Cherry-picked both value commits (a1fa49a8 + a91f6e93, incl. the fail-open guard for unparseable hostnames) onto the current tip, dropping the stale base-red sync commits. Focused tests: is-local-provider-11091 + the isLocalProvider consumer suites (provider-validation-specialty, ollama-local-provider) 127/127; mutation coverage no-drift. Fixes#11091 — RFC1918/CGNAT/link-local/mDNS hosts now classify as local via the existing isPrivateHost. Thank you @rqzbeh!
Cherry-picked the value commit (2c9202e4) onto the current tip, dropping the stale base-red sync commits. Focused tests: ollama-404-model-lockout 2/2 + the five sibling lockout suites (combo-provider-cooldown-sibling, 8247-model-unhealthy, vertex-passthrough, nvidia-410, account-fallback-service) 112/112; mutation coverage no-drift. Fixes#11071 — local/self-hosted 404s now scope to model lockout per the resilience doctrine. Thank you @rqzbeh!
Cherry-picked the value commit (55da60f5) onto the current tip, dropping the stale base-red sync commits. Focused tests: opencode-v2-config-11070 + merge-provider-guard + config-dir-single-source 12/12; mutation coverage gate no-drift (new test registered in tap.testFiles by the PR itself). Fixes#11070 — setup-opencode now emits both V1 and V2 blocks. Thank you @rqzbeh!
Cherry-picked the value commit (0e918816) onto the current tip, dropping the stale base-red sync commits. Focused tests: pollinations-api-key-required 1/1 plus the whole optional-key suite 142/142 (two legacy assertions in provider-route-schemas flipped to the new key-required contract, commented with the PR). Fixes#11096 — Pollinations answers 401 anonymously now. Thank you @rqzbeh!
Cherry-picked the value commit (ac09b6b5) onto the current release tip, dropping the stale base-red sync commits that no longer apply. Focused test cline-model-format-11099 2/2 green; check:provider-consistency OK (267/348/0). Fixes#11099 — the zai→z-ai namespace typo. Thank you @rqzbeh!
0ff0490ada (#10906) rewrote this suite keeping four module-level holders
typed as any for the dynamically imported src/lib/db/core.ts exports —
4 new unsuppressed @typescript-eslint/no-explicit-any errors that failed
the 'No new ESLint warnings' fast gate (npm run lint:json -- --max-warnings 0,
exit 1 on the pure tip; CI run 32403951361).
Replaces the any holders with a type-only CoreModule reference
(typeof import(...)), so the runtime dynamic-import timing that motivates
the pattern (DATA_DIR must be set before core.ts loads) stays untouched.
Purely typal: file lints clean and its 7 assertions still pass.
Two gaps left the vi locale red on the release tip:
- 6cd4d38e21 (#11069) added 14 providers.harImport* keys (HAR-import UX
for the m365 Copilot web flow) to en.json without vi translations, so
key parity, ICU placeholder and ICU parse tests all failed on the same
missing set. Added the 14 Vietnamese translations with the {minutes}
ICU placeholder preserved verbatim.
- agentSkills.catalog.omni-webhooks.description carried a literal
__MISSING__: fallback; replaced with a real Vietnamese translation,
keeping technical event names (request.completed, request.failed,
quota.exceeded) untouched.
tests/unit/i18n-vi-completeness.test.ts: 5/5 green.
The 'does not force 1M onto non-hyperagent providers' assertion expected
windsurf's per-model 200000 pin for claude-opus-4.7-max. That premise is
stale after two intentional base merges, both already on release/v3.8.50:
- acc066db3f (#8228) retired windsurf as a built-in provider (replaced by
devin-desktop), removing its explicit contextLength pin;
- 00dfdadf93 (#11034) added effort-suffix variant resolution in
getResolvedModelCapabilities: when no direct source resolves,
stripKnownEffortSuffix maps claude-opus-4.7-max to its base model
claude-opus-4.7 -> canonical claude-opus-4-7, whose real catalog window
is 1M (cursor catalogs the same id as 'Opus 4.7 1M Max').
