Validated on the combined 8-PR board + the branch itself: vertex-anthropic-models 4/4 (new — pushed to your branch as a fix-in-place commit per Hard Rule #8, covering the parser's global/project-scoped resource names, malformed-input handling, and the claude-* → targetFormat heuristic on vertex/vertex-partner), 88/88 across the board's focused suites, typecheck:core clean, gates within baseline. Anthropic partner models on Vertex AI now discover dynamically via the Model Garden publisher endpoint and route through the Claude translator even for models outside the static registry. Thank you @maci0!
Validated on the combined 8-PR board: ollama-local-capabilities-routing 3/3, managed-model-import 9/9 (including the integration with the carried Gemini-3.5-Flash cleanup from #11259), 88/88 across the board's focused suites, typecheck:core + dashboard-typecheck clean, gates within baseline. This brings #11088 to the release line — it had squash-merged to main by base error (mine) — AND fixes the two defects the port caught: the global filter drop that leaked image/video models into OpenAI chat selections (now scoped to self-hosted providers) and the unregistered hard-lease credential site. Exemplary port discipline: byte-identical carry + the corrections in a separate reviewable commit + the superpowers docs deliberately left out. main still needs the same two-line fix. Thank you @yourspraveen!
Validated on the combined 8-PR board: the declared-vocabulary clamp integrates with #11232's learned-set clamp (both fire in sequence; REASONING_EFFORT_ORDER import added), reasoning quartet 57/57 including the new opencode-go suite 7/7, 88/88 across the board's focused suites, typecheck:core + dashboard-typecheck clean, all static gates within baseline. Retargeted main→release/v3.8.50 and rebased onto the release tip (authorship preserved; the branch carried main history, so only the real commit was carried). The generic explicit-capability clamp closes the Console Go 400 [1210] loop for models with a declared effort vocabulary. Thank you @linhdmn — the live-probe table in the PR body is exactly the evidence standard we want.
Validated on the resolved merge against the current tip (527da656 + the post-#11281 rebaseline): the single conflict was a comment-only collision in providers/[id]/models/route.ts (kept the tip's #10828-ordering note). Focused suites 125/125 across all 13 touched test files (build-sqlite-stub, cc-compatible, copilot-claude-messages, copilot-gemini-route, executor-github, ghe-copilot, github-copilot-discovery-token, github-copilot-model-discovery, noauth-sibling-7620, provider-header-profiles, provider-models-config, request-log-payloads, upstream-error-passthrough), typecheck:core clean, file-size/changelog-integrity OK. Merged --admin over the inherited 2026-08-23 base-red cluster (#9985) — the reds are proven tip failures (CLI catalog cluster + @testing-library allowlist, being drained by #11280), not from this diff. Note: the rebase means several items the body listed (relay x-relay-path SSRF, /v1/search blocked-providers, #10736 rotation fence, #10903, #10865, #10899, #10916) already landed upstream and are NOT in this delta — the delta is: better-sqlite3 build guard + build heap/worker caps + telemetry-off (#10060 re-derived), credential-echo passthrough refusal + OCR/moderation redaction + call-log key redaction, Copilot CLI 1.0.81-6 wire identity + Claude→/v1/messages name-matched routing + discovery token fix, CC model_not_found 400, compat overrides for no-auth aliases (#7620-pinned). The Copilot wire-identity change is the one to watch in production. Thank you @arminanton — and the ported-author credits in the commit history (@rqzbeh, yidecode, the #10899/#10916 authors) are preserved. Your config-posture finding (REQUIRE_API_KEY default vs 0.0.0.0) is noted for a maintainer decision, as you scoped it.
TDD: 4 red-to-green tests reproducing the two review findings + 6 new cases; 30/30 in tests/unit/codex-app-server.test.ts; typecheck/eslint/env-doc-sync/file-size clean. Merged --admin over the inherited 2026-08-23 base-red cluster (#9985): the remaining reds (CLI catalog/registry tests, @testing-library allowlist) are proven tip failures unrelated to this diff — shard logs show only CLI-cluster failures, and this PR itself drains the mutation-test-coverage red (stryker registration for #11267's test). Behavior change for app-server deployments is documented in the PR body (sandbox default + binding refusal).
