@bitkyc08/opencodex 2.88.0 → 2.89.0

package/src/server/management/route-registry.ts

@@ -364,6 +364,7 @@ export const MANAGEMENT_ROUTES: readonly ManagementRoute[] = [
   { method: "PUT", path: "/api/oauth/accounts/pool", module: "server/management/oauth-account-routes", mutates: true },
   { method: "PUT", path: "/api/providers/keys/active", module: "server/management/oauth-account-routes", mutates: true },
   { method: "PUT", path: "/api/providers/keys/alias", module: "server/management/oauth-account-routes", mutates: true },
+  { method: "PUT", path: "/api/providers/keys/pause", module: "server/management/oauth-account-routes", mutates: true },
   // server/management/provider-routes
   { method: "DELETE", path: "/api/providers", module: "server/management/provider-routes", mutates: true },
   { method: "GET", path: "/api/provider-context-caps", module: "server/management/provider-routes", mutates: false },
@@ -373,6 +374,7 @@ export const MANAGEMENT_ROUTES: readonly ManagementRoute[] = [
   { method: "GET", path: "/api/providers", module: "server/management/provider-routes", mutates: false },
   { method: "PATCH", path: "/api/providers", module: "server/management/provider-routes", mutates: true },
   { method: "POST", path: "/api/providers", module: "server/management/provider-routes", mutates: true },
+  { method: "POST", path: "/api/providers/test-completion", module: "server/management/provider-routes", mutates: true },
   { method: "POST", path: "/api/providers/test", module: "server/management/provider-routes", mutates: true },
   { method: "PUT", path: "/api/providers", module: "server/management/provider-routes", mutates: true },
   { method: "PUT", path: "/api/provider-context-caps", module: "server/management/provider-routes", mutates: true },

