Update from upstream 2.80.0 while preserving fork integrations

[ar4ft]

Summary

Update the existing integration PR to upstream stable 2.80.0 (250f17afd) and merge the current fork main (7b8b5f59c, preview.4). Both tips are ancestors of this branch; merge commits retain attribution and the fork fixes added after the original PR. The fork package version stays 0.0.10-preview.4.

Preserved fork behavior:

• GitHub Copilot discovery advertises only chat models enabled in the authenticated subscription picker. Empty live results stay authoritative; configured seeds and combos cannot resurrect unavailable models. The Copilot API version header and credential destination restrictions remain.
• ocx sync / sync-cache refresh the proxy and enabled Grok/Claude integrations. Claude Desktop profile assignments and the Claude Code gateway cache refresh even when agent injection is disabled. Cold requests using the old claude-opus-4-8-20260330 Fable slot retain the Fable route instead of becoming Opus.
• Native Codex CLI, Codex Desktop and ChatGPT integration remains disabled, including old ON configs. Upstream remote connect/sync/disconnect catalog writes, hidden restart/filter helpers and native restart scopes are blocked. Startup does not start the native routing healer. Legacy proxies are refused before their old sync/shutdown hooks can run.
• Standalone POSIX installer, oxc / ocx aliases, checksum verification, atomic updates, and update-pre remain. scripts/install.sh, src/update/github-prerelease.ts and .github/workflows/tag-release.yml match fork main exactly.
• Only the fork tag-release workflow exists. Upstream CI, repository automation and workflow files are not restored, as requested. Documentation links describe upstream automation without claiming those files exist in this fork.

Upstream improvements retained include Claude ToolSearch references, expanded Desktop slots and non-date wire aliases with real Anthropic collision protection, Cursor Claude tool wording/redirects, Devin linear tool declarations/results, Jev effort handling, web-search pacing deadlines, service/lifecycle fixes, and the MCP SDK 1.31.0 security update. Claude first-party mode and sibling ownership are checked before gateway writes. Exact operator maps and explicitly Copilot-qualified dateless migration maps retain precedence. Broad non-Copilot fallbacks cannot claim unknown managed Desktop slots.

The broader verification found and fixed a Claude compatibility regression: cold recovery now applies to Copilot aliases without replacing an existing non-Copilot registry. Unknown managed slots retain upstream's unavailable response, and unrelated large provider catalogs do not consume the legacy Copilot date namespace. The endpoint tests cover the fork's disabled native enrichment while confirming that routed requests still work and never receive native credentials. All seven translated READMEs now carry the canonical fork notice and standalone installer instructions; their source hashes and structural parity are current.

Verification

Focused verification for b9263d17b06435b74b032c8fb5da478eb2c15d9c:

• 764 passed, 0 failed, across 41 runtime/regression files, including the complete Claude Messages endpoint tests, Copilot routing/discovery, native isolation, installer/prerelease handling, README parity, test layout, file-size ratchet, and core/Lab boundaries.
• 2,943 passed, 0 failed, across all 330 GUI test files with bun test --isolate --parallel=2 tests (127.24s).
• 6 passed, 0 failed in the compiled-binary native isolation regressions.
• Typecheck, privacy scan, structure check, generated skill check and diff whitespace checks: passed.
• Earlier GUI build/lint, documentation build (569 pages / 79,027 internal links), frozen installs and root/GUI high-severity dependency audits passed. Their inputs are unchanged by this verification follow-up; the GUI still matches upstream stable exactly.
• Installer, prerelease updater and tag-release workflow remain byte-identical to fork main. Both fork main and upstream stable remain ancestors of this head.

Broader runtime results and limitations

The full runtime suite is not green and this PR remains draft:

• bun run test at the previous head 293ec1873 hit the runner's 900-second limit, exit 124. The report listed 209 unstarted files and four interrupted files; no successful full-suite result is claimed.
• Continued the 209 unstarted files and three interrupted files through the isolated wrapper: 4,323 passed, 6 skipped, 223 failed, 4,552 tests / 212 files, exit 1 (297.83s). The interrupted legacy native hard-lock file remains incompletely verified.
• Ran all 20 serial-lane files separately through the wrapper: 654 passed, 23 skipped, 67 failed. Failures were in Codex shim/injection/toggle tests (65) and service tests (2). The same two service failures reproduce on unchanged upstream. Native Grok toggle (24 tests), release helper (40), live server (55), update stop-first (26), and the other passing lanes completed successfully.
• Many runtime failures expect native Codex writes, native credentials/claims, or upstream GitHub workflow files that this fork deliberately excludes. Native support and removed automation were not restored to satisfy those expectations.
• Compared 15 non-native/runtime failure files against unchanged upstream 250f17afd: 380 passed, 3 skipped, 24 failed, 1 error. Kiro model evidence, catalog-route and other failures also reproduce there. This is representative baseline evidence; not every remaining runtime failure has been fully classified.

The standard GUI command bun test tests failed with 2,767 passes / 176 failures. All observed failure names also reproduced in an unchanged-upstream run (2,766 passes / 177 failures). Isolating test files removed the cross-file React state/warning cascade. The first isolated run had one timing failure; its file passed separately on both branches, and the complete final isolated run passed all 2,943 tests.

React Doctor previously completed with exit 0 but reported 19 errors and 1 warning, all in unchanged upstream GUI test fixtures. That diagnostic result remains recorded; it is not described as clean.

No native macOS/Windows runtime or manual desktop/browser verification was performed in this Linux workspace. Only the release workflow is retained, so no PR CI result is claimed. The full suite has not been rerun from scratch after the focused fixes, and broader completion/readiness remains outstanding.