So the 1M now comes from the shared model catalog via legitimate base-model
resolution, not from the forbidden hyperagent-default leak the test guards
against. Discrimination evidence: reproduces on the pure tip with zero diff
(git log -p origin/release/v3.8.50 on both source and test); the test file
itself was untouched since d54a659804 while both source merges landed later.
The bluesminds/claude-opus-4-5 -> 200000 assertion is unchanged and still
guards the original #8496 intent: a non-hyperagent provider's own pin must
win over name heuristics and sibling windows.
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
Learned reasoning_effort caps mirror the merged learnedThinkingCaps mechanism: parse the upstream 4xx enum, clamp, retry once, consult proactively — covers custom openai-compatible connections the static registry can't. 22 new test cases + full regression list green. Fixes#11111. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
TDD red->green: parallel function_call items now get distinct stable index/id at .added time via a per-call Map, interleaved argument deltas no longer glue, dual item_id/output_index correlation. 120-test translator suite green. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
Operator-declared per-provider error rules via settings, consulted before the built-ins; the allowlist bypass is correct — declaring a rule is itself the opt-in, and no provider-specific rule is hardcoded. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
Config-audit survives restarts with bounded growth (migration 161 + OR IGNORE seed, retention wired into runAutoCleanup); route cabling deliberately out of scope. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
Strict schema + {sanitized, rejected} DB boundary — silent validation degradation now answers 400 with the offending keys. Caller audit done: only the providers write path consumes the sanitizers. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
Defaults summary and strips malformed ids on kept Responses input items — both 400s observed against live muse-spark traffic; the flipped legacy assertion is documented contract propagation. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
Real suggestionCount replaces the conflated link count (deprecated alias kept), dashboard deep-link fixed. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
Restores the models[0] dashboard default silently changed by #11051, with narrow guards instead of a brittle full snapshot. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
Closes the two #11050 blockers: dispatcher tests now derive from WEBHOOK_EVENT_VALUES (no more TypeError on removed events) and vi.json carries a real translation. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
Live-confirmed 400 (missing summary on freshly-built Chat->Responses reasoning items); the two legacy assertions updated are contract propagation, each commented. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean, typecheck:core clean, focused tests green.
Test-only followup to #11043 — 3 direct tests for splitConcatenatedToolCallArguments plus the index-normalization comment, exactly the two review nits. Thank you @maxmad64bis!
Validated on the combined batch board over release/v3.8.50 tip d91238b7: static gates clean (changelog-integrity, file-size, complexity 2619<=2774, cognitive 1178<=1223, dead-code 411<=416), typecheck:core clean, focused tests green (262 batch-touched tests pass; the only red is the pre-existing vi.json harImport key drift from #11069, reproduced on the pure tip — not this batch).
Follow-up fix pushed to the branch pre-merge: the changelog fragment was prose, now a bullet (fragment gate). Thank you @maxmad64bis — the root-cause writeup (per-call finish_reason vs general terminal marker making the in-flight guard a no-op) is exactly right.
Same inherited base-red set as #11148 (getTokenLimit + vi parity + ESLint gate — all reproduce on the pure base tip). Merge integrity, Docs Gates, Vitest, Build: green.
Inherited base-red at merge time (discriminated against the pure base tip, both reproduce WITHOUT this diff): getTokenLimit test + Vietnamese i18n key parity (new UI strings merged untranslated) + No new ESLint warnings gate. Merge integrity, Docs Gates, Vitest, Fast Production Build: green.
