replicas-engine 0.1.776 → 0.1.778

package/dist/src/{asp-host-HKSTCW5T.js → asp-host-LDNP6IN6.js}

@@ -5,11 +5,11 @@ import {
   getCodexAspHost,
   restartCodexAspHost,
   restartCodexAspHostIfRunning
-} from "./chunk-OVQGTUPL.js";
-import "./chunk-NPFAPMJ3.js";
-import "./chunk-RP4LIWKF.js";
+} from "./chunk-PRDBLI2T.js";
+import "./chunk-Z3ZWANDZ.js";
+import "./chunk-RXBUE7A3.js";
 import "./chunk-UZSNFLDQ.js";
-import "./chunk-3PHLWZEX.js";
+import "./chunk-7WLA6QVZ.js";
 import "./chunk-VEQXQN22.js";
 export {
   getCodexAspHost,

package/dist/src/{chunk-3PHLWZEX.js → chunk-7WLA6QVZ.js}

@@ -5993,6 +5993,9 @@ var CODEX_CATEGORY_COLORS = {
   input: "#3eeba3",
   output: "#56b6c2"
 };
+function isRecord2(value) {
+  return typeof value === "object" && value !== null;
+}
 function numberOrNull(value) {
   return typeof value === "number" && Number.isFinite(value) ? value : null;
 }
@@ -6072,6 +6075,123 @@ function buildCodexTokenUsageContextUsagePayload(usage) {
     updatedAt: usage.updatedAt
   };
 }
+function normalizeProvider(provider) {
+  return provider === "relay" ? "relay" : provider;
+}
+function resolveClaudeContextWindow(provider, model, reportedMaxTokens) {
+  if (provider !== "claude" && provider !== "relay") {
+    return { maxTokens: reportedMaxTokens, usedKnownWindow: false };
+  }
+  const knownWindow = getClaudeModelContextWindow(model);
+  return {
+    maxTokens: knownWindow ?? reportedMaxTokens,
+    usedKnownWindow: knownWindow !== null
+  };
+}
+function coerceContextUsagePayload(payload, timestamp) {
+  const provider = payload.provider;
+  if (typeof provider !== "string" || !isValidAgentProvider(provider)) return null;
+  const rawTotalTokens = numberOrNull(payload.totalTokens);
+  const percentageValue = numberOrNull(payload.percentage);
+  if (rawTotalTokens === null || percentageValue === null) {
+    return null;
+  }
+  const model = typeof payload.model === "string" ? payload.model : null;
+  const reportedMaxTokens = numberOrNull(payload.maxTokens);
+  const { maxTokens, usedKnownWindow } = resolveClaudeContextWindow(provider, model, reportedMaxTokens);
+  const rawMaxTokens = usedKnownWindow ? maxTokens : numberOrNull(payload.rawMaxTokens);
+  const clamped = clampTokensToWindow(rawTotalTokens, maxTokens);
+  const totalTokens = clamped.totalTokens;
+  const payloadTotalProcessedTokens = numberOrNull(payload.totalProcessedTokens);
+  const totalProcessedTokens = payloadTotalProcessedTokens !== null ? payloadTotalProcessedTokens : clamped.totalProcessedTokens ?? null;
+  const source = payload.source === "codex_token_count" || payload.source === "claude_usage" || payload.source === "provider_usage" ? payload.source : "claude_context";
+  const categories = Array.isArray(payload.categories) ? compactCategories(payload.categories.map((item) => {
+    if (!isRecord2(item)) return null;
+    const name = typeof item.name === "string" ? item.name : null;
+    const tokens = numberOrNull(item.tokens);
+    const categoryPercentage = numberOrNull(item.percentage);
+    if (!name || tokens === null || categoryPercentage === null) return null;
+    return {
+      name,
+      tokens,
+      percentage: usedKnownWindow ? percentage(tokens, maxTokens) : categoryPercentage,
+      ...typeof item.color === "string" ? { color: item.color } : {},
+      ...typeof item.isDeferred === "boolean" ? { isDeferred: item.isDeferred } : {}
+    };
+  })) : [];
+  const apiUsage = isRecord2(payload.apiUsage) ? {
+    ...numberOrNull(payload.apiUsage.inputTokens) !== null ? { inputTokens: numberOrNull(payload.apiUsage.inputTokens) } : {},