package/src/server/messages-native.ts

@@ -1,3 +1,4 @@
+import { captureAccountAutoPauseAttempt, wasAccountAutoPaused } from "../oauth/account-auto-pause";
 import { createPhysicalSendReporter } from "../lib/request-execution-budget";
 import { createInferenceSendBudget } from "./inference/context";
 /**
@@ -120,7 +121,7 @@ import {
   recordKeyWireAttemptUsage,
   type RequestLogContext,
 } from "./request-log";
-import { fetchWithHeaderTimeout, providerFetch, safeHostLabel, sendWithConnectionPolicy } from "./responses/fetch-helpers";
+import { fetchWithHeaderTimeout, physicalSendObserverInit, providerFetch, safeHostLabel, sendWithConnectionPolicy } from "./responses/fetch-helpers";
 import { anthropicBillingPhysicalFetch } from "../providers/anthropic-billing-ownership";
 import { rotateAnthropicBillingKeyOnResponse } from "../providers/anthropic-billing-recovery";
 import { linkAbortSignal } from "./responses/core-lifetime";
@@ -545,6 +546,7 @@ export async function handleNativeMessages(options: HandleNativeMessagesOptions)
     pendingDistributedFailure = undefined;
     record?.();
   };
+  /** Send the native Messages request under the transient retry policy, re-validating the selected OAuth account or API key at the dispatch boundary. */
   const send = async (recovery?: "rate-limit-429" | "oauth-401" | "oauth-account-403" | "key-429" | "key-401"): Promise<Response> => {
     const remaining = remainingTransientSends();
     if (requestTransientPolicy && remaining <= 0) {
@@ -616,6 +618,7 @@ export async function handleNativeMessages(options: HandleNativeMessagesOptions)
             }
             const releaseFamily = snapshot && oauthBinding ? anthropicModelQuotaFor(oauthBinding.instance).claimAnthropicFamilyRevalidation(snapshot.accountId, route.modelId) : () => {};
             if (!releaseFamily) throw new AnthropicAccountCooldownError(1);
+            const observeAutoPause = captureAccountAutoPauseAttempt(route.providerName, activeProvider, { config, oauth: sendingBinding?.snapshot }, headers);
             const selection = activeProvider._apiKeyAttempt;
             let dispatched: Response;
             const spendReport = createPhysicalSendReporter(physicalBudget, () => ({
@@ -641,7 +644,7 @@ export async function handleNativeMessages(options: HandleNativeMessagesOptions)
                   anthropicBillingPhysicalFetch((activeProvider as OcxProviderTransport).fetch ?? execute, route.providerName, activeProvider),
                   route.providerName, activeProvider, wire.url, activeProvider.apiKeyTransport === "bearer" ? "authorization" : "x-api-key"),
                 wire.url,
-                applyUpstreamRecoveryInit({ ...init, method: "POST", headers, body: wire.body }, transportRecovery),
+                applyUpstreamRecoveryInit({ ...init, method: "POST", headers, body: wire.body, ...physicalSendObserverInit(observeAutoPause, wire.url) }, transportRecovery),
                 { providerName: route.providerName, provider: activeProvider },
               );
             } finally {
@@ -659,7 +662,7 @@ export async function handleNativeMessages(options: HandleNativeMessagesOptions)
                   && credentialGeneration(current.credential) === snapshot.generation
                   && current.credential.accountId === providerAccountUuid) {
                   bindAnthropicRefusalCredentialForSend(dispatched, sendOwner, providerAccountUuid);
-                  recordAnthropicAccountQuotaFromHeadersForInstance(nativeInstance!, snapshot.accountId, dispatched.headers, writerGeneration, dispatched.status, route.modelId);
+                  recordAnthropicAccountQuotaFromHeadersForInstance(nativeInstance!, snapshot.accountId, dispatched.headers, writerGeneration, dispatched.status, route.modelId, snapshot.generation);
                 }
               } catch { /* Passive observation must not fail the response. */ }
             }
@@ -672,7 +675,7 @@ export async function handleNativeMessages(options: HandleNativeMessagesOptions)
               ? () => { recordDistributedResponseFailure(dispatched, config, route.providerName, req.signal); }
               : undefined;
             if (!dispatched.ok) await recordKeyAttemptFailure(logCtx, dispatched, init.signal ?? upstream.signal);
