- burn-rate-chart: pay the cold BurnRateChartInner import in beforeAll, so the
first test no longer times out on a loaded host
- typecheck:core includes src/shared/constants/iconMap.ts (TS2300 on duplicates)
- accent-ink guard checks pairs across separate ${...} expressions
- icon codemod keeps a directive first when code follows it on the same line
- vitest maps the kit's renamed exports (styles.css, tokens.css, brand) explicitly
- Playwright expect timeout is 30 s locally, as in CI
Refs #18079
## Summary
Add `POST /api/keys/import` so a management client can provision the same caller-supplied gateway key on multiple instances without invoking whole-database restore.
- Required `key`/`name`, optional UUID `id`; inherits every create-policy field and refinement through `createKeySchema.safeExtend`.
- Same management authentication as key creation, including `admin` access-token scope and the open-mode management-key credential requirement.
- One synchronous SQLite IMMEDIATE transaction checks identity and inserts the complete policy. Existing explicit id + same value is a read-only retry (200); conflicting id/value is 409 without another id or secret in the response; new import is 201.
- Existing crypto helpers supply encrypted-at-rest storage when configured, hash and prefix. Responses whitelist masked key + stored policy. Environment/self-loop credential collisions are rejected. New imports trigger best-effort asynchronous cloud sync.
- No rekey, settings/provider/combo/session import, or changes to `import-json` authorization. Existing creation remains server-generated.
Closes#18005
## Target branch and design
Thanks @xiaoyaner0201.
Resolve the hang-escalation conflict with #18960: keep the deterministic
mock-clock test, keep the SIGTERM handler at createClient, and use a single
handler so the term-ignored marker is written however early the SIGTERM lands.
fix(ws): bound /v1/ws frame input and keep buffering linear
Evidence: combined board on release/v3.8.54: 147 node:test and 11 vitest cases passed across the lot's test files; eslint, typecheck:core and open-sse typecheck clean; file-size OK; changelog integrity OK against origin/release/v3.8.54.
Keeps the logical-clock hang test and read-count hydration test from this branch
over the wall-clock bounds of #18440, and keeps #18440's early SIGTERM handler in
the hang fixture.
Adds an optional self-pinger for the Render free tier. It stays inert unless a URL is configured: no timer, no request, no log.
Maintainer rework: config and pinger moved to `scripts/build/runtime-env.mjs` (already shipped in the standalone bundle); `SELF_PING_PATH` can no longer replace the host; logs only on status changes; env vars documented in `.env.example` and ENVIRONMENT.md; 24 tests.
Evidence: removing the path guard fails 3 tests. Gate / CI, Gate / Quality and Merge integrity green on this head. Not exercised on a real Render service.
HTTP: on a keyless install (REQUIRE_API_KEY=false) any Authorization, x-api-key or x-goog-api-key header, valid or not, made client-API routes echo the request Origin while serving the request as anonymous. The allow path now relaxes CLIENT_API CORS only for a non-anonymous subject; route-level 401s that refuse a presented key still echo Origin (Vary: Origin, no credentials header) so browser clients can read them (#5242).
WebSocket: browsers do not apply CORS to WebSockets and the /v1/responses, /v1/ws and /api/v1/ws handshakes never read Origin. They now admit no-Origin clients, CORS-allowlisted origins, the gateway's own public origin and handshakes with a valid API key; everything else, null included, gets 403 ws_origin_forbidden. The bridge upgrades only when the route answers originChecked: true. /v1/responses now enforces REQUIRE_API_KEY. The service worker also bypasses the client-API aliases. CORS and PWA guide mirrors refreshed.
Co-authored-by: HouMinXi <19586012+HouMinXi@users.noreply.github.com>
The test:e2e script passed tests/e2e/*.spec.ts ahead of the caller's arguments, so
`npm run test:e2e -- <spec>` still ran every spec. The script now calls the runner with no
glob; the runner forwards spec paths and flags as `test` arguments and defers to the
config's testDir/testMatch when called bare.
