commits into
mainfrom
tomas/sync-upstream-2026-11-20

Conversation

@radys-s

@radys-s radys-s commented Nov 20, 2026

Copy link
Copy Markdown
Owner

Problem

After #22, fork main was 69 upstream commits behind (up to the v0.1.11 release), so it was missing upstream's newer server, client and provider fixes. This brings in upstream main 16d40942ac and keeps every fork feature.

Change

Fork features kept: Auto Low/Medium/High routed by live limits and weekly pace, failover to the next subscription on a usage limit, the Auto level: line in Auto-routed Claude prompts, Auto in the model picker, Resume on every provider, the pooled usage view and reset advice, unknown (not 0%) usage for Grok unified-billing accounts, and the fork's Resume-after-limits and Usage copy in every language.

Verification (hub2)

  • Fork diff before and after. The fork's diff against upstream is 76 files both before (66b8a94f8d..9b157f83e0) and after the merge (16d40942ac..04ba054acf). Per file, the lines the fork adds and removes are identical in 66. The other 10 are web catalogs where one key line counted as moved in the old diff. Parsed as JSON, all 20 web and mobile catalogs equal the old base plus the fork's key changes plus upstream's, with no key changed by both.
  • vp run check:changed 16d40942ac (the fork's 76 files): unused patches, format, lint, knip, graphql, typecheck of 42 packages, 1107 tests in 31 files and react compiler all pass.
  • vp run i18n:check: catalogs match the source in web, mobile and desktop.
  • Real boot on the worktree's own .t3, seeded by vp run migrate-dev-db from a read-only snapshot of the live database (never ~/.t3/userdata), with settings holding only the two Auto instances and auto-resume off. No agent turn was sent.
    • Migrations 101_DelegatedTaskReadIndexes (from chore: sync the fork with upstream 66b8a94f8d #22, not yet deployed) and 102_ThreadItemCounts applied. The new per-thread item totals match the per-run counts for all 28 seeded threads.
    • Auto's resolver replayed on the snapshots the server cached (A at 99% weekly, past the 90% guard; B at 78%): Low → Auto B Sonnet 5.5 at low, Medium → Auto B Opus 5.5 at medium, High → Auto B Opus 5.5 at max.
    • Trace: 3406 spans in about 4 minutes, 2 failures (the managed Codex install the fresh data dir lacks, one transport error from Claude's usage endpoint). All three Claude provider checks passed.
  • Review: one read-only review of the merge found no semantic conflicts and confirmed each call above. Failover sets no deliveryIntent, so fix(server): agent messages queued behind /compact keep send order pingdotgg/t3code#39417 never queues it. A stream's failure waits for its queued snapshots, so the error item commits before run.updated: failed. Its one finding, Auto routing to a broken runtime, is the fix above.

Notes

  • The CI workflow is disabled on the fork, so the local checks above stand in for it. The agent context budget test passed among the 31 files.
  • Switching the hubs: migrations 101 (two partial indexes) and 102 (a table filled from the existing counts, an index and five triggers). The app version is now 0.1.11. The Node pins and the lockfile did not change.
  • Antigravity A and B on hub2 stay unusable until they are updated to 1.5 or newer in Antigravity setup. The live build already refuses them through upstream's live manifest; with this PR, Auto also stops routing to them.
  • The real run's first boot inherited the agent's CLAUDE_CONFIG_DIR (Claude B), so every Claude instance without its own home path read account B. The numbers above come from a second boot with the Claude session variables unset. On a fresh data dir, the dev-only startup welcome thread is created about 4 s before the first Claude probe returns, so Auto rejects it and it is skipped with a warning. t3 serve, which the hubs run, never creates it, and the path is unchanged by this merge.
  • Upstream main moved 3 commits during the sync (f2d4df0367: fix(server): Codex web searches list the query once pingdotgg/t3code#39585, fix(web): keep reopened pairing link dialogs open on completion pingdotgg/t3code#39579, fix(composer): keep screenshots preparing after an unanswered question closes pingdotgg/t3code#39569). They touch none of the fork's files and go in the next sync.
  • The branch descends from fork main, so a merge commit (gh pr merge --merge) works without a force-push.

