replicas-engine 0.1.1248 → 0.1.1249

package/dist/src/{asp-host-DFIT2PB7.js → asp-host-UBXPKAXA.js}

@@ -8,9 +8,9 @@ import {
   listRunningCodexModels,
   restartCodexAspHost,
   restartCodexAspHostIfRunning
-} from "./chunk-3V6VDRU5.js";
+} from "./chunk-GH5EEUT2.js";
 import "./chunk-6PVW5AKP.js";
-import "./chunk-P3G4OERE.js";
+import "./chunk-5GHRPLRX.js";
 import "./chunk-3C2WTKME.js";
 import "./chunk-UZSNFLDQ.js";
 import "./chunk-7QJWP5ZL.js";

package/dist/src/{chunk-EVNSRGKX.js → chunk-5ANTWWVQ.js}

@@ -3,7 +3,7 @@ import { createRequire as __createRequire } from 'node:module';
 const require = __createRequire(import.meta.url);
 import {
   AspClient
-} from "./chunk-P3G4OERE.js";
+} from "./chunk-5GHRPLRX.js";
 import {
   ASTER_BASE_URL,
   ASTER_PROVIDER,

package/dist/src/{chunk-P3G4OERE.js → chunk-5GHRPLRX.js}

@@ -240,7 +240,7 @@ function getCodexModelProvider(method) {
   if (method === "foundry") return "azure";
   return method === "ollama" || method === "openai-compatible" ? "replicas_openai_compatible" : "openai";
 }
-var ENGINE_PACKAGE_VERSION = "0.1.1248";
+var ENGINE_PACKAGE_VERSION = "0.1.1249";
 var INITIALIZE_METHOD = "initialize";
 var INITIALIZED_NOTIFICATION = "initialized";
 var ACCOUNT_LOGIN_START_METHOD = "account/login/start";

package/dist/src/{chunk-3V6VDRU5.js → chunk-GH5EEUT2.js}

@@ -14,7 +14,7 @@ import {
 import {
   AppServerProcess,
   buildCodexAgentEnv
-} from "./chunk-P3G4OERE.js";
+} from "./chunk-5GHRPLRX.js";
 import {
   SUBPROCESS_MAX_BUFFER,
   execFileAsync
@@ -291,7 +291,7 @@ var CodexTokenManager = class extends BaseRefreshManager {
     await this.applyCredentialUpdate(++this.credentialGeneration, async () => {
       await this.applyCredentialsResponse(data);
       setAgentCredentialSnapshot("codex", data);
-      const { restartCodexAspHostIfRunning: restartCodexAspHostIfRunning2 } = await import("./asp-host-DFIT2PB7.js");
+      const { restartCodexAspHostIfRunning: restartCodexAspHostIfRunning2 } = await import("./asp-host-UBXPKAXA.js");
       await restartCodexAspHostIfRunning2();
     });
     await this.start(true);
@@ -313,7 +313,7 @@ var CodexTokenManager = class extends BaseRefreshManager {
       const accountChanged = ENGINE_ENV.REPLICAS_AGENT_CREDENTIALS.codex?.credentialId !== data.credentialId;
       await this.applyCredentialsResponse(data);
       if (restartOnChange && (accountChanged || CODEX_AUTH_ENV_KEYS.some((key, index) => process.env[key] !== previousEnv[index]))) {
-        const { restartCodexAspHostIfRunning: restartCodexAspHostIfRunning2 } = await import("./asp-host-DFIT2PB7.js");
+        const { restartCodexAspHostIfRunning: restartCodexAspHostIfRunning2 } = await import("./asp-host-UBXPKAXA.js");
         await restartCodexAspHostIfRunning2();
       }
       setAgentCredentialSnapshot("codex", data);

package/dist/src/headless-agent.js

@@ -13,11 +13,11 @@ import {
   recoverCompletedTurn,
   spawnMuseMsp,
   startMuseSession
-} from "./chunk-EVNSRGKX.js";
+} from "./chunk-5ANTWWVQ.js";
 import {
   AppServerProcess,
   buildCodexAgentEnv
-} from "./chunk-P3G4OERE.js";
+} from "./chunk-5GHRPLRX.js";
 import "./chunk-3C2WTKME.js";
 import {
   AGENT,