The runner also spawned the runtime `playwright` 1.62.1 CLI while specs import
`@playwright/test` 1.63.0 ("did not expect test() to be called here"). It now spawns the
@playwright/test CLI (what `npx playwright` resolves to in CI), and @playwright/test is
pinned to 1.62.1, the exact runtime pin, so the lockfile carries one Playwright.
check-test-discovery anchors the e2e collector on the new script and playwright.config.ts;
gate-manifest follows the script. tests/unit/playwright-runner-args.test.ts guards the
argument building and the version pin (KEA-7886 follow-up, P19a).
Conflicts resolved mechanically: the release side only renamed github.com/diegosouzapw/OmniRoute links to cheaper-inference/OmniRoute (#17871) in QuotaSharePageClient, providers/page, privacy and terms; applied the rename on the migrated markup. src/shared/components/Footer.tsx stays deleted (dead shared Footer removed in F4, pinned by dashboard-kit-sweep).
Review follow-ups for KEA-7886 F4-A:
- M1: an unmapped <Icon> name draws SquareDashedBottom (exported as MISSING_ICON_GLYPH)
instead of CircleDashed, which is ICON_MAP.pending; a test pins that the placeholder
is neither an ICON_MAP value nor the same drawing as one.
- M2: restore the #3695 guard lost with material-icons-self-hosted.test.ts: neither
src/app/layout.tsx nor globals.css loads fonts from fonts.googleapis/gstatic.com.
- M3: icon-map-coverage and gen-icon-map point non-icon strings to NOT_ICON_NAMES
instead of a fake Lucide mapping.
- N6: resolveIcon resolves own keys only (constructor/toString return null), pinned.
- N1/N2: stale dashboard-layout-flag reference; single-file loop inlined.
The <Icon> adapter loses its Material ligature fallback: a name missing from ICON_MAP
renders a neutral CircleDashed glyph (data-icon-missing, same a11y) and warns once
outside production. The 32 remaining null entries of ICON_MAP are mapped to Lucide
(ICON_MAP is now Record<string, LucideIcon>). gen-icon-map no longer reads the
vocabulary from the material-symbols package: every scanned name must map to Lucide,
except a documented NOT_ICON_NAMES list (provider logo keys, a size key), and the
data-*-icon attributes are no longer mistaken for icon props.
The import is now inserted after the directive prologue (leading comments + "use client"/'use server' lines) instead of at offset 0, with a guard that throws if the output's directives differ from the input's. Leftover detection scans whole JSX opening tags (brace/string aware) instead of single lines, so multi-line dynamic spans are reported at their <span line and the CLI exits 2; multi-line spans with a static className (including {`...`} without ${}) are normalised and converted.
Release v3.8.52 — 1,442 documented changes (135 features, 1,112 fixes, 195 maintenance) from 1,386 cycle commits and 306 external contributors; full changelog in CHANGELOG.md → [3.8.52]. Local pre-flight on c754707d: check:release-green --quick --full-ci RELEASE-GREEN (0 HARD / 0 drift), webpack production build + build:cli OK; the release-PR ci.yml run was never dispatched (Actions FIFO backlog), merged with --admin by owner decision at STOP #1. ⚠️ base-red inherited: #15306 (i18n translation drift pinned by 3 unit tests + Docs Sync strict — translation backlog, not code).
Merged via /merge-batch (2026-10-16 wave, maintainer PRs). Validated on a second combined board of 8 PRs over release/v3.8.52@16d9205b8: typecheck:core, open-sse typecheck, eslint on changed files, file-size, changelog-integrity, complexity, cognitive-complexity, env-doc-sync and agent-skills-sync all green; focused tests 249/250 (node:test; the one red, kiro-content-stall-15260, passes 3/3 isolated — a 100 ms stall-timer flake under load) + 16/16 (vitest UI); boundary 774 = tip (inherited base-red). Remaining CI reds are the inherited base-red #15306.
Merged via /merge-batch (2026-10-16 wave, maintainer PRs). Validated on one combined board of 20 PRs over release/v3.8.52@76e905b15: typecheck:core, open-sse typecheck, eslint on changed files, file-size, changelog-integrity, complexity, cognitive-complexity, env-doc-sync and agent-skills-sync all green; focused tests 263/263 (node:test) + 26/26 (vitest UI). Remaining CI reds are the inherited base-red #15306.
Thanks @shipsfromrio.