Validated on the combined 12-PR batch board + the resolved merge against the post-#11232 tip: focused suites 73/73 (learned-reasoning-effort-caps, synced-capabilities-learned-effort-override, synced-effort-suffix-learned-validation, effort-tiers-loop-catalog-e2e, reasoning-effort-clamp-and-retry, reasoning-effort-learned-capability) + opencode-plugin effort-tier-variants 4/4, typecheck:core clean, gates within baseline. The stacked-branch conflict after #11232 squash-landed was resolved by hand (the learned-caps module keeps both the Set API and the new model-scoped lookup). The effort_tiers loop is closed end-to-end: catalog advertises exactly what the upstream accepts, and -<tier> suffix variants resolve against the learned set. Thank you @maxmad64bis!
The opencode-go quota preflight ignored the dashboard quota snapshots, so
priority combos kept selecting connections whose weekly window was already
drained. Two gaps, two fixes:
A) fetchOpencodeQuota only consulted the live upstream endpoint, which has
no public quota API (404 — the module JSDoc already admits this). Every
preflight therefore evaluated null and proceeded (fail-open) even with a
connection at 0% weekly remaining in plain sight on the dashboard. The
fetcher now synthesizes its triple-window QuotaInfo from the cached
dashboard snapshots when the live endpoint yields nothing, mapping
session→window_5h, weekly→window_weekly, mcp_monthly→window_monthly and
mirroring getQuotaWindowStatus semantics (expired resetAt = window rolled
over = must not count as exhausted; fractionReported=false = unknown,
never exhaustion). Read-only via src/domain/quotaCache.ts accessors —
never SQL, never a re-scrape on the hot path — and gated on the
connection actually having dashboard scrape config, so unconfigured
connections never touch the snapshot store. Fail-open is preserved:
no snapshots → null, exactly as before. The quotaCache import is dynamic
because a static edge would close an initialization cycle
(fetcher → quotaCache → usage.ts → usage/opencode.ts → fetcher).
B) The sibling-selection latency gate in getProviderCredentialsWithQuotaPreflight
never consulted resilience.quotaPreflight.enabled
(QUOTA_PREFLIGHT_CUTOFF_ENABLED) — that flag only armed the auto-strategy
candidate builder and the per-target cutoff for pinned connections, so a
priority combo over sibling opencode-go connections (connectionId null at
combo level) skipped preflight entirely. The flag now arms the gate as
well; the default (flag off) is unchanged.
TDD (Hard Rule #18), tests/unit/quota-exhaustion-cutoff-opencode.test.ts:
- fetcher 404 + seeded snapshots weekly=0%/session=80% → cutoff blocks
(RED before, GREEN after); also asserts the bridge is read-only (single
upstream fetch, no re-scrape).
- weekly 0% with next_reset_at in the past → not blocked (window dropped).
- per-window threshold override resolves against the mapped window_weekly
key (50% override blocks at 40% remaining; factory 2% does not).
- fail-open guard: configured dashboard with no snapshots still returns null.
- selector level: flag on, two opencode-go sisters, priority-1 exhausted →
selection skips to the healthy one (RED before, GREEN after).
Sibling suites green: opencode-quota-fetcher (18), quota-preflight,
combo-priority-quota-exhaustion-cutoff-5923, issue-6686, 8431, throttle-6911,
sse-auth*, quota fetchers, combo strategies, snapshot/hydration tests
(~500 tests). eslint (with suppressions) and typecheck:core clean.
Closes#11234
Co-authored-by: Xiangzhe <bakryun0718@proton.me>
Two defects exposed by #11226 (the 401 "User not found." itself is upstream
OpenRouter's response to a bad key — not an OmniRoute bug):
1. OpenRouter key validation was vacuous. Both the dashboard Check (via
validateProviderApiKey -> validateOpenAILikeProvider) and 'omniroute
providers test' (bin/cli/provider-test.mjs) probed /api/v1/models, which is
PUBLIC and answers 200 to any key — so invalid keys were saved/marked valid
and only failed on real chat traffic. OpenRouter's authenticated key-info
endpoint (GET /api/v1/auth/key, 200 = valid / 401 = invalid) is now the
probe: registered as testKeyModelsUrl on the openrouter registry entry
(same mechanism as perplexity) and as keyCheckPath in the CLI test configs.
No other provider's probe changes; error results keep using canned strings,
never the raw upstream body (Hard Rule #12).