+    ...numberOrNull(payload.apiUsage.outputTokens) !== null ? { outputTokens: numberOrNull(payload.apiUsage.outputTokens) } : {},
+    ...numberOrNull(payload.apiUsage.cacheCreationInputTokens) !== null ? { cacheCreationInputTokens: numberOrNull(payload.apiUsage.cacheCreationInputTokens) } : {},
+    ...numberOrNull(payload.apiUsage.cacheReadInputTokens) !== null ? { cacheReadInputTokens: numberOrNull(payload.apiUsage.cacheReadInputTokens) } : {},
+    ...numberOrNull(payload.apiUsage.cachedInputTokens) !== null ? { cachedInputTokens: numberOrNull(payload.apiUsage.cachedInputTokens) } : {},
+    ...numberOrNull(payload.apiUsage.reasoningOutputTokens) !== null ? { reasoningOutputTokens: numberOrNull(payload.apiUsage.reasoningOutputTokens) } : {},
+    ...numberOrNull(payload.apiUsage.totalTokens) !== null ? { totalTokens: numberOrNull(payload.apiUsage.totalTokens) } : {}
+  } : null;
+  const compactsAutomatically = typeof payload.compactsAutomatically === "boolean" ? payload.compactsAutomatically : void 0;
+  const autoCompactThreshold = normalizeAutoCompactThreshold(payload.autoCompactThreshold, maxTokens);
+  return {
+    provider,
+    source,
+    ...model ? { model } : {},
+    totalTokens,
+    ...totalProcessedTokens !== null ? { totalProcessedTokens } : {},
+    maxTokens,
+    rawMaxTokens,
+    percentage: usedKnownWindow ? percentage(totalTokens, maxTokens) : clampPercentage(percentageValue),
+    ...compactsAutomatically !== void 0 ? { compactsAutomatically } : {},
+    ...autoCompactThreshold !== null ? { autoCompactThreshold } : {},
+    categories,
+    apiUsage,
+    updatedAt: typeof payload.updatedAt === "string" ? payload.updatedAt : timestamp
+  };
+}
+function codexTokenCountToContextUsage(event) {
+  const payload = event.payload;
+  if (payload.type !== "token_count") return null;
+  const info = isRecord2(payload.info) ? payload.info : null;
+  if (!info) return null;
+  const lastUsage = isRecord2(info.last_token_usage) ? info.last_token_usage : null;
+  if (!lastUsage) return null;
+  const totalUsage = isRecord2(info.total_token_usage) ? info.total_token_usage : null;
+  const maxTokens = numberOrNull(info.model_context_window);
+  const inputTokens = numberOrNull(lastUsage.input_tokens) ?? 0;
+  const outputTokens = numberOrNull(lastUsage.output_tokens) ?? 0;
+  const cumulativeTotalTokens = numberOrNull(totalUsage?.total_tokens);
+  return buildCodexTokenUsageContextUsagePayload({
+    last: {
+      inputTokens,
+      outputTokens,
+      totalTokens: numberOrNull(lastUsage.total_tokens) ?? inputTokens + outputTokens,
+      cachedInputTokens: numberOrNull(lastUsage.cached_input_tokens),
+      reasoningOutputTokens: numberOrNull(lastUsage.reasoning_output_tokens)
+    },
+    total: cumulativeTotalTokens !== null ? { totalTokens: cumulativeTotalTokens } : null,
+    modelContextWindow: maxTokens,
+    updatedAt: event.timestamp
+  });
+}
+function extractLatestContextUsage(events, provider) {
+  if (!events || events.length === 0) return null;
+  let latest = null;
+  for (const event of events) {
+    if (event.type === CONTEXT_USAGE_EVENT_TYPE) {
+      const usage = coerceContextUsagePayload(event.payload, event.timestamp);
+      if (usage && usage.provider === normalizeProvider(provider)) {
+        latest = usage;
+      }
+      continue;
+    }
+    if (provider === "codex") {
+      const usage = codexTokenCountToContextUsage(event);
+      if (usage) {
+        latest = usage;
+      }
+      continue;
+    }
+  }
+  return latest;
+}
 