juliusmarminge and others added 30 commits November 19, 2026 14:09
…rs (pingdotgg#33568)

Co-authored-by: Julius Marminge <julius0216@outlook.com>
…dotgg#39395)

Co-authored-by: Julius Marminge <julius0216@outlook.com>
…el (pingdotgg#39415)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…elegating (pingdotgg#39413)

Co-authored-by: Simone <185146821+Lucenx9@users.noreply.github.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>
…ect loop for good (pingdotgg#39420)

Co-authored-by: Julius Marminge <julius0216@outlook.com>
…ain (pingdotgg#37419)

Co-authored-by: Claude Haiku 5.5 <noreply@anthropic.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…t and @ (pingdotgg#38323)

Co-authored-by: Yash Singh <saiansh2525@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…ider versions (pingdotgg#39427)

Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
…fest (pingdotgg#39434)

Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
…urn ends (pingdotgg#38869)

Co-authored-by: Julius Marminge <julius0216@outlook.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Yash-Singh1 and others added 29 commits November 19, 2026 21:20
…g#39493)

Co-authored-by: Julius Marminge <julius0216@outlook.com>
…"Ran command" (pingdotgg#39474)

Co-authored-by: Julius Marminge <julius0216@outlook.com>
…erseded ones (pingdotgg#39502)

Co-authored-by: Cursor <cursoragent@cursor.com>
…6499)

Co-authored-by: Julius Marminge <julius0216@outlook.com>
…ents that failed (pingdotgg#39536)

Co-authored-by: Julius Marminge <julius0216@outlook.com>
…dotgg#39552)

Co-authored-by: Julius Marminge <julius0216@outlook.com>
Co-authored-by: Julius Marminge <julius0216@outlook.com>
…ingdotgg#39548)

Co-authored-by: MarcoDotIO <20867620+MarcoDotIO@users.noreply.github.com>
pingdotgg#39562)

Co-authored-by: Julius Marminge <julius0216@outlook.com>
Syncs the fork with pingdotgg/t3code main, 69 commits newer than the
last sync base 66b8a94. A real merge keeps fork main an ancestor, so
the PR merges without a force-push.

No conflicts. The 29 files changed on both sides merged cleanly: the
web and mobile i18n catalogs, ClaudeAdapterV2.ts, Orchestrator.ts,
server.ts, ProviderRegistry.test.ts, ChatView.tsx, settingsSearch.ts,
docs/user/thread-sidebar.md, and threadExecution.ts and its test.

Origin: Tomáš asked to sync the fork with upstream.
…oken

Upstream pingdotgg#38970 rates Antigravity below 1.5 broken for T3 0.1.10 and up,
now in the bundled manifest too, and pingdotgg#39427 makes the server refuse a
session on any provider whose version the manifest rates broken. Auto's
usable() did not read that rating, so once both Claude accounts were
out, Auto could pick a runtime the server refuses, and each turn failed
with a setup error that failover does not handle.

On hub2, Antigravity A and B run 1.4.5 and still list Gemini 3.8 Flash,
and B's Gemini pool is 7% used. Replaying Auto on hub2's live provider
snapshots with both Claude accounts exhausted, Auto Low went to
Antigravity B before this change and goes to OpenCode with it.

Origin: review of the upstream 16d4094 sync (merge 04ba054).
What the sync merged, the checks of the fork's diff before and after,
the Auto fix for runtimes the manifest rates broken, the real run on a
seeded copy of the live database (migrations 101 and 102), the review
result, and what switching the hubs needs.

Origin: Tomáš asked to sync the fork with upstream.
@radys-s
radys-s merged commit f4e573d into main Nov 20, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.