Validated in local merge-train merge-train-20261016-020918 on devbox-vm-06-dev002 @ 86de1a0e. Fast gates green. Inherited typecheck:core and i18n drift match origin/release/v3.8.52 (#15306). Focused node tests and vitest passed.
Thanks @tiagolabsit.
Validated in local merge-train merge-train-20261016-020918 on devbox-vm-06-dev002 @ 86de1a0e. Fast gates green. Inherited typecheck:core and i18n drift match origin/release/v3.8.52 (#15306). Focused node tests and vitest passed.
Thanks @omnitrixAG — the standalone launcher applies the heap auto-calibration.
Validated in local merge-train .claude/worktrees/merge-train-20261015-235925 on devbox-vm-06-dev002 @ cb939401be036aa8caa1a9a972b0f13d85f5c914. Static gates green except inherited typecheck:core (same errors on origin/release/v3.8.52) and inherited i18n translation drift. Focused node tests passed. Vitest 499 passed; 1 timeout in provider-family-combos auto/glm (20s). Isolated rerun of that file passed 2/3, third was the same timeout, not an assertion. These PRs do not touch autoCombo. Inherited base-red #15306. Fast train: daily full unit suite still required.
Thanks @Junior-HJ — accepted connections now enable TCP keepalive. The change is in `src/shared/utils/runtimeTimeouts.ts` plus the dev server scripts, covered by the three main-server keepalive and timeout-parity tests.
Boarded together on release/v3.8.52 at 4833845bda. Focused unit tests: 133 passed, 0 failed (`node --import tsx/esm --test --test-force-exit`). File-size gate: only the inherited `open-sse/services/combo/executeTargetAttempt.ts` overage (1288 > frozen 1284), unchanged by this pull request.
⚠️ base-red inherited: #15306
* feat(dev): local candidate build/validate/promote loop with rollback
Add scripts/dev/candidate.mjs (npm run dev:candidate), the local side of
RFC #8084: build one candidate from the current tree (npm pack) or a
prebuilt tarball into an isolated npm prefix under _artifacts/candidate/<id>,
boot that packaged CLI on a free port with an isolated DATA_DIR and fake
secrets, require /api/health and /v1/models to answer 200, promote the same
validated artifact with compensated renames, and roll back automatically
when the promoted slot fails its health check. Every command supports
--dry-run and --json; child processes use argument arrays only (Hard Rule #13).
stopChild is now exported from check-pack-boot.mjs so the loop reuses the
same process-group shutdown instead of duplicating it.
* docs(ops): document the local candidate loop in the golden path and release checklist
* fix(dev): isolate the candidate boot from operator .env files
The first real validate run against the published 3.8.51 tarball booted and
answered /api/health, but /v1/models returned 401: the packaged CLI fills
undefined keys from <packageRoot>/.env (postinstall copies .env.example with
INITIAL_PASSWORD=CHANGEME), and any password is a credential surface for the
model catalog. It also reads the default data dir's .env, derived from
HOME/XDG_CONFIG_HOME/APPDATA, i.e. the operator's real ~/.omniroute/.env.
The candidate now gets an isolated HOME (plus XDG/APPDATA) under its slot,
INITIAL_PASSWORD pinned to an empty value, a fresh DATA_DIR on every
validation, and --log so failures show server output. run --dry-run no
longer claims the candidate is active.
Validated in a combined board with the other PRs of this wave on release/v3.8.52: the PR's own tests pass; typecheck:core, open-sse typecheck, api-typecheck, complexity, changelog-integrity, env-doc-sync, test-discovery, vitest-exclusions, error-helper, test-masking, lockfile, deps, gate-manifest, cli-i18n and i18n coverage green. Any file-size ceiling or stryker registration this wave needs lands in the wave's follow-up PR. Thank you for the contribution!
Onda 6. Este squash entra com o crédito de @QuangBlue (fix(api): standalone server owns SIGTERM so graceful-shutdown cleanup runs).
O board único validou os 19 PRs (sem o #15412): file-size OK, changelog OK, i18n drift OK, typecheck:core OK, complexity e cognitive dentro do baseline. Testes focados: 390 passaram, 1 skip. O teste novo de GHE Copilot Sonnet 5.5 falhou no head original e foi corrigido no board; esse ajuste entra com o #15478, fora deste squash.