 // ../shared/src/pricing.ts
 var AUTOMATED_SOURCES = ["api", "automation"];
@@ -6387,7 +6507,7 @@ var MAX_WARM_HOOK_TIMEOUT_MS = 120 * 60 * 1e3;
 var DEFAULT_START_HOOK_TIMEOUT_MS = 5 * 60 * 1e3;
 var DEFAULT_HOOK_OUTPUT_PREVIEW_CHARS = 1e5;
 var HOOK_EXEC_MAX_BUFFER_BYTES = 10 * 1024 * 1024;
-function isRecord2(value) {
+function isRecord3(value) {
   return typeof value === "object" && value !== null;
 }
 function clampWarmHookTimeoutMs(timeoutMs) {
@@ -6419,7 +6539,7 @@ function parseWarmHookConfig(value, filename = "replicas.json") {
   if (typeof value === "string") {
     return value;
   }
-  if (!isRecord2(value)) {
+  if (!isRecord3(value)) {
     throw new Error(`Invalid ${filename}: "warmHook" must be a string or object`);
   }
   if (!Array.isArray(value.commands) || !value.commands.every((entry) => typeof entry === "string")) {
@@ -6446,11 +6566,11 @@ function resolveWarmHookConfig(value) {
 
 // ../shared/src/replicas-config.ts
 var REPLICAS_CONFIG_FILENAMES = ["replicas.json", "replicas.yaml", "replicas.yml"];
-function isRecord3(value) {
+function isRecord4(value) {
   return typeof value === "object" && value !== null;
 }
 function parseReplicasConfig(value, filename = "replicas.json") {
-  if (!isRecord3(value)) {
+  if (!isRecord4(value)) {
     throw new Error(`Invalid ${filename}: expected an object`);
   }
   const config = {};
@@ -6467,7 +6587,7 @@ function parseReplicasConfig(value, filename = "replicas.json") {
     config.systemPrompt = value.systemPrompt;
   }
   if ("startHook" in value) {
-    if (!isRecord3(value.startHook)) {
+    if (!isRecord4(value.startHook)) {
       throw new Error(`Invalid ${filename}: "startHook" must be an object with "commands" array`);
     }
     const { commands, timeout, separate } = value.startHook;
@@ -7629,6 +7749,7 @@ export {
   normalizeAutoCompactThreshold,
   clampTokensToWindow,
   buildCodexTokenUsageContextUsagePayload,
+  extractLatestContextUsage,
   fetchModelsDevCatalog,
   SANDBOX_PATHS,
   REPLICAS_RUNTIME_ENV_ALIASES,

package/dist/src/{chunk-GICTTMFH.js → chunk-DTXZFR5B.js}

@@ -3,12 +3,12 @@ import { createRequire as __createRequire } from 'node:module';
 const require = __createRequire(import.meta.url);
 import {
   monolithRequest
-} from "./chunk-NPFAPMJ3.js";
+} from "./chunk-Z3ZWANDZ.js";
 import {
   extractCommandProtectionCommandText,
   findGitCommitSignals,
   findPrMergeSignals
-} from "./chunk-3PHLWZEX.js";
+} from "./chunk-7WLA6QVZ.js";
 
 // src/services/command-protection-service.ts
 var DEFAULT_COMMAND_PROTECTION_BLOCK_MESSAGE = "Blocked by Replicas command protection.";

package/dist/src/{chunk-OVQGTUPL.js → chunk-PRDBLI2T.js}

@@ -5,18 +5,18 @@ import {
   ENGINE_ENV,
   monolithRequest,
   setAgentCredentialSnapshot
-} from "./chunk-NPFAPMJ3.js";
+} from "./chunk-Z3ZWANDZ.js";
 import {
   AppServerProcess,
   buildCodexAgentEnv
-} from "./chunk-RP4LIWKF.js";
+} from "./chunk-RXBUE7A3.js";
 import {
   CODEX_AUTH_ENV_KEYS,
   CODEX_AUTH_ENV_KEYS_BY_METHOD,
   codexAuthEnvFromResponse,
   createErrorResult,
   createSuccessResult
-} from "./chunk-3PHLWZEX.js";
+} from "./chunk-7WLA6QVZ.js";
 