2. 'omniroute auth export' crashed with "cmd.optsWithGlobals is not a
function". .command("auth export") does not register a two-word command:
commander parses the bare word 'export' as a required positional argument,
so the action received (exportArgValue, options, command) while expecting
(options, command). Registered 'export' as a proper nested subcommand of
'auth' — the documented CLI surface 'omniroute auth export [--force]
[--id] [--format] [--out]' is unchanged, and unknown positionals
(e.g. 'omniroute auth bogus') are now rejected instead of silently running
the export.
TDD: tests/unit/openrouter-key-validation-auth-endpoint.test.ts (stub mimics
the real OpenRouter: /models public-200, /auth/key 401 'User not found.') and
tests/unit/cli-auth-export-wiring.test.ts (real commander wiring via
createProgram) were RED before the fix and are GREEN after.
Co-authored-by: Xiangzhe <bakryun0718@proton.me>
Validated on the combined 12-PR batch board: combo-max-global-attempts-config 4/4 plus the neighboring combo suites green, typecheck:core clean, gates within baseline. clampGlobalAttempts mirrors clampComboDepth with a 200 hard cap, so a bad config can never disable the budget. Closes the configurability item from #11134. Thank you @pacocartones!
The dashboard stores LM Studio connections under the hyphenated provider id
"lm-studio", but the embedding registry keys the provider as "lmstudio"
with no alias. As a result, "lm-studio/<model>" embedding requests failed
with a 400 unknown-provider error, and "lmstudio/<model>" requests always
hit the hardcoded http://localhost:1234/v1/embeddings endpoint, ignoring the
baseUrl of the configured connection.
Mirror the ollama-local pattern from #2824/#9225:
- embeddingRegistry: add "lm-studio" -> "lmstudio" to
EMBEDDING_PROVIDER_ALIASES (registry key unchanged so existing
"lmstudio/<model>" clients keep working).
- embeddings service: extend the optional keyless-connection hydration to
lmstudio; getProviderCredentials("lmstudio") already resolves the
"lm-studio" connection via the provider search pool/alias, and a
selection/rate-limit failure still proceeds without credentials.
- embeddings handler: apply the same baseUrl override + normalization
(strip trailing slashes and /v1, /v1/chat/completions, /v1/embeddings
suffixes, then rebuild <host>/v1/embeddings) to lmstudio, keeping the
static localhost fallback when no connection or empty baseUrl.
TDD: tests/unit/lmstudio-connection-baseurl-11233.test.ts failed on the
alias, override and service-hydration asserts before the fix and passes
after; ollama-local (#2824) and lmstudio registry (#7601) sibling tests
remain green.
Co-authored-by: Xiangzhe <bakryun0718@proton.me>
Validated on the combined 12-PR batch board: focused node:test suites green (incl. provider-rate-limit-overrides-schema + ratelimit-admission-control-6593 isolated 11/11 + the new EditConnectionModal vitest 3/3), typecheck:core clean, all static gates within baseline. Pushed one follow-up commit to your branch with the real Vietnamese translations for the two new keys (repo convention: vi never ships __MISSING__ placeholders). Per-connection maxWaitMs override lands, folding the zai-web exception into the general mechanism. Thank you @maxmad64bis!
Validated on the combined 12-PR batch board (worktree off origin/release/v3.8.50 @ 8f390eff): focused node:test suites 183/184 (the single red was a load-induced pollUntil flake — 11/11 when re-run isolated, this file included), typecheck:core clean, file-size/changelog/complexity/cognitive gates all within baseline, mutation-coverage gate no-drift. Learned effort sets now cost at most one 400 per provider+model. Thank you @maxmad64bis!
Merged after conflict resolution: the 5 conflicting test files were the base-red drains that #11201 already landed on the tip — kept the tip versions; the feature content is untouched. Validated on the combined batch board + this branch: codex-app-server + codex-gpt56-catalog 25/25, typecheck:core clean, docs-counts green (351 providers), provider-consistency 268/351/0. The opt-in codex-app-server transport (JSON-RPC-over-WS, turn/completed-awaited close, Responses SSE bridge) leaves the default codex path untouched. Thank you @arminanton — a 3.4k-line transport with the docs wave and tests to match!