O rebaseline de file-size e o adopt de i18n ficam num follow-up.
⚠️ base-red inherited: #15306
Obrigado, @QuangBlue.
Onda 6. Este squash entra com o crédito de @QuangBlue (fix(api): keep Next's upgrade listener off the run-next server).
O board único validou os 19 PRs (sem o #15412): file-size OK, changelog OK, i18n drift OK, typecheck:core OK, complexity e cognitive dentro do baseline. Testes focados: 390 passaram, 1 skip. O teste novo de GHE Copilot Sonnet 5.5 falhou no head original e foi corrigido no board; esse ajuste entra com o #15478, fora deste squash.
O rebaseline de file-size e o adopt de i18n ficam num follow-up.
⚠️ base-red inherited: #15306
Obrigado, @QuangBlue.
Onda 5. Este squash entra com o crédito de @grapeslush (fix(cli): prefer an existing ~/.omniroute over XDG_CONFIG_HOME in every data-dir resolver).
O board único validou os 28 PRs juntos: file-size OK, changelog OK, i18n drift OK, typecheck:core OK, complexity 2976/3218, cognitive 1358/1437. Testes focados: 545 passaram, 0 falharam, 3 skip em 52 arquivos. Cinco timeouts de resilience-http-e2e eram carga e passaram isolados.
O rebaseline de file-size e o adopt de i18n ficam num follow-up, fora deste squash.
⚠️ base-red inherited: #15306
Obrigado, @grapeslush.
Onda 3 (woodsonl) validada num único board em `.claude/worktrees/batch-w3` sobre `origin/release/v3.8.52` (`4f4c983347b9`), 27 PRs, superfícies protegidas 0.
Evidência: `typecheck:core` limpo. i18n drift limpo no board. changelog-integrity ok. file-size ok (170 congelados). complexity 2975/3218 e cognitive-complexity 1357/1437 dentro do baseline. node:test focado: 1868 passaram, 0 falharam. Vitest da UI: 150 passaram em 12 arquivos. Os 38 do RTK caem no `useCallback` nulo de `node_modules/node_modules/react` via zustand, o mesmo React duplicado desta worktree. Onde dois PRs mexiam no mesmo save, o board ficou com os dois comportamentos: escrita parcial de engine por id e a fila/resync do PR que entrou depois.
O hash de `.i18n-state.json` do guia de quality-gates entra no follow-up, fora do squash.
⚠️ base-red inherited: #15306 (timeouts de suíte no tip, não neste diff).
Obrigado, @woodsonl.
Validated in a local merge-train on a clean-install host (FAST gates green: static + changed tests + vitest on the combined tree; audit.test.ts verified passing on base and PR tree when the machine is idle).
⚠️ base-red inherited: #15306
Validated in a local merge-train on tomni-proxmox-113 (static gates + changed tests + vitest green on the 31-PR combined tree; the FULL unit run on the same set showed only load-timing flakes, the base-red consoleInterceptor-writes, and one regression since fixed in #15087).
⚠️ base-red inherited: #15306
Release v3.8.51 — 2,022 documented changes (235 features, 1,454 fixes, 333 maintenance) from 1,973 cycle commits and 301 external contributors; full changelog in CHANGELOG.md → [3.8.51]. Full ci.yml matrix green on 2f42a9ac (57 checks).
responses-ws-proxy.mjs (shipped as dist/responses-ws-proxy.mjs) sent raw exception text to the client in the failed-upgrade 500 body (CodeQL #1021/#1022) and in the upstream-connect response.failed frame, which could carry the upstream proxy URL with credentials. Both now return a fixed message; details stay server-side. Remaining CI reds reproduce on the pure release tip (#15032).
Maintainer rework: reconciled with #15038/#15040 on the tip — the wider forwarding-header set only sets the via-proxy marker from a peer that may be a proxy (hasProxyHopHeader && isTrustedProxyPeer). Loopback without headers stays local. 35 peer-stamp/route-guard/live-WS test files green. Board-validated with the rest of the batch on release/v3.8.51: typecheck:core, check:open-sse-typecheck, check-file-size, eslint on changed files, i18n:check-keys all green. Thank you @HouMinXi!