⭐5 — M365 Copilot (BizChat) individual/consumer path — 3 itens: (1) forma de invocação do #10718 derivou de novo (2026-08-21 capture): optionsSets 14→34, allowedMessageTypes 6→30, tone "magic"→"Magic", plugins []→[{BingWebSearch}], disconnectBehavior em todos os tiers, +8 keys de clientInfo; verificado contra conta real com round-trip WebSocket (ping-then-close → resposta real). (2) Aviso sobre o alias Antigravity gemini-3.1-pro-high ainda não publicado (3.8.49 pré-data). (3) Botão "Import .har file" no modal de credencial M365.
Conflito resolvido em copilot-m365-frames.ts (board vs release tip): mantive o forwarding de opts.plugins/toolChoice/customInstructions do HEAD com os NOVOS defaults da captura (BingWebSearch builtin, tone "Magic"). Alinhei 3 testes pré-existentes que afirmavam o contrato antigo (m365-bizchat-frames-4042 clientInfo, m365-tone-model-variants tone, copilot-m365-tool-calls plugins) — propagação de contrato, não mascaramento. Rebaselinei AddApiKeyModal 1073→1080 (crescimento próprio da parte 3, ~Har import button) com anotação.
Validação: typecheck limpo, 142/142 testes m365/copilot verdes, changelog-integrit/file-size/eslint OK.
⭐5 — Fecha #10265: chat do command-code migra do endpoint CLI-only /alpha/generate (version-gated + proxy-blocked para callers externos) para o documentado /provider/v1/chat/completions (OpenAI format). Removido o envelope CLI reverse-engineered (config/memory/taste/skills + headers CLI-impersonation), substituído por passthrough OpenAI plano com normalização de model id vendor-prefixed (#10809), clamp de max_tokens (#5166), sanitização de reasoning_effort. commandCode.ts 1037→171 linhas.
Validado no worktree board sobre tip: typecheck:core limpo; 175/175 testes focados (command-code executor/vision/usage/maxtokens/user-array/validation-specialty/responses-handler/provider-models-scoping); changelog-integrity/file-size/complexity/cognitive todos OK. TDD RED→GREEN documentado.
⭐5 — /v1/combos strip connectionId de propósito; dois passos pinando contas DIFERENTES do mesmo provider viram objetos byte-idênticos e um cliente conclui que não há failover. Adiciona accountPinned (boolean derivado do connectionId, nunca vaza o id) em cada passo model; sempre true/false, nunca em combo-ref. Sem rota/schema/mudança de caller. TDD. Fecha #10968.
⭐4 — Suporte de backend nativo Bun + Dockerfile.bun multi-stage + fallback dinâmico de driver SQLite (better-sqlite3 prioritário sob Bun, bun:sqlite fallback; Node preservado) + correção de estabilidade do DAST CI smoke.
Validado a fundo (worktree board sobre tip): bun-support 4/4, typecheck:core limpo, dashboard-typecheck OK (220 dentro do baseline), open-sse-typecheck OK (5 pré-existentes), gate de runtime OK sob Node, changelog-integrity OK, file-size/complexity/cognitive/dead-code OK. Verificado que o driver preserva a cadeia Node/falback conforme AGENTS.md; teste bun-support presente. Baselines de typecheck removidos são ratchet honesto (erros não existem mais).
OBS: destravei 2 base-reds do tip neste turno (push direto 7ffa3ef): movi o changelog fragment da #11050 da seção inválida breaking/ para fixes/, e rebaselinei AddApiKeyModal 1067->1073 (crescimento da #11056). Sem isso a #11039 e o resto da fila ficariam vermelhos.
⭐5 — Atualiza domínio do provider hailuo-web para chat.minimax.io (#11000). Merge autorizado pelo operador (ordem explícita de prosseguir sem o smoke VPS da Hard Rule #18).
⭐5 — computeConnectionDefaultName aceita array de nomes existentes e acha o próximo não-conflitante, evitando overwrite de nome de conexão. TDD. Fecha #11033.