package/dist/src/index.js

@@ -64,7 +64,7 @@ import {
   spawnMuseMsp,
   startMuseSession,
   uuidV7
-} from "./chunk-EVNSRGKX.js";
+} from "./chunk-5ANTWWVQ.js";
 import {
   publishedPullRequestUrls
 } from "./chunk-FQZ6YTN2.js";
@@ -140,7 +140,7 @@ import {
   timestampFromSeconds,
   upsertCredentialFileLines,
   writeSecureCredentialFile
-} from "./chunk-3V6VDRU5.js";
+} from "./chunk-GH5EEUT2.js";
 import {
   ENGINE_ENV,
   monolithRequest,
@@ -152,7 +152,7 @@ import {
 } from "./chunk-6PVW5AKP.js";
 import {
   getCodexModelProvider
-} from "./chunk-P3G4OERE.js";
+} from "./chunk-5GHRPLRX.js";
 import {
   SUBPROCESS_MAX_BUFFER,
   execAsync,
@@ -8844,7 +8844,7 @@ var ClaudeTokenManager = class extends BaseRefreshManager {
     await this.applyCredentialUpdate(generation, () => this.applyCredentialsResponse(data));
     console.log(`[ClaudeTokenManager] Credentials refreshed (method=${data.type})`);
   }
-  async fetchFreshCredentials(failureReason, failureKind = "rejected", failedCredential = ENGINE_ENV.REPLICAS_AGENT_CREDENTIALS.claude, allowedMethods) {
+  async fetchFreshCredentials(failureReason, failureKind = "rejected", failedCredential = ENGINE_ENV.REPLICAS_AGENT_CREDENTIALS.claude, allowedMethods, resetsAt) {
     const configuredMethod = ENGINE_ENV.REPLICAS_CLAUDE_AUTH_METHOD;
     const failedMethod = failedCredential?.method === "ollama" || failedCredential?.method === "oauth" || failedCredential?.method === "api_key" || failedCredential?.method === "bedrock" || failedCredential?.method === "foundry" ? failedCredential.method : configuredMethod === "ollama" || configuredMethod === "oauth" || configuredMethod === "api_key" || configuredMethod === "bedrock" || configuredMethod === "foundry" ? configuredMethod : void 0;
     return this.swapCredentials({
@@ -8856,6 +8856,7 @@ var ClaudeTokenManager = class extends BaseRefreshManager {
         ...failedCredential?.method === failedMethod ? { failedCredential } : {},
         failureReason,
         failureKind,
+        ...resetsAt ? { resetsAt } : {},
         ...exclusions
       } : exclusions),
       isOauthNow: () => ENGINE_ENV.REPLICAS_CLAUDE_AUTH_METHOD === "oauth"
@@ -11177,12 +11178,12 @@ var AuthFallbackCoordinator = class {
   record;
   context;
   /** `attempted` is the credential the failed turn ran on, which another chat may already have replaced. */
-  fallBack(reason, detail, attempted) {
+  fallBack(reason, detail, attempted, resetsAt) {
     let lock = fallbackLocks.get(this.provider);
     if (!lock) fallbackLocks.set(this.provider, lock = new AsyncLock());
-    return lock.run(() => this.fallBackNow(reason, detail, attempted));
+    return lock.run(() => this.fallBackNow(reason, detail, attempted, resetsAt));
   }
-  async fallBackNow(reason, detail, attempted) {
+  async fallBackNow(reason, detail, attempted, resetsAt) {
     const tokenManager = SWAPPABLE_AGENTS[this.provider];
     const live = ENGINE_ENV.REPLICAS_AGENT_CREDENTIALS[this.provider];
     const exhausted = attempted ?? live;
@@ -11219,7 +11220,8 @@ var AuthFallbackCoordinator = class {
         // Skipped via the exclusion above, not rested, so Settings doesn't show a usage limit.
         reason === "auth_failed" || reason === "credits_required" ? "rejected" : "exhausted",
         exhausted,
-        step.methods.length > 0 ? step.methods : void 0
+        step.methods.length > 0 ? step.methods : void 0,
+        resetsAt
       );
       if (!swap.ok) {
         failed ||= swap.error.code !== "no_credentials";
@@ -11574,9 +11576,9 @@ var CodingAgentManager = class {
       this.getHistorySink()?.append(event);
     }, options.fallbackContext ?? (() => ({ excludeProviders: [], forkable: false }))) : null;
   }
-  async fallBackCredential(reason, detail, attempted) {
+  async fallBackCredential(reason, detail, attempted, resetsAt) {
     if (!this.authFallback) return { outcome: "unavailable" };
-    const result = await this.authFallback.fallBack(reason, detail, attempted);
+    const result = await this.authFallback.fallBack(reason, detail, attempted, resetsAt);
     if (result.outcome === "switched" && result.model) await this.onModelSwitched(result.model);
     return result;
   }
@@ -12806,12 +12808,14 @@ var ClaudeMissingSessionError = class extends Error {
   }
 };
 var ClaudeQuotaError = class extends Error {
-  constructor(message, kind) {
+  constructor(message, kind, resetsAt) {
     super(message);
     this.kind = kind;
+    this.resetsAt = resetsAt;
     this.name = "ClaudeQuotaError";
   }
   kind;
+  resetsAt;
 };
 var MAX_AUTH_RETRIES = 2;
 var MAX_TRANSIENT_RETRIES = 2;
@@ -13260,7 +13264,7 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
