ci(t3x): raise the test timeout for the fork's slower runners

The sync branch's CI failed on apps/web/src/lib/stashImageCompression.test.ts
timing out at 15s. A control run proved this is upstream-inherited and not a
rebase defect: branch t3x/ci-control-upstream — pristine upstream/main plus
only this workflow file, zero fork patches — fails identically (run
30386125253).

Cause is an environment mismatch, not a bug. apps/web/vite.config.ts pins
testTimeout to 15s, tuned for upstream's blacksmith runners. This fork runs on
2-core ubuntu-latest. The test is the worst-case path in its file (every
quality step fails, so it runs the full nested loop) at ~3s locally — 6x the
next slowest test there — which leaves too little headroom on slower hardware.

Fixed in the fork-owned workflow rather than by patching apps/web/vite.config.ts,
which would add a 35th row to the seam ledger and permanent rebase cost for what
is purely a CI-environment problem. 60s is what the root vite.config.ts already
uses, so this only stops apps/web's tighter override from applying on hardware
it wasn't tuned for; timeout-minutes remains the backstop for a hung test.

Also documents the control experiment in the sync runbook, so the next red
verify gets diagnosed instead of assumed.

Review fixes (found by an independent review of this work):

  - SEAMS.md: the settings.ts risk was 70, but the row is +7/-2 = 9 lines
    against churn 14 = 126. Corrected and re-sorted. The other 33 rows and both
    totals were verified correct.
  - SEAMS.md: the header pinned merge-base 89c5a19, which this very sync makes
    stale. Updated to 887dd6e, with a note that churn/risk still reflect the
    old base until the absorbed commits age into the window.
  - SEAMS.md: the regeneration command used `--since="60 days ago"` (anchored to
    today) while the prose defines churn relative to the merge-base, so the
    ledger was only reproducible on the day it was written.
  - claude.md: attributed account identification to `claude auth status`; the
    app actually reads the SDK initialization result. Reworded so the command is
    presented as a way to observe the effect, not as the mechanism.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: radroid

.github/workflows/t3x-ci.yml

@@ -47,6 +47,17 @@ jobs:
         if: ${{ !cancelled() }}
         run: vp run lint
 
+      # `--testTimeout` is raised because `apps/web/vite.config.ts` pins testTimeout to 15s, which is
+      # tuned for upstream's blacksmith runners. This fork runs on 2-core ubuntu-latest, where
+      # upstream's `stashImageCompression` worst-case test (~3s on an M-series Mac; 6x the next
+      # slowest in its file) reliably exceeds 15s and fails the whole gate.
+      #
+      # Proven upstream-inherited, not a fork regression: branch `t3x/ci-control-upstream` — pristine
+      # upstream/main plus this workflow file and nothing else — fails identically (run 30386125253).
+      #
+      # 60s is not arbitrary: it is what the ROOT vite.config.ts already uses. This only stops
+      # apps/web's tighter local override from being applied on slower hardware than it assumes.
+      # The job's own `timeout-minutes` remains the backstop against a genuinely hung test.
       - name: Test
         if: ${{ !cancelled() }}
-        run: vp run test
+        run: vp run test --testTimeout=60000

docs/providers/claude.md

@@ -33,9 +33,11 @@ normal login.
 
 > **Do not type `~/.claude` into that field.** Leaving it empty and pointing it at the default
 > directory are not the same thing. Setting `CLAUDE_CONFIG_DIR` explicitly changes how Claude Code
-> looks up its stored credentials, and the account stops being identified — `claude auth status`
-> still reports `"loggedIn": true` but returns `"email": null`, so T3 Code can no longer show you
-> which account the instance is using. Leave the field empty for your main account.
+> looks up its stored credentials, and the account stops being identified: you stay logged in, but
+> the account's email is no longer reported, so T3 Code cannot show you which account the instance is
+> using. You can see this for yourself — `claude auth status` returns your email with no prefix, but
+> `CLAUDE_CONFIG_DIR=~/.claude claude auth status` returns `"loggedIn": true` with `"email": null`.
+> Leave the field empty for your main account.
 
 ## I Want Work And Personal Claude Accounts
 

docs/t3x/SEAMS.md

@@ -3,8 +3,12 @@
 **The authoritative list of every upstream-owned file this fork edits.**
 
 Measured, not asserted: **34 upstream-owned files, +1466 / -112 lines**, against merge-base
-`89c5a192f`. Everything else the fork adds lives in new files upstream has never seen and cannot
-conflict.
+`887dd6e45` (the 2026-07-28 upstream sync). Everything else the fork adds lives in new files upstream
+has never seen and cannot conflict.
+
+The churn and risk columns are still measured against the _previous_ merge-base `89c5a192f`, because
+the 45 commits absorbed by that sync have not yet aged into a 60-day window. Recompute them at the
+next sync.
 
