Skip to content

chore(sync): merge upstream dev through cbe71595e1 - #494

Merged
shuv1337 merged 44 commits into
integration-v2from
upstream-sync
Nov 14, 2026
Merged

shuv1337 merged 44 commits into
integration-v2from
upstream-sync

Conversation

@shuv1337

@shuv1337 shuv1337 commented Nov 14, 2026 •

Copy link
Copy Markdown
Owner

Summary

Normal two-parent merge of upstream dev through cbe71595e1, following #484, #487 and #489. Sync-only: no version bump, no release, no publish. Targets only shuv1337/shuvcode:integration-v2. No default-branch push, rebase, force-push, tag, release, publish, deployment or live-service change. Stop before merge.

  • 40 upstream commits past 73abe63ac2 (the task reported 38; the fetched drift is 40). Upstream introduced no new version in this window, so root manifests stay 2.3.0 and the CLI manifest/lockfile stay 2.3.0-shuv.1.
  • Include: timed-out shell commands moving to the background by default (upstream graduated and deleted shell_timeout_background), read-tool default limits and paged-read fixes, six-byte GIF detection, the media MIME shortcut, connect-dialog sign-in polish, submenu/tooltip fixes, image jobs addressed by provider ID, the xAI video entrypoint, OPENCODE_SESSION_ID told to the model, and the stories/components-lab expansions.
  • Preserve Shuvcode identity, npm-only updates and the publish guard, bundled Claude/Antigravity/Quota/UI integrations, durable session/execution rules, and all three Claude Agent SDK hiding layers.
  • chore(ci): run the full suite on sync and release PRs #490's sync/release PR gate, test(core): fix the stopped-command race and report stalled PowerShell calls #491's stopped-command race fix plus stalled-PowerShell reporting, and fix(ci): authenticate Bun release downloads in the CLI build #492's GH_TOKEN build fix are untouched by this upstream range and still present.

Ancestry

Durable audit, conflict resolutions and complete branch dispositions.

Conflict resolutions

Seven paths conflicted:

Path Resolution
CHANGELOG.md Keep the fork 2.3.0-shuv.1 note and all earlier fork notes; keep one upstream ## v2.3.0 section, verified byte-identical to the tag copy already in the fork. No duplicate, no new fork entry for a sync that does not release.
bun.lock Keep shuvcode / 2.3.0-shuv.1; take upstream's @hona/strobe 0.39.0 hashes and the new @kobalte/core patch. Frozen install validates it.
packages/cli/package.json Keep fork name/bin/version; adopt the rest of upstream's manifest.
packages/app/src/runtime/i18n/en.ts Keep the fork's removal of provider.connect.claudeAgent.* (hidden provider); adopt upstream's ellipsis-free provider.connect.command.waiting; the unused fork-only title.anthropicProMax key goes with upstream's removal.
packages/app/src/providers/connect/dialog.tsx Adopt upstream's hoisted STEP_CLASS/methodLabel/METHOD_ICON; keep the fork deletion of the Claude Agent setup UI and its narrow("claude") match.
packages/app/e2e/regression/settings.spec.ts Keep the fork deletion of the Claude Agent provider case; upstream's rewrite still drives a provider the fork hides.
packages/tui/test/app-lifecycle.test.tsx Keep both fork update/quit "Run shuvcode to use …" tests; adopt upstream's ttft removal above them.

Every other both-sides-changed file auto-merged with the fork delta intact, including the isHiddenProvider filters in the TUI session route and the connect controller, the ToolFailure re-export in @opencode/ai, the tab-popover e2e assertion and the Go-upsell art width. #491's delta in packages/core/test/tool-shell.test.ts (147 changed lines) is byte-identical after the merge.

Managed policy and fork preservation

Keep #484's upstream behaviour: model.list, model.default, provider.list, generate.*, voice.* and websearch.query wait up to 5 seconds, then return 503 if activation is incomplete; other catalog reads stay non-waiting. No extra waits added.

Global.app = "shuvcode", XDG roots, ports 0x1337/0x1338, client Service.fallback() and the ["shuvcode", "serve", "--service"] spawn, npm-only updates, the shuv1337/shuvcode publish guard and the bundled plugins are unchanged. The Claude Agent SDK tree stays on disk, unregistered, still filtered by ProviderPlugins, the config provider plugin and every list that calls isHiddenProvider.

Validation

Bun 1.4.4, package suites run from their package directories with an isolated environment (env -i, private HOME/TMPDIR, no inherited OPENCODE_*): this worktree's shell carries the live OPENCODE_CONFIG_DIR, which changes the SDK suite's result. No privileged/systemd container and no host /dev, /sys, /run/udev or /run/dbus mounts.