The /v1/ws and Responses WebSocket relays now report the real client address and drop client-written forwarding headers. 10 failures on the tip, green with the change; responses-ws suites green on the batch board. Thank you @HouMinXi!
Forwarding headers only count as a proxy signal from loopback/private/Cloudflare peers or OMNIROUTE_TRUSTED_PROXIES, and the IP filter judges a server-stamped client address. 16 failures on the tip, green with the change; ip-filter and peer-stamp suites green. Note for operators: a reverse proxy on a public address in front of an IP whitelist must now be listed in OMNIROUTE_TRUSTED_PROXIES. Thank you @HouMinXi!
Maintainer rework: merged the release/v3.8.51 tip (twice, the tip moved mid-rework) and resolved the file-size-baseline conflict (kept every annotation; stream.ts frozen at 3297 as the PR measured; chatCore.ts measures 6383 on the reconciled tip, so its ceiling stays at the tip's 6441 and the annotation says so). Boarded with the 2026-09-28 release-drain batch (typecheck:core, check:open-sse-typecheck, ESLint clean); on the reconciled head sse-output-signal 6/6, responses-ws-proxy-turn-timing 1/1, stream-readiness 22/22 and chatcore-streaming-ttft 2/2 (one timing flake in four runs on the loaded host). The only file-size red on the reconciled head is tests/unit/response-sanitizer.test.ts, which comes from the tip (#14813), not this PR. Thank you @QuangBlue!
Boarded on the release/v3.8.51 tip together with the 2026-09-28 release-drain batch (30 PRs): typecheck:core, check:open-sse-typecheck, file-size, migration numbering, i18n keys and ESLint on the changed files are clean, and this PR's own test files (1) pass on the combined tree. Thank you @rqzbeh!
Merge-batch 2026-09-23 (PRs do mantenedor, Trilha B, retrabalhada).
Um worker corrigiu o defeito apontado na revisão direto no branch da PR; o detalhe está na seção "Rework (merge-batch 2026-09-23)" do corpo, com a prova de red→green. Revalidado numa árvore combinada das PRs da 2ª leva da Trilha B (#14415#14472#14468#14493#14403) sobre o tip atual de `release/v3.8.51`:
- 104/104 testes node focados + 46/46 vitest (as 6 suítes de UI que saíram da quarentena); `check:provider-consistency` OK (359)
- `typecheck:core`: só o herdado `cliproxyAccountHealth.ts:157`; `check:open-sse-typecheck`: só o herdado `auggie.ts` (base-red #14547, corrigido pela #14511)
- eslint nos arquivos alterados limpo; file-size OK
* fix(dev): allow Ctrl+C to promptly kill dev server by closing active connections and adding force-exit timeout
* test(dev): add regression assertions for prompt dev server exit on Ctrl+C
* docs(changelog): add fragment for #13020
* feat(security): warn at boot when the inference server is exposed anonymously
`GET /v1/models` follows the dashboard login posture
(`isAuthRequired()` / `requireAuthForModels`) while the inference routes
follow `REQUIRE_API_KEY`. On an instance with an admin password set and
`REQUIRE_API_KEY=false`, `/v1/models` answers 401 while `/v1/responses`
is open to anyone who can reach the port — so the most natural probe an
operator runs reports the opposite of the truth.
#12568 added a boot warning for exactly this combination, but wired it
only into the API bridge and the live dashboard WebSocket. The Next
server that actually answers `/v1/chat/completions` and `/v1/responses`
never reached it, and it is the one that binds every interface by
default (`process.env.HOST || "0.0.0.0"`).
Wire the existing guard into the Next boot hook, and document the split.
Resolving the bound host needed care: two entrypoints bind that server
and they read different variables. `run-next.mjs` honours `HOST`; the
Docker entrypoint delegates to Next's generated `server.js`, which reads
`HOSTNAME`. `run-next.mjs` now publishes what it actually binds as
`OMNIROUTE_BOUND_HOST`, and the guard reads that, then `HOSTNAME`, then
the shared `0.0.0.0` default. `HOST` is deliberately absent from the
chain: the standalone server ignores it, so consulting it there would
warn about an interface the server is not on — and one false warning
teaches an operator to ignore the next one.
Closes#13695
* docs(changelog): add changelog.d entry for #13820