Exact focused regression command

bun test \
  tests/test-layout.test.ts \
  tests/test-layout-tooling.test.ts \
  tests/server/server-jev-combo-e2e.test.ts \
  tests/server/combo-requested-effort.test.ts \
  tests/claude-integration/claude-desktop-picker.test.ts \
  tests/claude-integration/claude-copilot-model.test.ts \
  tests/claude-integration/claude-desktop-policy.test.ts \
  tests/claude-integration/claude-desktop-picker-routes.test.ts \
  tests/claude-integration/claude-desktop-request-model.test.ts \
  tests/claude-integration/claude-inbound-tool-reference.test.ts \
  tests/claude-integration/claude-desktop-first-party-guards.test.ts \
  tests/claude-integration/claude-desktop-1m.test.ts \
  tests/claude-integration/messages-native-scope.test.ts \
  tests/claude-integration/claude-desktop-native-context.test.ts \
  tests/claude-integration/claude-desktop-discovery.test.ts \
  tests/claude-integration/claude-fork-copilot-wire.test.ts \
  tests/claude-integration/claude-desktop-first-party.test.ts \
  tests/claude-integration/claude-desktop-picker-profile.test.ts \
  tests/ci-workflows/file-size-ratchet.test.ts \
  tests/ci-workflows/tag-release-prerelease.test.ts \
  tests/ci-workflows/standalone-installer.test.ts \
  tests/codex-integration/github-copilot-model-discovery.test.ts \
  tests/codex-integration/fork-native-client-isolation.test.ts \
  tests/routing/jev-decision.test.ts \
  tests/cli/fork-live-policy.test.ts \
  tests/service/autostart-health.test.ts \
  tests/lab/core-lab-boundary.test.ts \
  tests/web-search/web-search-pacing-deadline.test.ts \
  tests/providers/devin-chat-wire-fixes.test.ts \
  tests/adapters/openai/openai-chat-qwen38-leading-system.test.ts \
  tests/providers/cursor/cursor-blob-claude-client-tools.test.ts \
  tests/providers/cursor/cursor-tool-definitions.test.ts \
  tests/providers/cursor/cursor-tool-wording.test.ts \
  tests/providers/github-copilot/github-copilot-oauth.test.ts \
  tests/providers/github-copilot/github-copilot-wire-defaults.test.ts \
  tests/providers/github-copilot/github-copilot-sse-rewrite.test.ts \
  tests/providers/github-copilot/github-copilot-account-origin.test.ts \
  tests/providers/github-copilot/github-copilot-context-tier.test.ts \
  tests/providers/github-copilot/github-copilot-stream-contract.test.ts \
  tests/claude-integration/claude-messages-endpoint.test.ts \
  tests/ci-workflows/docs-readme-translation-parity.test.ts

Additional validation commands:

bun run typecheck
bun run privacy:scan
bun run structure:check
bun run skill:surface:check
bun build --compile ./src/cli/index.ts --outfile /tmp/oxc-pr9-binary
OXC_TEST_BINARY=/tmp/oxc-pr9-binary bun test tests/codex-integration/fork-native-client-isolation.test.ts
(cd gui && bun test --isolate --parallel=2 tests)
# Earlier successful checks; these inputs are unchanged by this follow-up:
bun install --frozen-lockfile
bun audit --audit-level=high
(cd gui && bun install --frozen-lockfile && bun run build && bun run lint && bun audit --audit-level=high && bun run doctor)
(cd docs-site && bun install --frozen-lockfile && bun run build)
git diff --check
git merge-base --is-ancestor origin/main HEAD
git merge-base --is-ancestor upstream/main HEAD
git diff --exit-code origin/main -- scripts/install.sh src/update/github-prerelease.ts .github/workflows/tag-release.yml

The timed-out run's remaining-file list was taken directly from its report; it was executed as bun run test --parallel=2 <209 unstarted files plus three interrupted files>. Serial lanes were executed one file at a time as bun run test --parallel=1 <file>, using the 20 entries in SERIAL_FULL_SUITE_FILES from scripts/test.ts. Comparison runs used a separate detached worktree at upstream stable with the same Bun 1.4.0 and installed dependency versions. The unchanged build/audit commands and their results are recorded above.

Checklist

• Latest reviewed fork main and upstream stable included; native-client exclusion and release-only automation preserved.
• Integration regressions found during broader verification fixed and covered by passing focused tests.
• Complete isolated GUI suite passed; runtime failures and limits documented without claiming a full pass.
• Full runtime integration suite green / all remaining failures classified.
• Native platform/manual validation completed.
• Ready for review / merge.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-07T16:32:44.058659Z	b9263d1	Draft marked ready
🔒 Security Review	✅ Completed 2026-10-07T16:34:20.083712Z	b9263d1	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

opencodex/src/cli/dispatch.ts

Lines 994 to 996 in b9263d1

	if (["restore", "recover-history", "codex-shim", "codex-log-guard", "chatgpt", "connect", "disconnect", "catalog"].includes(resolveDispatchCommand(command) ?? command)) {
	console.error("Native Codex and ChatGPT client integration is disabled in this fork.");
	return 2;

Preserve the documented remote-hub commands

This blanket guard makes every ocx connect invocation—including connect status and --clients claude—and every ocx disconnect exit with status 2 before their handlers run. However, this same commit adds docs-site/src/content/docs/guides/remote-hub.md, which instructs users to enroll, inspect, rotate, and disconnect remote clients with those commands, so the newly documented remote-hub workflow is unusable in this fork. Either remove that advertised surface or narrow the guard to operations that actually mutate native Codex state.

AGENTS.md reference: src/AGENTS.md:L28-L29

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".