Gate Result
Root bun run check (lint + 40-package typecheck) Pass, 40/40 tasks, lint exit 0
bun run generate (packages/client) Pass, no generated drift
bun install --frozen-lockfile Pass, 1597 packages, no lockfile change
git diff --check Pass
Forced Turbo unit gate, concurrency 3 31 of 32 reporting suites green; SDK failed only under the inherited OPENCODE_* env; Turbo hit known SIGSEGV pc_4bef4e2422a7 before core/TUI/CLI finished, so those ran individually
Core 7,633 pass / 16 skip / 11 fail across 333 files
CLI 477 pass / 7 skip / 0 fail across 69 files
TUI 1,753 pass / 2 skip / 0 fail across 211 files
Server 139 pass / 1 skip / 0 fail
SDK 40 pass / 0 fail isolated
App unit + browser 2,718 / 360 pass, 0 fail
ai / codemode / gui-plugins / ui / session-ui / merman / client / site / util / latex / plugins 2,241 / 1,641 / 1,130 / 626 / 602 / 508 / 274 / 242 / 146 / 123 / 205 pass, 0 fail

The 11 core failures are all GitLabPlugin OAuth cases that need the fixed local callback port: GitLab login needs local port 8080, but it is already in use. Something listens on 127.0.0.1:8080 on this dev host; the same file fails identically (8 pass / 11 fail) on unmodified origin/integration-v2, and a GitHub runner has the port free. The SDK workerd-seeded case behaves the same way under the inherited environment and passes isolated. No existing test budget, retry or skip was weakened.

The initial merge changes 343 files (the audit and review repairs add more paths), above the 300-file cap on GitHub's diff endpoint: gh pr diff and Plannotator --no-local will omit hunks. Review with git fetch origin pull/<n>/head && git diff origin/integration-v2...FETCH_HEAD, or git show --remerge-diff 6782796613 for the conflict resolutions.

No full browser E2E/component suite, cross-platform execution or all-target build is claimed. Focused Chromium component regressions were run as described below. packages/app/e2e/regression/settings.spec.ts changed only by keeping the fork's deletion, so the e2e job should be unaffected; CI's unit and e2e jobs run on this PR because its title starts with chore(sync).

Independent review and repairs

Astra identified two upstream regressions at the initial head; both are repaired in eb4e543f3a and undergoing a fresh independent Astra re-check:

  • Preview find now matches escaped case-insensitive Unicode patterns on original DOM text, avoiding shifted ranges/throws after case expansion in both rendered documents and sandboxed HTML.
  • The new timeline again shows the existing localized nonzero-exit, timeout and stopped badges for user-run commands, even without output.

The two preview cases and three shell-outcome cases failed before the production fixes. Afterward, preview tests passed 20/20 attempts and the entire shell component spec passed 120/120 attempts at two workers, without retries or dependency-optimizer reloads. Root bun run check passed 40/40; whitespace passed. Evidence and limitations are recorded in the durable audit. A strobe comparison bounded out during setup; no performance verdict claimed. Required replacement-head CI and independent approval remain gates.

Visual

<svg xmlns="http://www.w3.org/2000/svg" width="240" height="72" viewBox="0 0 240 72">
  <style>@keyframes focus{0%,100%{opacity:.2}50%{opacity:.7}}@media(prefers-reduced-motion:no-preference){.hit{animation:focus 3s ease-in-out infinite}}</style>
  <g fill="none" stroke="currentColor" opacity=".6"><rect x="4" y="4" width="232" height="64" rx="6"/><path d="M120 10v52"/></g>
  <rect class="hit" x="56" y="22" width="14" height="24" rx="2" fill="currentColor" opacity=".2"/>
  <g fill="currentColor" font-family="monospace" font-size="16"><text x="22" y="40">İ x</text><text x="135" y="30" font-size="12">$ exit 1</text><text x="135" y="51" font-size="12">Exit 1</text></g>
</svg>

Release

Not in this PR. Recommendation: if the hosts should get these 40 commits, a separate release-prep PR bumps to 2.3.0-shuv.2 on channel latest — the suffix sorts above 2.3.0-shuv.1 and below the next upstream base, and upstream's base is still 2.3.0. Release prep stays out so this merge is the only review surface and host upgrades can wait out the post-publish npm tarball 404 window.

Feature flags

No new fork flag. Upstream graduated and deleted shell_timeout_background (timed-out shell commands move to the background by default); the fork adopts the upstream default and keeps the flag's history only in past release notes.

rekram1-node and others added 30 commits November 13, 2026 02:12
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…ay in (anomalyco#63739)

Co-authored-by: Filip Hejmowski <fhejmowski@simplito.com>
@socket-security

socket-security Bot commented Nov 14, 2026 •

Copy link
Copy Markdown

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff Package Supply Chain
Security
Vulnerability Quality Maintenance License
Updatednpm/​@​hona/​strobe@​0.34.0 ⏵ 0.39.010010099 +196 +1100

View full report

The New timeline boundary test reported timeline/terminal-work.stories.tsx
as an unauthorized importer. Restore the old story's pre-repair route and own
the shell outcome story in timeline-next/stories, which the three component
cases mount. The localized outcome badge is unchanged.
@shuv1337
shuv1337 merged commit 617bc15 into integration-v2 Nov 14, 2026
7 checks passed
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.

8 participants