-            return dispatched;
+            return observeAutoPause ? observeAutoPause(dispatched) : dispatched;
           },
         }),
       ),
@@ -700,9 +703,19 @@ export async function handleNativeMessages(options: HandleNativeMessagesOptions)
     let billingRecoveryActive = false;
     const keySendAvailable = () => transientSendAvailable()
       && (!billingRecoveryActive || Math.max(physicalSendsUsed, physicalBudget.used) < physicalBudget.policy.maxTotalModelSends);
+    let autoPauseMoves = 0;
     const recoveryPasses = activeProvider._apiKeyAttempt?.distributed ? activeProvider.apiKeyPool?.length ?? 1 : 1;
     for (let pass = 0; ; pass++) {
       const priorResponse = response;
+      for (; wasAccountAutoPaused(response) && !oauthBinding && config.providers[route.providerName]?.apiKey && autoPauseMoves < 3 && keySendAvailable(); autoPauseMoves++) {
+        const fresh = resolveCurrentProviderApiKeyTransport(config, route.providerName, { ...activeProvider, _apiKeyAttempt: undefined });
+        const next = fresh && (selectProactiveApiKeyTransport(config, route.providerName, fresh) ?? fresh);
+        if (!next) break;
+        try { void response.body?.cancel().catch(() => {}); } catch { /* closed */ }
+        rebuildFor(next);
+        response = await send("key-401");
+      }
+
       // A credential-scoped 401 on a static key pool says nothing about the sibling keys.
       while (response.status === 401 && distributedResponseOwnsKey && hasKeyPoolFailover(activeProvider) && keySendAvailable()) {
         const rotated = rotateProviderTransportOn401(config, route.providerName, activeProvider, {
@@ -762,6 +775,7 @@ export async function handleNativeMessages(options: HandleNativeMessagesOptions)
       if (response === priorResponse || isNonReplayableResponse(response) || !keySendAvailable()) break;
       // Billing class transitions share the physical-send ceiling; distributed-only
       // transitions additionally retain their original bounded pool-pass allowance.
+      if (wasAccountAutoPaused(response) && autoPauseMoves < 3) continue;
       if (distributedResponseOwnsKey && (
         (billingRecoveryActive && [401, 429].includes(response.status))
         || (activeProvider.adapter === "anthropic" && hasKeyPoolFailover(activeProvider) && [400, 402].includes(response.status))

package/src/server/messages-shared.ts

@@ -13,6 +13,7 @@
  * other.
  */
 import { sniffImageDimensions } from "../adapters/anthropic-image-guard";
+import { anthropicCacheWrite1hTokens } from "../adapters/anthropic/cache-retention";
 import { createToolCallIdAllocator } from "../adapters/tool-call-id";
 import { idleDeadline } from "../lib/abort";
 import {
@@ -29,12 +30,13 @@ import type { RequestLogContext } from "./request-log";
 import { recordGenerationEvent } from "./request-log-generation-window";
 import type { OcxConfig } from "../types";
 
-export function anthropicUsageToOcx(usage: Rec | undefined): { inputTokens: number; outputTokens: number; cachedInputTokens?: number; cacheReadInputTokens?: number; cacheCreationInputTokens?: number } | undefined {
+export function anthropicUsageToOcx(usage: Rec | undefined): { inputTokens: number; outputTokens: number; cachedInputTokens?: number; cacheReadInputTokens?: number; cacheCreationInputTokens?: number; cacheCreation1hInputTokens?: number } | undefined {
   if (!usage) return undefined;
   const num = (v: unknown) => typeof v === "number" ? v : 0;
   const hasCache = usage.cache_read_input_tokens !== undefined || usage.cache_creation_input_tokens !== undefined;
   const read = num(usage.cache_read_input_tokens);
   const write = num(usage.cache_creation_input_tokens);
+  const write1h = anthropicCacheWrite1hTokens(usage, write);
   // Anthropic input_tokens excludes cache read/write; normalize to the canonical
   // inclusive convention (types.ts OcxUsage / devlog 070). cached = READS only.
   return {
@@ -44,6 +46,7 @@ export function anthropicUsageToOcx(usage: Rec | undefined): { inputTokens: numb
       cachedInputTokens: read,
       cacheReadInputTokens: read,
       cacheCreationInputTokens: write,
+      ...(write1h !== undefined ? { cacheCreation1hInputTokens: write1h } : {}),
     } : {}),
   };
 }

