[danspark]

Follow upstream nightly https://github.com/pingdotgg/t3code/releases/tag/v0.1.7-nightly.20261101.4048, source 88f6113.

Reviewed all 908 upstream commits from v0.1.4-nightly.20261027.3796 (b88da91) to v0.1.7-nightly.20261101.4048 (88f6113). No new analytics, crash-reporting, trace or metrics exporter, telemetry SDK or external collection endpoint was added. Upstream added managed policies that can only turn product analytics and relay tracing off, desktop shell-consent gating for analytics, server-side npm release lookups that read public package metadata the web client already fetched, user-initiated triage that clones the public upstream source and asks before posting, more URL credential redaction, and desktop update-track switching with an optional administrator feed URL. None of these export private data, and fork update destinations are unchanged. Re-anchored the AnalyticsService no-op guard ahead of the new consent and policy reads, and the fixed T2 artifact name after upstream added afterPack. Re-anchored privacy tests after upstream changes: the browser-trace test import, the relay-span cases now that the bridge method is required, the analytics tests including new shell-consent and managed-policy cases that now expect nothing sent, and the private task-cache hook test after upstream deleted the native-runner test. Retired three release-CI repairs the pinned upstream now contains: the saved-usage clock pin and the desktop launch settings write, replaced upstream by a fixture that disables auto-settle. Kept the storage-cleanup Git 2.56 fix, prior security fixes, fork-owned privacy files, Linux x64/ARM64 packaging, dependency versions and lockfiles.

Application changes are limited to the recorded analytics patches and installer metadata. Full CI must pass before merging.

Model: Claude Opus 5.5. Harness: Claude Code in T3 Code.

🤖 Generated with Claude Code

[danspark]

Follow-up for 87a5d475:

• Test Server 2: two tests added upstream expected exports that T2 blocks: an OTLP log export of a redacted request failure, and relay tracing with no managed policy. 9ed051f3c keeps both scenarios but asserts nothing leaves the machine. All 102 telemetry, observability, relay, policy and update test files pass locally (693 tests).
• CodeQL: of 36 annotations, 30 are alerts already open on main in upstream code this large diff touches; this PR doesn't change them. Six are new:
  • Add placeholder testing notes to AGENTS.md pingdotgg/t3code#153 (SSH agent-signing regex): fixed in 23818e80c. The old regex was quadratic on repeated prefixes (27.8 s on an 820 KB line). The new check scans each line once, matches the old regex on 400,000 random inputs, and adds a regression test that timed out on the pinned source.
  • Scope git checkout effects and update Effect/Node dependency set pingdotgg/t3code#148, Guard worktree branch rename and fix branch toolbar active state pingdotgg/t3code#149, Virtualize branch picker list and forward ComboboxList ref pingdotgg/t3code#150, docs: add no-op dummy note to AGENTS.md pingdotgg/t3code#152: dismissed as false positives, with the evidence recorded on each alert.
  • Add dummy testing comment to AGENTS guide pingdotgg/t3code#151: dismissed as used in tests: it is in the eval harness, which no T2 artifact includes.

[danspark]

CI control change in 7d9f5951: Test failed three times on 23818e80c, each time at the 15 s browser-test limit in the PDF test upstream added in this sync (pingdotgg#25555). Upstream runs its web suite on 8-vCPU runners; ours are standard ubuntu-24.04. The same test passed in 12.0 s on 87a5d475, and the web sources were identical in all four runs. The web package now runs as its own step in the same Test job with --testTimeout=30000. Every test still runs and must pass. No application code changed.