Merged after conflict resolution onto the post-#10961 tip (verified no clobber of today's plaintext-wins work — the resolver functions were appended and the chatCore call-site swapped; a first --theirs attempt was caught reverting #10961 and redone hunk-by-hunk). resolveIncompatibleReasoningAction now defaults single-target incompatible reasoning to drop while combos keep their explicit strategy, with the x-omniroute-reasoning-fallback header override. 21/21 reasoning suites green, typecheck clean. Fixes#10959. Thank you @adevwithpurpose!
Validated on the combined batch board: gates clean, typecheck clean, focused tests green.
Exactly one Market id rewritten at the createTask seam (google-imagen/nano-banana-2 → nano-banana-2); every other namespaced id forwards byte-identical — the contract is stated and tested precisely. Fixes#11225. Thank you @xiaoyaner0201!
Validated on the combined batch board over tip 0b41259f: static gates clean (changelog, file-size 160 frozen, complexity 2628<=2774, cognitive 1187<=1223, dead-code 408<=416, docs-counts green at 351 providers, provider-consistency 268/351/0), typecheck:core clean, 430+ focused tests green across 5 groups.
muse-spark's all-reasoning empty-answer payloads (41 failures in 2h captured live) now floor the output budget so the model can actually emit content — no more empty-content 502 churn. Thank you @linhdmn!
Validated on the combined batch board over tip 0b41259f: static gates clean (changelog, file-size 160 frozen, complexity 2628<=2774, cognitive 1187<=1223, dead-code 408<=416, docs-counts green at 351 providers, provider-consistency 268/351/0), typecheck:core clean, 430+ focused tests green across 5 groups.
GLM-family upstreams reject message arrays with no user turn (400 [1214], verified live 2026-08-23); the synthetic user turn keeps Claude Code sessions on opencode-go alive. Thank you @linhdmn!
Validated on the combined batch board over tip 0b41259f: static gates clean (changelog, file-size 160 frozen, complexity 2628<=2774, cognitive 1187<=1223, dead-code 408<=416, docs-counts green at 351 providers, provider-consistency 268/351/0), typecheck:core clean, 430+ focused tests green across 5 groups.
max_context_window now wins over the pricing-tier context_window in Codex catalog parsing — context-aware fallback stops demoting codex behind smaller targets. Live evidence (390K served past 272K) plus the live-shape regression test. Thank you @excessivechaos!
Validated on the combined batch board over tip c92bd40b: static gates clean (changelog, file-size 158 frozen, complexity 2626<=2774, cognitive 1183<=1223, dead-code 409<=416), typecheck:core clean, 70 focused tests green (PR suites 49/49 + auth/combo neighbors 21/21).
Keyless custom-compatible connections stay in auto/* pools — the credential filter now recognizes a registry-free keyless endpoint instead of dropping the connection before pool construction. Fixes#11180. Thank you @pacocartones!
Validated on the combined batch board over tip c92bd40b: static gates clean (changelog, file-size 158 frozen, complexity 2626<=2774, cognitive 1183<=1223, dead-code 409<=416), typecheck:core clean, 70 focused tests green (PR suites 49/49 + auth/combo neighbors 21/21).
Operator-configured positive minTime floor now survives the plenty-of-headroom relaxation (resolveMinTime instead of a hard 0). Fixes#9763. Thank you @pacocartones!
Validated on the combined batch board + this branch: perf-a-b-c-d + account-fallback-service 93/93, gates + typecheck clean. Owner approved the full bundle including item A (async proxy-log batching, 1s/100-item flush with an unref'd timer — reviewed the implementation: flush helper exists for shutdown wiring, buffered logs are the accepted tradeoff). B (lazy modals), C (O(1) alias maps), D (pre-compiled regex — this is also the entire content of #11187, being closed as subsumed) ride along. Conflict with the tip was only stale provider-count docs. Thank you @rqzbeh!
Validated on the combined batch board + this branch: antigravity-dynamic-session-id + proxy-fetch-dns-retry green; file-size gate green with the proxyFetch 1244 frozen entry (dated annotation for the +5 retry-classification lines, owner-authorized). Static per-account sessionId unpinning ends the concurrent-turn 429s and EmptyStreamError drops on the Hermes→Antigravity path; EAI_AGAIN/ENOTFOUND/ETIMEDOUT now classified retryable. Conflict with the tip was only stale provider-count docs. Resolves the remaining #10443 root causes. Thank you @rqzbeh!