package/src/server/request-log.ts

@@ -982,7 +982,7 @@ export function usageFromResponsesPayload(usage: unknown): OcxUsage | undefined
   const raw = usage as {
     input_tokens?: unknown;
     output_tokens?: unknown;
-    input_tokens_details?: { cached_tokens?: unknown; cache_write_tokens?: unknown };
+    input_tokens_details?: { cached_tokens?: unknown; cache_write_tokens?: unknown; cache_write_1h_tokens?: unknown };
     output_tokens_details?: { reasoning_tokens?: unknown };
     total_tokens?: unknown;
     prompt_tokens?: unknown;
@@ -1004,6 +1004,9 @@ export function usageFromResponsesPayload(usage: unknown): OcxUsage | undefined
       ...(typeof raw.input_tokens_details?.cache_write_tokens === "number"
         ? { cacheCreationInputTokens: raw.input_tokens_details.cache_write_tokens }
         : {}),
+      ...(typeof raw.input_tokens_details?.cache_write_1h_tokens === "number"
+        ? { cacheCreation1hInputTokens: raw.input_tokens_details.cache_write_1h_tokens }
+        : {}),
       ...(typeof raw.output_tokens_details?.reasoning_tokens === "number"
         ? { reasoningOutputTokens: raw.output_tokens_details.reasoning_tokens }
         : {}),
@@ -1864,7 +1867,7 @@ export function aggregateAttemptUsage(
   if (usages.length === 0) return { status };
 
   const sumOptional = (
-    key: "cachedInputTokens" | "cacheReadInputTokens" | "cacheCreationInputTokens"
+    key: "cachedInputTokens" | "cacheReadInputTokens" | "cacheCreationInputTokens" | "cacheCreation1hInputTokens"
       | "reasoningOutputTokens" | "providerCredits",
   ): number | undefined => {
     const present = usages.flatMap(usage => (
@@ -1875,6 +1878,7 @@ export function aggregateAttemptUsage(
   const cachedInputTokens = sumOptional("cachedInputTokens");
   const cacheReadInputTokens = sumOptional("cacheReadInputTokens");
   const cacheCreationInputTokens = sumOptional("cacheCreationInputTokens");
+  const cacheCreation1hInputTokens = sumOptional("cacheCreation1hInputTokens");
   const reasoningOutputTokens = sumOptional("reasoningOutputTokens");
   const providerCredits = sumOptional("providerCredits");
   const totalTokens = usages.reduce(
@@ -1888,6 +1892,7 @@ export function aggregateAttemptUsage(
     ...(cachedInputTokens !== undefined ? { cachedInputTokens } : {}),
     ...(cacheReadInputTokens !== undefined ? { cacheReadInputTokens } : {}),
     ...(cacheCreationInputTokens !== undefined ? { cacheCreationInputTokens } : {}),
+    ...(cacheCreation1hInputTokens !== undefined ? { cacheCreation1hInputTokens } : {}),
     ...(reasoningOutputTokens !== undefined ? { reasoningOutputTokens } : {}),
     ...(providerCredits !== undefined ? { providerCredits } : {}),
     ...(status === "estimated" ? { estimated: true } : {}),

package/src/server/responses/account-auto-pause-recovery.ts

@@ -0,0 +1,26 @@
+import type { ResponsesSendBudget } from "./request-send-budget";
+
+const moves = new WeakMap<object, number>();
+/** One same-provider credential move, funded by the existing recovery owner. */
+export async function recoverAutoPausedAccount<T>(options: {
+  requestKey: object;
+  remaining?: number;
+  reserve: ResponsesSendBudget["reserveCredentialHop"];
+  refresh: () => Promise<boolean>;
+  refetch: () => Promise<T>;
+  setPermit: (permit: ResponsesSendBudget["pendingHopPermit"]) => void;
+}): Promise<T | undefined> {
+  const count = moves.get(options.requestKey) ?? 0;
+  if (count >= 3 || options.remaining !== undefined && options.remaining <= 0) return;
+  const hop = options.reserve("auth-recovery", "account-auto-pause", true);
+  if (!hop.allowed) return;
+  try {
+    if (!await options.refresh()) return;
+    moves.set(options.requestKey, count + 1);
+    options.setPermit(hop.permit);
+    return await options.refetch();
+  } finally {
+    options.setPermit(undefined);
+    hop.permit?.release();
+  }
+}