⭐5 — Preenche os 3 test.skip com asserções reais (resetDbInstance troca o singleton preservando a linha no disco, journal_mode WAL, schema_version=1). Além do valor pretendido, o autor redesenhou o setup()/cleanup() do arquivo corrigindo um bug de isolamento pré-existente que eu apontei em review: DATA_DIR/SQLITE_FILE são const de topo de módulo; o cleanup() usava require() CJS que nunca resetava a instância ESM-importada, então os testes 1-4 passavam "por acidente" contra a conexão nunca fechada. Agora: tempDir compartilhado definido antes do primeiro import, resetDbInstance importado via ESM uma vez, handle fechado antes de cada reopen, e o catch{} silencioso removido. 7/7 verdes no arquivo inteiro.
⭐5 — LIMIT negativo = "sem limite" no SQLite: ?limit=-1 retornava a leaderboard inteira em endpoint management autenticado; ceil por Math.min só no upper. Duas camadas: route rejeita não-inteiro/fora de range com 400 (mesmo contrato de parseListLimit), getTopN clampeia como backstop defense-in-depth + exporta LEADERBOARD_MAX_LIMIT. TDD red→green, 5 casos novos, 84/84 suíte gamification. Fecha #11058.
⭐5 — Dedupe dos 3 entries byte-idênticos (kimi-k2.7-code, qwen3.5-plus, qwen3.6-plus) entre opencode-zen e opencode-go via OPENCODE_ZEN_GO_SHARED_MODELS (padrão GLM_SHARED_MODELS), frozen e testado por deep-equal. Pure move, sem mudança de comportamento; check:provider-consistency OK (267 entries, 348 providers). TDD 3/3. Follow-up de #11049/#11048. Base-red #9985 inherited.
⭐5 — Muse Spark responde só na Responses API; #10874 corrigiu só o provider `opencode`, mas `opencode-zen` (faltava targetFormat em 2 entries) e `opencode-go` (6 entries effort-tier) nunca receberam. Espelha a declaração. freebuff NÃO tocado (backend codebuff.com /chat/completions distinto, verificado). TDD 2/2 + 18/18 irmãos. Fecha #11048. Base-red #9985 inherited.
⭐5 — Providers que não bumpam index/id em tool calls repetidas do mesmo nome colam N arguments JSON num só ({...}{...}{...}); leitores a jusante pegam só o primeiro e dropam o resto em silêncio. Detecta N objetos concatenados e divide de volta em N tool_calls. TDD 20/20 + 86/86 irmãos. Fecha #11044. Base-red #9985 inherited.
⭐5 — OpencodeExecutor buscava targetFormat com o id bruto do provider em vez do alias público ("oc"), caindo em default openai → corpo Responses no URL chat/completions (400). resolveOpencodeTargetFormat resolve via PROVIDER_ID_TO_ALIAS. TDD 4/4 + 32/32 irmãos, lint/tsc limpos. Fecha #11046. Base-red #9985 inherited.
⭐5 — Drain de base-reds v3.8.50 (#9985): corrige build-break do #10939 (import fantasma @/lib/db/connections → reroteado por getProviderConnectionById real), headers Anthropic no fallback GLM, fallthrough do reject de listing CC-compatible, no-auth leak no auto-combo, retry de mesmo-account nos combo live tests, pin onnxruntime-node ~1.24.3; + alignments de teste + gates (env-doc, docs-counts 347→348, openapi floor, i18n vi/pt-BR).
Reconciliado com o tip: 5 conflitos de contagem gerada (SVGs + PROVIDER_REFERENCE — HEAD já em 348, branch em 347 stale) resolvidos ficando com o HEAD. typecheck limpo, gates verdes, 5/5 testes focados (#10939 regressão + utilização), lint limpo nos tocados.
Também rebaselined neste turno via push direto ao tip: config/quality/file-size-baseline.json modelCapabilities.ts 1016→1072 (drift de PRs ja mergeadas, ex: #11034), com changelog fragment — tip estava vermelho nesse gate e travava esta PR.