Merged after conflict triage: the six base-red repair files (vi.json, opencode.ts JSDoc, context-manager test, the three webhook dispatcher tests, the uncloseai orphan-test rename) were already drained on the tip by today's #11130/#11157/#11160/#11113 — those hunks resolved to the tip shape. What lands is the production-fix set: GLM transport-aware Anthropic headers, Claude Code-compatible model-listing rejection, combo live-test single-probe, zero-cost Auto-Combo interval normalization, recovery-clearing union handling, LLMLingua real-path compare, macOS netstat PID discovery, AI Horde R2 strict public-host validation. Sweep of every touched test file: 243/243 green; typecheck + file-size clean. (guide-settings-route's 4 reds reproduce on the pure tip — pre-existing drift from #11079, not from here.) Thank you @backryun!
Merged after count reconciliation: the branch's regenerated docs claimed 57 free forever / 157 migrations from its older base; gate-verified values on the current tip are 56 free forever (Logfare carries a Free badge via gateways.ts freeNote but has no freeModelCatalog per-model entries, so the live-code counter stays at 56) and 159 migrations — the README/SVGs now match the check:docs-counts output exactly. provider-consistency OK at 267 registry / 349 canonical. logfare-registry 4/4, icon + KNOWN_PNGS + discovery-set membership all verified present. Closes#10644. Thank you @jonlwheat2-gif!
Validated on the combined batch board + this branch: 231/231 across chatcore-translation-paths, reasoning-cache, strip-reasoning-blobs, and both Responses translator suites. Pre-merge: propagated the #11110/#11129 summary:[] defaults into five assertions here (each commented with its PR) — without it this branch red against the tip, and as a bonus the merge drains the 4 reasoning reds that were live on the tip from those merges. Plaintext now wins over a coexisting opaque companion; opaque-only drops cleanly for plaintext targets; combos keep explicit Skip. Fixes#10949 and #10959. Thank you @jackjinke!
Merged after conflict resolution in modelMetadataRegistry.ts: the tip's effortTiers chain (declared efforts → declared tiers → undefined-if-thinking-declared → codex extension) now carries this PR's GLM guard as the final-fallback override — GLM-family models without a provider-declared contract get the authoritative empty tier list instead of generic OpenAI tiers. GLM/ZCode suites 40/40 on the resolved branch. Closes#10962. Thank you @xz-dev!
Validated on the combined batch board + this branch: ccr-non-mcp-full-prompt-loss + ccr-retrieval-ramp 20/20, file-size gate green with the ccr/index listing (1024, dated annotation — owner-authorized). The callerSupportsCcrRetrieve gate now skips the whole engine for callers whose tools[] cannot reach omniroute_ccr_retrieve — no more 15KB prompt arriving upstream as 112 tokens. Production-measured root cause, textbook TDD. Thank you @HouMinXi!
Compaction-V2 output now counts as real model output (no synthetic response.failed after response.completed), the Codex SSE filter handles CRLF framing, and terminal detection runs before scan-state bounding. 88/88 stream/readiness suites on the board. Thank you @jackjinke!
SSE comment lines (OpenRouter keep-alives) and leading whitespace no longer fail the combo quality gate's JSON fallback. First contribution — clean minimal fix with test. Thank you @asorourx, welcome aboard!
Merged after sibling #11151 landed: streamRecovery.ts auto-merged byte-identical to the validated combined board; the test-file conflict (both PRs added suites at the same anchor) resolved keeping all 11 tests — #11151's four clean-stop cases plus this PR's three threshold cases, with the PR's updated partial-tail fixture for the pre-existing overlap test. Full chain green: 32/32 (wiring + continuation + toolcall regression). The documented 8-char overlap threshold ends the silent mid-word gluing. Thank you @maxmad64bis!
Merged after conflict resolution against the tip's #11109 (per-call tool_call tracking): scanOpenAiSseText keeps the per-call finish_reason special-case AND gains reasoningText + literal finishReason; canContinue uses the in-flight predicate with the new reasoning-only-clean-stop escape. One integration fix on the branch: the PR's hallucinatedEmptyStop referenced emittedToolCall, which #11109 had renamed — the branch now tracks emittedSawToolCall at the emitted level (any tool_call delta, complete or not), preserving the PR's don't-recover-after-tool-calls intent. Chain suites green: stream-continuation-wiring + stream-continuation + stream-recovery-toolcall 29/29. Thank you @maxmad64bis!