package/src/server/responses/adapter-dispatch.ts

@@ -1,5 +1,8 @@
+import { recoverAutoPausedAccount } from "./account-auto-pause-recovery";
+import { wasAccountAutoPaused, wasPermanentAccountRefusal } from "../../oauth/account-auto-pause";
 import { bindAnthropicRejectionResponse } from "../../adapters/anthropic-rejection-memo";
 import { deferPhysicalSendReporter } from "./deferred-send-reporter";
+import { usesApiKeyAccount } from "../../providers/label";
 import { spendLedgerStorageErrorResponse } from "./spend-storage-error";
 import { classifyAnthropic429 } from "../../oauth/anthropic-rate-limit-policy";
 import { rotateAnthropicAccountOnResponseForInstance } from "../../oauth/anthropic-account-refusal";
@@ -141,6 +144,7 @@ export async function prepareAdapterExchange(
     | "sameTargetToken"
     | "transportToken"
     | "oauthDispatch"
+    | "recoverAutoPausedSelection"
     | "imageTierBias"
     | "oauth401ReplayAttempted"
     | "isOAuth401ReplayProvider"
@@ -394,11 +398,12 @@ export async function prepareAdapterExchange(
     upstreamResponse = new Response(text, { status: original.status, statusText: original.statusText, headers });
     return upstreamResponse;
   };
+  const autoPauseEnabled = config.providers[route.providerName]?.accountAutoPause?.enabled === true;
   const billingPoolRequest = isAnthropicBillingKeyPool(route.provider);
   let billingRecoveryActive = false;
-  const genericRefetchCountsSends = () => sendBudgetState.adapterSendBudget?.spendEnforced === true
+  const genericRefetchCountsSends = () => options.comboAttempt === true || sendBudgetState.adapterSendBudget?.spendEnforced === true
     || transientRetryPolicyFor(route.provider) !== null || resetReplayPolicyFor(route.provider) !== null
-    || billingPoolRequest || billingRecoveryActive;
+    || billingPoolRequest || billingRecoveryActive || autoPauseEnabled;
   let handoffPermit: ResponsesSendBudget["pendingHopPermit"];
   try {
     if (transportState.activeAdapter.fetchResponse) {
@@ -520,10 +525,10 @@ export async function prepareAdapterExchange(
           label: safeHostLabel(builtInitialRequest.url),
           claimAmbiguousResend: claimPreHeaderResend,
           ...(sendBudgetState.adapterSendBudget?.spendEnforced || options.comboDispatchPermit || transientPolicy || resetPolicy || compactPrepaid || options.comboAttempt
-            || billingPoolRequest
+            || billingPoolRequest || autoPauseEnabled
             ? {
               // A pending compaction permit already booked this leg's first physical send.
-              ...(transientPolicy || resetPolicy || compactPrepaid || options.comboAttempt
+              ...(transientPolicy || resetPolicy || compactPrepaid || options.comboAttempt || autoPauseEnabled
                 ? {
                   attempts: Math.min(initialSendCap,
                     sendBudgetState.initialSendAllowance(initialSendCap) + (compactPrepaid ? 1 : 0)),
@@ -758,7 +763,9 @@ export async function prepareAdapterExchange(
             : undefined;
           let refetchPermitUsed = false;
           const refetchReporter = refetchAllowance ? transientSendReporter(refetchAllowance.permit) : undefined;
-          const deferredRefetch = billingPoolRequest && refetchReporter ? deferPhysicalSendReporter(refetchReporter) : undefined;
+          const deferredRefetch = (billingPoolRequest || options.comboAttempt) && refetchReporter
+            ? deferPhysicalSendReporter(refetchReporter) : undefined;
+          const pendingKeyMetadata = deferredRefetch && usesApiKeyAccount(route.provider);
           try {
             return await refetchWithPolicy(
               recoveryKind => {
@@ -767,14 +774,14 @@ export async function prepareAdapterExchange(
                 });
                 // Prepare key metadata before the guarded transport commits its attempt;
                 // counters and permits still wait for the physical executor after pacing.
-                if (deferredRefetch) transportState.noteRoutedAttemptSend(retryEstimate, recoveryKind ?? recovery);
+                if (pendingKeyMetadata) transportState.noteRoutedAttemptSend(retryEstimate, recoveryKind ?? recovery);
                 const confirmDispatch = () => {
                   if (refetchAllowance?.permit && !refetchPermitUsed) {
                     if (!refetchAllowance.permit.use()) throw new SendBudgetExhaustedError(safeHostLabel(retryRequest.url));
                     refetchPermitUsed = true;
                   }
-                  if (!deferredRefetch) transportState.noteRoutedAttemptSend(retryEstimate, recoveryKind ?? recovery);
-                  // Eligible pools confirm after pacing at the physical executor; legacy
+                  if (!pendingKeyMetadata) transportState.noteRoutedAttemptSend(retryEstimate, recoveryKind ?? recovery);
+                  // Counted Combo/pool sends confirm at the physical executor; legacy
                   // helper paths retain their existing thunk admission boundary.
                   onDispatch?.();
                   if (preserveFailureResponse) replacementAdmitted = true;
@@ -887,7 +894,7 @@ export async function prepareAdapterExchange(
       const adapterOwnsDispatch = transportState.activeAdapter.fetchResponse !== undefined;
       const hop = reserveCredentialHop("auth-recovery",
         `${route.providerName}|${route.modelId}|antigravity-auth`,
-        !adapterOwnsDispatch && (transientRetryPolicyFor(route.provider) !== null || resetReplayPolicyFor(route.provider) !== null));
+        !adapterOwnsDispatch && genericRefetchCountsSends());
       if (!hop.allowed) return null;
       try {
         const snapshot = await failoverAccountSnapshot(route.providerName, nextId);
@@ -923,6 +930,15 @@ export async function prepareAdapterExchange(
     };
    // Keep recovery kinds in sync with the native Responses `passthroughRecovery:` loop above.
    recovery: for (;;) {
+      if (wasAccountAutoPaused(upstreamResponse) && route.provider.authMode !== "forward") {
+        const recovered = await recoverAutoPausedAccount({ requestKey: transportState, reserve: reserveCredentialHop,
+          remaining: transientSendCapFor(transientRetryPolicyFor(route.provider)?.attempts, sendBudgetState.targetSendsUsed),
+          refresh: () => transportState.recoverAutoPausedSelection(parsed), refetch: () => rebuildAndRefetch("oauth-account-429"), setPermit: permit => { sendBudgetState.pendingHopPermit = permit; } });
+        if (!recovered) break;
+        if ("failed" in recovered) return recovered.failed;
+        upstreamResponse = recovered; continue recovery;
+      }
+
       // Preserve the terminal verdict through adapter and combo error formatting.
       // This also covers a reset reached by a 401/429/413 recovery refetch.
       if (isNonReplayableResponse(upstreamResponse)) {
@@ -942,7 +958,7 @@ export async function prepareAdapterExchange(
         continue recovery;
       }
      if (
-       upstreamResponse.status === 401
+       upstreamResponse.status === 401 && !wasPermanentAccountRefusal(upstreamResponse)
        && isOAuth401ReplayProvider
         && transportState.sentOAuthSnapshot
         && !oauth401ReplayAttempted
@@ -978,7 +994,7 @@ export async function prepareAdapterExchange(
               const adapterOwnsDispatch = transportState.activeAdapter.fetchResponse !== undefined;
               const hop = reserveCredentialHop("auth-recovery",
                 `${route.providerName}|${route.modelId}|terminal-refresh-account`,
-                !adapterOwnsDispatch && (transientRetryPolicyFor(route.provider) !== null || resetReplayPolicyFor(route.provider) !== null));
+                !adapterOwnsDispatch && genericRefetchCountsSends());
               if (hop.allowed) {
                 try {
                   const admitted = await applyFailoverSnapshot(alternate);
@@ -1062,7 +1078,7 @@ export async function prepareAdapterExchange(
         continue recovery;
       }
 