 Regenerate this ledger before trusting it — see [Regenerating](#regenerating) at the bottom. An
 earlier version of this file claimed the surface was 2 files and "Contracts / persistence: _None._"
@@ -45,10 +49,10 @@ Sorted by risk, worst first.
 | `apps/desktop/src/backend/DesktopBackendConfiguration.ts`               | +29      | 5     | **145**  | Backend heap headroom (`NODE_OPTIONS`)                                                                                                           |
 | `apps/web/src/components/chat/ComposerPrimaryActions.tsx`               | +53/-13  | 2     | **132**  | Queue button; extracts upstream's inline stop button                                                                                             |
 | `apps/desktop/src/backend/DesktopBackendConfiguration.test.ts`          | +42      | 3     | **126**  | Heap-headroom assertions                                                                                                                         |
+| `packages/contracts/src/settings.ts`                                    | +7/-2    | 14    | **126**  | `notifyOnNeedsInput` (**persisted schema**) + Claude `homePath` placeholder/description                                                          |
 | `packages/shared/src/composerTrigger.test.ts`                           | +31/-1   | 3     | **96**   | `replaceTextRange` newline coverage                                                                                                              |
 | `apps/server/src/sourceControl/SourceControlProviderDiscovery.ts`       | +20/-8   | 3     | **84**   | Issue #4: CLI probe timeout + spawn-error classification                                                                                         |
 | `apps/server/src/server.ts`                                             | +3       | 25    | **75**   | The intended mount point: one import, one `Layer.provideMerge`, one route entry                                                                  |
-| `packages/contracts/src/settings.ts`                                    | +7/-2    | 14    | **70**   | `notifyOnNeedsInput` (**persisted schema**) + Claude `homePath` placeholder/description                                                          |
 | `apps/desktop/src/main.ts`                                              | +4       | 17    | **68**   | `ElectronNotification` layer                                                                                                                     |
 | `apps/web/src/connection/platform.ts`                                   | +7/-1    | 7     | **56**   | Lazy `import()` of outbox cleanup to dodge a module-init cycle                                                                                   |
 | `apps/web/src/routes/__root.tsx`                                        | +6       | 9     | **54**   | Mounts `<NotificationCoordinator>`, `<ThreadOutboxDrain>`, `<PushSubscriptionManager>`                                                           |
@@ -115,11 +119,17 @@ git diff --numstat "$MB"..HEAD | while read -r a d p; do
 done
 ```
 
-That prints exactly the upstream-owned files the fork edits. Churn for any one of them:
+That prints exactly the upstream-owned files the fork edits. Churn for any one of them — anchored to
+the merge-base date, **not** to today, so the number is reproducible after the fact:
 
 ```bash
-git log --oneline --since="60 days ago" "$MB" -- <path> | wc -l
+SINCE=$(git show -s --format=%cI "$MB")
+git log --oneline --since="$SINCE" --until="$SINCE" --before="$SINCE" "$MB" -- <path> | wc -l
+# simpler and equivalent: the 60 days of upstream history ending at the merge-base
+git log --oneline "$MB" --since="$(git show -s --format=%cI "$MB") -60 days" -- <path> | wc -l
 ```
 
-Re-run both after every upstream sync and update this file. If the ledger and this document
-disagree, the ledger is right.
+**Re-run both after every upstream sync and update this file — including the merge-base hash in the
+header.** Every churn and risk figure shifts once the newly absorbed commits fall inside the window,
+so a ledger quoting an old merge-base is stale even when its file list is still right. If the
+regenerated ledger and this document disagree, the ledger is right.

docs/t3x/sync-agent-runbook.md

@@ -68,6 +68,28 @@ created) shows up in the Actions tab as **skipped**, not as an error. Check ther
 a rebase that is not the problem. Read the `**Result:**` / `**kind:**` line first; a `weekly-build`
 failure means fix the build.
 
+### When verify goes red on code the fork does not own
+
+A sync can import an upstream test that fails **in this fork's CI but not upstream's**, because
+upstream runs on `blacksmith-*` runners and the fork runs on 2-core `ubuntu-latest`. Before treating
+a red verify as a bad rebase, run the control experiment — it takes two minutes and gives a
+definitive answer:
+
+```
+git switch --detach upstream/main
+git checkout <sync-branch> -- .github/workflows/t3x-ci.yml
+git switch -c t3x/ci-control-upstream && git commit -am "ci: control run" && git push -u origin HEAD
+gh workflow run t3x-ci.yml -R radroid/t3code --ref t3x/ci-control-upstream
+```
+
+That branch is pristine upstream plus one fork-owned file. If it fails the same way, the failure is
+upstream-inherited and no fork patch caused it. Delete the branch afterwards.
+
+Fix such failures in `.github/workflows/t3x-ci.yml` (fork-owned) rather than by patching the upstream
+file — patching adds a row to `docs/t3x/SEAMS.md` and permanent rebase cost for a CI-environment
+problem. The `--testTimeout` override on the Test step exists for exactly this reason; its comment
+records the case.
+
 ## What the agent does (and what a human doing it locally should do)
 
 This is the checklist the workflow prompt mirrors — follow it if you resolve locally instead.