Validated on the combined batch board: purify-system-first suite 4/4, typecheck clean. Pre-merge: file-size baseline gained a frozen entry for contextManager.ts at 1001 (+1, this PR's merge-into-leading-system branch) with a dated annotation — the gate caps unlisted files at 1000. Producer side of the live-confirmed TokenRouter 400 class: no internal path emits a mid-array system message anymore. Thank you @ggdayup — the call-log evidence made this airtight!
Validated on the combined batch board over tip 92ef3c71: static gates clean (changelog, file-size, complexity 2624<=2774, cognitive 1182<=1223, dead-code 411<=416), typecheck:core clean, focused tests green.
Empty-envelope 400 (no error field, empty content, finish_reason null) now rotates/retries instead of propagating as success; 200/streaming path never buffered; real-error 400s untouched. account-rotation + new rotation suite 34/34 on the board. Thank you @maxmad64bis!
Validated on a worktree over the current tip: account-fallback-service 91/91 plus the five sibling lockout suites 24/24. The measurement in the body (40 of 111 passthroughModels providers uncovered on this branch) is the clincher — one lookup via getProviderById().passthroughModels beside the existing checks, closing the #11071 remainder for shared-registry gateways (port of #11075 which had only landed on main). Thank you @yourspraveen!
Validated on a worktree over the current tip: the red it fixes reproduced exactly as described (media-page-client-browser-bundle red since #11122 — providerRegistry became reachable from the dashboard client bundle via node:net). Post-fix: bundle test 2/2 green, new ip-parity suite + is-local-provider 7/7, all 7 outboundUrlGuard consumer suites 76/76 (the moved normalizeHost/isPrivateHost keep their re-exports; routing behavior untouched). Thank you @yourspraveen — clean surgical extraction with a pure-JS ipVersion mirroring Node's own regexes.
Cherry-picked onto the current tip (authorship preserved), generated-count noise stripped. Two pre-merge adjustments: (1) dropped the unrelated localDb.ts re-export hunk (nothing in this PR uses those symbols); (2) automated security review flagged the blocked-provider list resolving once at server creation — the handler now rebuilds the schema per invocation via the resolver (advertised tools/list schema stays a creation-time snapshot, which is inherent to MCP). Vitest: new runtime-blocked-schema suite 3/3, full MCP __tests__ 118/118; contract suites (mcp-web-search-provider-enum-contract, search-blocked-providers-11100) 6/6; typecheck clean. This closes the residual gap noted when #11120 was closed. Thank you @rqzbeh!
Cherry-picked onto the current tip (authorship preserved), noise files stripped. Pre-merge addition: regenerated the golden snapshot with UPDATE_GOLDEN=1 because the branch's snapshot predated two legitimate tip changes — the dify bare-root from #11065 and the hackclub removal from #11123. The regen'd delta contains exactly those two (audited). This also drains a live base-red: provider-translate-path-golden was failing on the pure tip. 3/3 green. Thank you @rqzbeh!
Merged after conflict resolution validated on the combined board (50/50 casing tests green, typecheck:core clean). Two pre-merge adjustments on the branch: (1) the utilization route conflict resolved to the tip shape — its asNullableString/displayName version is newer than the branch's; (2) dropped the newly-added src/lib/db/connections.ts, orphaned once the route kept the tip shape (tip already uses getProviderConnectionById) — nothing imported it. The casing fix itself lands intact: non-streaming OpenAI→Claude conversion now restores canonical tool names, identity echoes no longer pin lowercase, and TOOL_RENAME_MAP gained the Task* tools. Fixes the live-reproduced Claude Code 'No such tool available: bash' failures. Thank you @linhdmn — outstanding repro and root-cause writeup!
Validated on the combined board over tip 80d931ae: kimi suites green (executor-kimi-web, kimi-partner-aff-links, token-health-check-kimi 33 assertions), vitest providerPageHeaderKimiPartnerLink green, typecheck:core clean. Audited the diff: only kimi-web switches to the international www.kimi.ai (Connect-RPC base, website, auth hints); kimi-coding / kimi-coding-apikey affiliate links stay on kimi.com as intended — asserted by the updated tests. Owner approved merge without the VPS smoke. Thank you @MeRezaRezaei!