-      if (route.providerName === "google-antigravity" && upstreamResponse.status === 401
+      if (route.providerName === "google-antigravity" && upstreamResponse.status === 401 && !wasPermanentAccountRefusal(upstreamResponse)
         && oauth401ReplayAttempted && !antigravityAuthRotationAttempted
         && !isNonReplayableResponse(upstreamResponse)) {
         antigravityAuthRotationAttempted = true;
@@ -1348,12 +1364,31 @@ export async function prepareAdapterExchange(
         && anthropicInstance
         && transportState.anthropicPoolAccountId
       ) {
-        const nextAccountId = await rotateAnthropicAccountOnResponseForInstance(anthropicInstance, upstreamResponse, {
-          config, accountId: transportState.anthropicPoolAccountId, sessionKey: anthropicSessionKey,
-          model: route.modelId, requestKey: transportState, decision: transportState.anthropicRouteDecision, currentDecision: currentAnthropicRouteDecision, signal: upstream.signal,
-          canRetry: !sendBudgetExhausted() && transportState.anthropicPoolFailovers < ANTHROPIC_POOL_MAX_FAILOVERS_PER_REQUEST,
-        });
-        if (!nextAccountId) break;
+        // Same admission as the other rotation ladders: the roster cap cannot see the shared
+        // send budget. A canRetry term, not a break -- the call also records the refusal.
+        const adapterOwnsDispatch = transportState.activeAdapter.fetchResponse !== undefined;
+        const hop = reserveCredentialHop(
+          "auth-recovery",
+          `${route.providerName}|${route.modelId}|anthropic-pool-rotation`,
+          !adapterOwnsDispatch && genericRefetchCountsSends(),
+        );
+        let nextAccountId: string | null;
+        try {
+          nextAccountId = await rotateAnthropicAccountOnResponseForInstance(anthropicInstance, upstreamResponse, {
+            config, accountId: transportState.anthropicPoolAccountId, sessionKey: anthropicSessionKey,
+            model: route.modelId, requestKey: transportState, decision: transportState.anthropicRouteDecision, currentDecision: currentAnthropicRouteDecision, signal: upstream.signal,
+            // A denied reservation must not rotate on base headroom its bounds cannot see.
+            canRetry: hop.allowed && (hop.permit !== undefined || !sendBudgetExhausted())
+              && transportState.anthropicPoolFailovers < ANTHROPIC_POOL_MAX_FAILOVERS_PER_REQUEST,
+          });
+        } catch (rotationError) {
+          hop.permit?.release();
+          throw rotationError;
+        }
+        if (!nextAccountId) {
+          hop.permit?.release();
+          break;
+        }
         try {
           const admitted = await commitResolvedOAuthSelection(await anthropicRoutingFor(anthropicInstance).getAnthropicPoolAccessSnapshot(nextAccountId));
           if (!admitted) throw new Error("OAuth selection changed during recovery");
@@ -1376,12 +1411,26 @@ export async function prepareAdapterExchange(
           });
           sealRequestAttemptIdentity(logCtx.activeAttempt, logCtx.provider, transportState.activeAdapter.name, logCtx.accountLogLabel);
           recordAttemptCredentialSource(logCtx.activeAttempt, route.providerName, route.provider, transportState.activeAdapter.name);
-         const result = await rebuildAndRefetch(upstreamResponse.status === 401 ? "oauth-401"
-           : upstreamResponse.status === 403 ? "oauth-account-403" : "anthropic-oauth-429");
-         if ("failed" in result) return result.failed;
-         upstreamResponse = result;
+          // The replay IS this hop's send; the dispatching layer settles the reservation.
+          sendBudgetState.pendingHopPermit = hop.permit;
+          let result: Response | { failed: Response };
+          try {
+            result = await rebuildAndRefetch(upstreamResponse.status === 401 ? "oauth-401"
+              : upstreamResponse.status === 403 ? "oauth-account-403" : "anthropic-oauth-429",
+              () => { if (!adapterOwnsDispatch) hop.permit?.use(); });
+          } finally {
+            sendBudgetState.pendingHopPermit = undefined;
+          }
+          if ("failed" in result) {
+            // A no-op once the boundary was reached; a refund when the rebuild died first.
+            hop.permit?.release();
+            return result.failed;
+          }
+          upstreamResponse = result;
           if (isNonReplayableResponse(upstreamResponse)) continue recovery;
-       } catch {
+        } catch {
+          // release() refunds a pre-send hop and is a no-op once the send settled it.
+          hop.permit?.release();
           break;
         }
       }
@@ -1421,9 +1470,9 @@ export async function prepareAdapterExchange(
         const hop = reserveCredentialHop(
           "auth-recovery",
           `${route.providerName}|${route.modelId}|adapter-recovery-oauth-429`,
-          // Transient and opted-in reset helpers report physical sends; an adapter-owned
+          // Counted generic helpers report physical sends; an adapter-owned
           // ladder settles its own booking.
-          !adapterOwnsDispatch && (transientRetryPolicyFor(route.provider) !== null || resetReplayPolicyFor(route.provider) !== null),
+          !adapterOwnsDispatch && genericRefetchCountsSends(),
         );
         if (!hop.allowed) break;
         const nextAccountId = rotateGenericOAuthAccountOnRefusal(
@@ -1525,9 +1574,9 @@ export async function prepareAdapterExchange(
         const hop = reserveCredentialHop(
           "auth-recovery",
           `${route.providerName}|${route.modelId}|adapter-recovery-oauth-429`,
-          // Transient and opted-in reset helpers report physical sends; an adapter-owned
+          // Counted generic helpers report physical sends; an adapter-owned
           // ladder settles its own booking.
-          !adapterOwnsDispatch && (transientRetryPolicyFor(route.provider) !== null || resetReplayPolicyFor(route.provider) !== null),
+          !adapterOwnsDispatch && genericRefetchCountsSends(),
         );
         if (!hop.allowed) break;
         const nextAccountId = rotateGenericOAuthAccountOn429(
@@ -1646,7 +1695,7 @@ export async function prepareAdapterExchange(
         const hop = reserveCredentialHop(
           "auth-recovery",
           `${route.providerName}|${route.modelId}|adapter-recovery-oauth-verify`,
-          !adapterOwnsDispatch && (transientRetryPolicyFor(route.provider) !== null || resetReplayPolicyFor(route.provider) !== null),
+          !adapterOwnsDispatch && genericRefetchCountsSends(),
         );
         if (!hop.allowed) break;
         // Auth-refusal rotation, not the rate-limit one: a verification refusal