             const detail = error instanceof ClaudeQuotaError ? error.message : extractErrorText(error);
             const repeated = !!attemptedCredential && failedCredentials.some((failed) => isSameCredential(failed, attemptedCredential));
             if (attemptedCredential) failedCredentials.push(attemptedCredential);
-            const fallback = repeated ? { outcome: "unavailable" } : await this.fallBackCredential(reason, detail, attemptedCredential);
+            const fallback = repeated ? { outcome: "unavailable" } : await this.fallBackCredential(reason, detail, attemptedCredential, error instanceof ClaudeQuotaError ? error.resetsAt : void 0);
             if (fallback.outcome === "switched") {
               if (fallback.model) currentRequest = { ...currentRequest, model: fallback.model };
               authRetries = 0;
@@ -13673,6 +13677,7 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
     let loopError = null;
     let suppressLinearResponseFlush = false;
     let creditsRequired = false;
+    let resetsAt;
     let turnFailed = false;
     try {
       while (true) {
@@ -13680,10 +13685,15 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
         if (next.done || this.activeQuery !== response) break;
         const msg = next.value;
         if (!msg) break;
-        if (msg.type === "rate_limit_event") creditsRequired = msg.rate_limit_info.errorCode === "credits_required";
+        if (msg.type === "rate_limit_event") {
+          creditsRequired = msg.rate_limit_info.errorCode === "credits_required";
+          if (msg.rate_limit_info.status === "rejected" && msg.rate_limit_info.resetsAt) {
+            resetsAt = new Date(msg.rate_limit_info.resetsAt * 1e3).toISOString();
+          }
+        }
         const authErrorMessage = _ClaudeManager.detectAuthErrorInMessage(msg);
         const transientErrorMessage = _ClaudeManager.detectTransientTurnErrorInMessage(msg);
-        const turnError = authErrorMessage ? new ClaudeAuthError(authErrorMessage) : msg.type === "result" && msg.subtype !== "success" && msg.num_turns === 0 && msg.errors.some((err) => err.startsWith("No conversation found with session ID")) ? new ClaudeMissingSessionError(msg.errors.join("\n")) : _ClaudeManager.detectQuotaErrorInMessage(msg, creditsRequired) ?? (transientErrorMessage ? new ClaudeTransientTurnError(transientErrorMessage, this.pendingTurn?.sawActivity ?? false) : null);
+        const turnError = authErrorMessage ? new ClaudeAuthError(authErrorMessage) : msg.type === "result" && msg.subtype !== "success" && msg.num_turns === 0 && msg.errors.some((err) => err.startsWith("No conversation found with session ID")) ? new ClaudeMissingSessionError(msg.errors.join("\n")) : _ClaudeManager.detectQuotaErrorInMessage(msg, creditsRequired, resetsAt) ?? (transientErrorMessage ? new ClaudeTransientTurnError(transientErrorMessage, this.pendingTurn?.sawActivity ?? false) : null);
         if (turnError) {
           suppressLinearResponseFlush = true;
           turnFailed = true;
@@ -13977,17 +13987,18 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
     if (error instanceof ClaudeQuotaError) return error.kind;
     return detectAgentQuotaLimit(extractErrorText(error));
   }
-  static detectQuotaErrorInMessage(message, creditsRequired) {
+  static detectQuotaErrorInMessage(message, creditsRequired, resetsAt) {
+    const quotaError = (text2, kind2) => new ClaudeQuotaError(text2, creditsRequired ? "credits_required" : kind2, resetsAt);
     if (message.type === "assistant" && "error" in message) {
       const content = Array.isArray(message.message?.content) ? message.message.content.map((block) => "text" in block ? block.text : "").join("\n") : "";
       const text2 = [typeof message.error === "string" ? message.error : "", content].filter(Boolean).join("\n");
       const kind2 = detectAgentQuotaLimit(text2);
-      if (kind2) return new ClaudeQuotaError(text2, creditsRequired ? "credits_required" : kind2);
+      if (kind2) return quotaError(text2, kind2);
     }
     if (message.type !== "result" || !message.is_error) return null;
     const text = message.subtype === "success" ? message.result : message.errors.join("\n");
     const kind = detectAgentQuotaLimit(text);
-    return kind ? new ClaudeQuotaError(text, creditsRequired ? "credits_required" : kind) : null;
+    return kind ? quotaError(text, kind) : null;
   }
   static isTransientTurnError(error) {
     if (error instanceof ClaudeTransientTurnError) return true;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "replicas-engine",
-  "version": "0.1.1248",
+  "version": "0.1.1249",
   "description": "Lightweight API server for Replicas workspaces",
   "type": "module",
   "main": "dist/src/index.js",