chore: sync upstream v0.1.3 (T3O-74) - #199

High-level summary

This PR merges upstream v0.1.3 into t3o as a two-parent merge commit with six follow-up commits. It re-seats every T3o seam on upstream's reworked code: gitenv now sits on GitHubRequestPolicy.credential and the GraphQL batchKey, the T3O-8/T3O-54 scope fallback is rebuilt on upstream's generated operations, and the board settings route, worker prefill, Tiptap destroyed-editor guard and migration 031 are updated for upstream's changes. Fork-only lint and Effect-diagnostic deferrals sit in clearly marked override blocks.

I traced the riskiest seams end to end:

• checksUnread still reaches boardMergeState through the provider's spread.
• The summary fallback's filled commits: {nodes: []} decodes to checksState: null (unknown), not "passing".
• isMissingScopeRefusal matches the GitHubApiResponseError that upstream's fromClientError produces.
• The Add environment prefill isn't clobbered by the dialog's open handler.
• Migration 031 now skips the carry-over cleanly when upstream has dropped projection_state.

The merge looks sound. This round has one improvement and one nitpick. The improvement is a board test whose clock advance was widened during the sync with a rationale that doesn't hold for one of the two tests. The nitpick is board copy that still names the old "Connections" page. Nothing else blocks.

High-level summary — round 2 (re-review at 03901c5)

This PR merges upstream v0.1.3 into t3o and re-seats the fork's seams: the gitenv credential and batchKey, the GitHub scope fallback rebuilt on upstream's generated operations, migration 031's guard for upstream dropping projection_state, the board MCP registration test, the settings-route moves, and the worker copy. Both round-1 fixes hold at this head. The draft-merge tests now advance exactly to the first 30s sweep tick that can follow the jittered rung, and the board copy names Environments and Sharing & access. Reading the merge with fresh eyes turned up one new improvement. The fork's root test.maxWorkers: 4 conflicts with the root projects list that upstream added in this release. From the repo root, any vp test run that touches both the root project and a package project now fails before running a single test, and so does the test step of vp run check:changed. Triage put that failure down to something this card didn't touch, but this merge is what causes it. There are also two nitpicks: the board's lint override switches off two test-hygiene rules alongside the style rules, and one doc comment still says "Connections". That makes 1 blocking improvement and 2 non-blocking nitpicks.

Status of round-1 findings

Findings GitHub can't anchor inline

This PR changes 6,242 files, and GitHub's review API only resolves the first 3,000. vite.config.ts falls outside that, so these two findings can't be inline comments. Each one gives its exact file and lines instead.

vite.config.ts:102-107

Upstream v0.1.3 added a root test.projects list, used when isRepoRootTestRun. It holds every package in TEST_CONFIG_PACKAGES plus a root project with extends: true. The root project inherits this fork-only maxWorkers: 4, but @t3tools/web (unit) and the other package projects do not. Vitest refuses to mix projects that have different maxWorkers and the same sequence.groupOrder. Reproduced at this head:

$ vp test run packages/shared/src/boardThreadState.test.ts apps/web/src/board/workers/workerCopy.test.ts
Error: Projects "@t3tools/web (unit)" and "root" have different 'maxWorkers' but same 'sequence.groupOrder'.
 Test Files  no tests

With line 107 removed, the same command passes 2 files and 18 tests. origin/t3o has no root projects, so the merge introduced this break. It takes out the test step of vp run check:changed, which runs one root vp test run across all the changed packages. That is the pre-push gate in AGENTS.md, and this repo runs no CI checks on the PR to catch it instead. The triage note filed it as "not touched by this card", but the fork's seam colliding with upstream's new config is exactly what this card has to resolve.

Suggested fix: keep the memory bound without making projects disagree. Apply pool/maxWorkers only when !isRepoRootTestRun (package runs and vp run -r test still get it). Or give the root project its own sequence.groupOrder. Then record the decision in the seams.md v0.1.3 notes, and rerun vp run check:changed to confirm the test step runs.

vite.config.ts:667-668 (and :698)

This block is described as deferring upstream's style rules (tokens, React Compiler, copy), and T3O-75 removes it later. It also switches off t3code/no-wall-clock-in-tests and t3code/no-real-time-wait-in-tests for all of apps/web/src/board/**. The server block at line 698 does the same for wall-clock checks. Those two rules prevent flaky tests rather than enforce style. While the override stands, any new board test can sleep or read Date.now() without lint objecting. Only 8 existing call sites trip them: 5 in BoardCardItem.test.tsx, plus boardCardNotice.test.ts:232, BoardModeTabs.test.ts:49 and gitenv.test.ts:137.

Suggested fix: fix those 8 sites, or put per-line oxlint-disable-next-line directives with a reason on them. Then drop the two rules from both override blocks, so the deferral covers only style rules.

High-level summary — round 3 (re-review at 8cf062e)

This PR merges upstream v0.1.3 into t3o. It re-seats every T3o seam: gitenv on GitHub request policy, the scope fallback on the generated GraphQL operations, board migration 031, the settings routes and the worker prefill. It also adjusts the fork's test and lint config so they fit upstream's new tooling. This round re-reviewed the latest triage commit (8cf062e0f5) and re-checked the whole diff for leftover conflict markers and mis-cased T3O: markers. The cap in vite.config.ts now applies only to per-package runs, which matches the vp run -r --concurrency-limit 1 test path it was written for. The eight board test sites now pin their clocks instead of reading wall time. BoardModeTabs.test.ts now awaits router.latestLoadPromise. I checked that against TanStack router-core 1.171.8: with no history subscribers, commitLocation calls load() synchronously, so the promise is set before the test awaits it. From the repo root, the touched tests plus the draft-merge retry suite pass: 5 files, 108 tests, and the groupOrder error is gone. There are no new findings this round, so nothing blocks. The open items are the ones the build already flagged: the manual browser checklist, disabling the three inherited workflows after landing, and landing with a merge commit rather than a squash.

Status of round-2 findings