package/src/server/responses/combo-session-recall.ts

@@ -110,6 +110,31 @@ export function recallComboForLane(
   return entry.responseModel === model ? entry.comboId : undefined;
 }
 
+/**
+ * The concrete target that completed the lane's last turn for `comboId`, under the same
+ * freshness and ownership rules as `recallComboForLane`. A lane answered by another combo, or
+ * whose target left the combo or was disabled, recalls nothing.
+ */
+export function recallComboTargetForLane(
+  config: OcxConfig,
+  lane: string | undefined,
+  comboId: string,
+): Pick<OcxComboTarget, "provider" | "model"> | undefined {
+  if (!lane || !comboId) return undefined;
+  const entry = recall.get(lane);
+  if (!entry || entry.comboId !== comboId) return undefined;
+  const combo = getCombo(config, entry.comboId);
+  const provider = config.providers[entry.target.provider];
+  if (Date.now() - entry.at >= RECALL_TTL_MS
+    || !Object.hasOwn(config.providers, entry.target.provider)
+    || !provider || provider.disabled === true
+    || !combo?.targets.some(target => targetKey(target) === targetKey(entry.target))) {
+    deleteEntry(lane);
+    return undefined;
+  }
+  return { provider: entry.target.provider, model: entry.target.model };
+}
+
 /**
  * Periodic expiry. Without it a lane that is never read again and never touched by a config
  * reconciliation holds its entry for the life of the process: the existing TTL is only

package/src/server/responses/combo-stream-preflight.ts

@@ -115,12 +115,14 @@ function retryableZeroOutputTerminal(payload: unknown): boolean {
   if (!payload || typeof payload !== "object" || Array.isArray(payload)) return false;
   const event = payload as {
     type?: unknown;
-    response?: { incomplete_details?: { reason?: unknown } };
+    response?: { incomplete_details?: { reason?: unknown }; output?: unknown };
   };
   if (bareErrorIsRetryable(event)) return true;
   if (event.type === "response.failed") return true;
   if (event.type !== "response.incomplete") return false;
   const reason = event.response?.incomplete_details?.reason;
+  // A terminal-only upstream can carry output without preceding delta/item events.
+  if (reason === undefined) return !Array.isArray(event.response?.output) || event.response.output.length === 0;
   return typeof reason === "string" && RETRYABLE_ZERO_OUTPUT_INCOMPLETE_REASONS.has(reason);
 }
 

package/src/server/responses/combo-tool-loop-pin.ts

@@ -0,0 +1,70 @@
+/**
+ * Combo target pinning for an in-flight tool loop (#7298).
+ *
+ * A combo picks a target per request. A tool loop is several requests that together form one
+ * assistant turn: the model emits a call, the client runs it, and the next request carries the
+ * result back. Re-picking between those requests hands the second half of a turn to a model that
+ * did not produce the first half. That is wasteful at best (a cold prompt cache, a different
+ * plan) and a hard failure at worst: DeepSeek's thinking mode refuses a tool continuation whose
+ * call carries no reasoning it can read, and a GPT turn never leaves one.
+ *
+ * So a continuation whose input ends in a tool output keeps the target that completed the lane's
+ * previous turn, as long as that target is still selectable right now: enabled, still in the
+ * combo, out of cooldown and passing every per-request eligibility filter the caller supplies.
+ * When it is not, normal selection (and normal failover) runs, and the strict serializers make
+ * the foreign history acceptable instead: see `backfillBareToolCallTurnReasoning` and the Chat
+ * adapter's reasoning placeholder.
+ *
+ * A new user message is never pinned; re-deciding per user turn is what a combo is for.
+ */
+import { pickComboTarget, targetKey, type ComboPick } from "../../combos";
+import type { NormalizedComboTarget } from "../../combos/types";
+import type { OcxConfig } from "../../types";
+import { recallComboTargetForLane } from "./combo-session-recall";
+
+export const COMBO_TOOL_LOOP_PIN_REASON = "combo-tool-loop-pin";
+
+function isToolOutputItem(item: unknown): boolean {
+  if (!item || typeof item !== "object" || Array.isArray(item)) return false;
+  const type = (item as { type?: unknown }).type;
+  return typeof type === "string" && type.endsWith("_output");
+}
+
+/**
+ * Whether this request resumes an assistant turn by returning tool results: its Responses
+ * `input` (or Chat `messages`) ends with a tool output.
+ */
+export function isToolLoopContinuation(body: unknown): boolean {
+  if (!body || typeof body !== "object" || Array.isArray(body)) return false;
+  const { input, messages } = body as { input?: unknown; messages?: unknown };
+  if (Array.isArray(input)) return input.length > 0 && isToolOutputItem(input[input.length - 1]);
+  if (Array.isArray(messages) && messages.length > 0) {
+    const last = messages[messages.length - 1];
+    return !!last && typeof last === "object" && (last as { role?: unknown }).role === "tool";
+  }
+  return false;
+}
+
+/**
+ * The pick that keeps a tool loop on its previous target, or undefined to select normally.
+ * Uses a non-mutating preview so a pinned request never disturbs round-robin rotation state.
+ */
+export function toolLoopPinnedPick(args: {
+  config: OcxConfig;
+  comboId: string;
+  lane: string | undefined;
+  body: unknown;
+  eligible: (target: NormalizedComboTarget) => boolean;
+  now?: number;
+}): ComboPick | undefined {
+  if (!isToolLoopContinuation(args.body)) return undefined;
+  const previous = recallComboTargetForLane(args.config, args.lane, args.comboId);
+  if (!previous) return undefined;
+  const previousKey = targetKey(previous);
+  const pick = pickComboTarget(args.config, args.comboId, {
+    eligible: target => targetKey(target) === previousKey && args.eligible(target),
+    now: args.now,
+    preview: true,
+  });
+  return pick ? { ...pick, attempted: [previousKey] } : undefined;
+}

package/src/server/responses/compact.ts

@@ -1,3 +1,4 @@
+import { captureAccountAutoPauseAttempt, wasAccountAutoPaused, wasPermanentAccountRefusal } from "../../oauth/account-auto-pause";
 import { spendLedgerStorageErrorResponse } from "./spend-storage-error";
 import { createPhysicalSendReporter, rebindPhysicalSend } from "../../lib/request-execution-budget";
 import { createInferenceSendBudget } from "../inference/context";
@@ -1058,6 +1059,7 @@ export async function handleResponsesCompact(
     ) => {
       // A gateway response is not evidence about any Codex account.
       if (gatewayTransport || !usesCodexForwardPoolAuth(ctx, route.provider)) return;
+      if ((outcome === 401 || outcome === 403) && upstream && wasPermanentAccountRefusal(upstream)) return;
       recordCodexUpstreamOutcome(config, ctx.accountId, outcome, {
         ...meta,
         threadId: ctx.kind === "pool" || ctx.kind === "main-pool" ? ctx.affinityKey : undefined,
@@ -1166,6 +1168,7 @@ export async function handleResponsesCompact(
         providerFetch(sendProvider, undefined, {
           providerName: route.providerName,
           modelId: route.modelId,
+          accountAutoPause: (name, provider, headers) => captureAccountAutoPauseAttempt(name, provider, { config, codex: sendAuthCtx }, headers),
           beforeDispatch: isCanonicalOpenAiForwardProvider(sendProvider)
             ? createCodexAuthDispatchGuard(sendAuthCtx, config, selectedModelId, admission) : undefined,
           dispatchOverride: sendProvider.authMode !== "forward" && sendProvider.apiKey
@@ -1287,7 +1290,7 @@ export async function handleResponsesCompact(
 
     recordDistributedResponseFailure(upstream, config, route.providerName, req.signal);
     if (
-      upstream.status === 401
+      upstream.status === 401 && !wasPermanentAccountRefusal(upstream)
       && (authCtx.kind === "main-pool" || authCtx.kind === "pool")
       && !gatewayTransport
       && usesCodexForwardPoolAuth(authCtx, compactProvider)
@@ -1411,12 +1414,12 @@ export async function handleResponsesCompact(
     // surfaces to the client, which retries the compact task OUTSIDE the logical request
     // — reporting exhausted retries while another pool account sat idle (#913).
     if (
-      (upstream.status === 429 || upstream.status === 402)
+      (upstream.status === 429 || upstream.status === 402 || wasAccountAutoPaused(upstream))
       // A replay refusal this proxy synthesized carries 429 for the client's benefit only.
       // It is not pool quota evidence, and the alternate account below is another send of a
       // compact turn that may already have been processed.
       && !isNonReplayableResponse(upstream)
-      && !storedPool401ReplayAttempted
+      && (!storedPool401ReplayAttempted || wasAccountAutoPaused(upstream))
       && !gatewayTransport
       && usesCodexForwardPoolAuth(authCtx, route.provider)
       && !authCtx.fixedAccount
@@ -1521,7 +1524,7 @@ export async function handleResponsesCompact(
     // the same guards as the quota move, and a 5xx whose body confirms quota keeps today's path.
     if (
       shouldRetryCodexPoolAccountTransient(upstream)
-      && !storedPool401ReplayAttempted
+      && (!storedPool401ReplayAttempted || wasAccountAutoPaused(upstream))
       && !gatewayTransport
       && usesCodexForwardPoolAuth(authCtx, route.provider)
       && !authCtx.fixedAccount
@@ -1642,7 +1645,7 @@ export async function handleResponsesCompact(
       inspectResponseLogJson(logCtx, compactText);
       if (!options.compactionRoutingOverride) forgetCompactHandoffRoute(req, admission);
       rememberServingConversationStateIssuer(outcomeCtx, codexPoolAffinityKey(req.headers));
-    } else if (!options.compactionRoutingOverride && quotaFailure && !storedPool401ReplayAttempted) {
+    } else if (!options.compactionRoutingOverride && quotaFailure && (!storedPool401ReplayAttempted || wasAccountAutoPaused(upstream))) {
       const fallbackModel = compactHandoffRoute(req, admission, raw.model);
       if (fallbackModel && !req.signal.aborted) {
         const fallbackReq = new Request(req.url, {