 // src/managers/codex-token-manager.ts
 import { promises as fs } from "fs";
@@ -232,7 +232,7 @@ var CodexTokenManager = class extends BaseRefreshManager {
     const data = await response.json();
     await this.applyCredentialsResponse(data);
     if (restartOnChange && CODEX_AUTH_ENV_KEYS.some((key, index) => process.env[key] !== previousEnv[index])) {
-      const { restartCodexAspHostIfRunning: restartCodexAspHostIfRunning2 } = await import("./asp-host-HKSTCW5T.js");
+      const { restartCodexAspHostIfRunning: restartCodexAspHostIfRunning2 } = await import("./asp-host-LDNP6IN6.js");
       await restartCodexAspHostIfRunning2();
     }
     if (data.scope) {

package/dist/src/{chunk-RP4LIWKF.js → chunk-RXBUE7A3.js}

@@ -4,7 +4,7 @@ const require = __createRequire(import.meta.url);
 import {
   HOOK_EXEC_MAX_BUFFER_BYTES,
   isVersionBelow
-} from "./chunk-3PHLWZEX.js";
+} from "./chunk-7WLA6QVZ.js";
 
 // src/utils/codex-agent-env.ts
 function buildCodexAgentEnv(source = process.env) {
@@ -241,7 +241,7 @@ var DEFAULT_CODEX_ARGS = [
 var MIN_CODEX_CLI_VERSION = "0.153.3";
 var CODEX_UPGRADE_TIMEOUT_MS = 12e4;
 var codexCliVersionEnsured = null;
-var ENGINE_PACKAGE_VERSION = "0.1.776";
+var ENGINE_PACKAGE_VERSION = "0.1.778";
 var INITIALIZE_METHOD = "initialize";
 var INITIALIZED_NOTIFICATION = "initialized";
 var ACCOUNT_LOGIN_START_METHOD = "account/login/start";

package/dist/src/{chunk-OX5DTL5R.js → chunk-XAJWJBND.js}

@@ -4,7 +4,7 @@ const require = __createRequire(import.meta.url);
 import {
   findCodeHostPullRequestUrls,
   mayCreatePullRequest
-} from "./chunk-3PHLWZEX.js";
+} from "./chunk-7WLA6QVZ.js";
 
 // src/services/post-tool-pr-notifier.ts
 async function notifyPostToolUse(toolName, toolInput, toolResult) {

package/dist/src/{chunk-NPFAPMJ3.js → chunk-Z3ZWANDZ.js}

@@ -15,7 +15,7 @@ import {
   isValidRelayBaseProvider,
   parsePosixEnvFile,
   readReplicasRuntimeEnv
-} from "./chunk-3PHLWZEX.js";
+} from "./chunk-7WLA6QVZ.js";
 
 // src/engine-env.ts
 import { readFileSync as readFileSync2 } from "fs";

package/dist/src/command-protection-hook.js

@@ -3,12 +3,12 @@ import { createRequire as __createRequire } from 'node:module';
 const require = __createRequire(import.meta.url);
 import {
   evaluateCommandProtection
-} from "./chunk-GICTTMFH.js";
-import "./chunk-NPFAPMJ3.js";
+} from "./chunk-DTXZFR5B.js";
+import "./chunk-Z3ZWANDZ.js";
 import {
   isRecord
 } from "./chunk-UZSNFLDQ.js";
-import "./chunk-3PHLWZEX.js";
+import "./chunk-7WLA6QVZ.js";
 import "./chunk-VEQXQN22.js";
 
 // src/command-protection-hook.ts

package/dist/src/deepseek-command-protection-plugin.js

@@ -3,13 +3,13 @@ import { createRequire as __createRequire } from 'node:module';
 const require = __createRequire(import.meta.url);
 import {
   evaluateCommandProtection
-} from "./chunk-GICTTMFH.js";
+} from "./chunk-DTXZFR5B.js";
 import {
   notifyPostToolUse
-} from "./chunk-OX5DTL5R.js";
-import "./chunk-NPFAPMJ3.js";
+} from "./chunk-XAJWJBND.js";
+import "./chunk-Z3ZWANDZ.js";
 import "./chunk-UZSNFLDQ.js";
-import "./chunk-3PHLWZEX.js";
+import "./chunk-7WLA6QVZ.js";
 import "./chunk-VEQXQN22.js";
 
 // src/deepseek-command-protection-plugin.ts

package/dist/src/headless-agent.js

@@ -8,12 +8,12 @@ import {
 import {
   AppServerProcess,
   buildCodexAgentEnv
-} from "./chunk-RP4LIWKF.js";
+} from "./chunk-RXBUE7A3.js";
 import {
   AGENT,
   getMemoryOutputSafetyViolation,
   headlessAgentRequestSchema
-} from "./chunk-3PHLWZEX.js";
+} from "./chunk-7WLA6QVZ.js";
 import "./chunk-VEQXQN22.js";
 
 // src/headless-agent.ts

package/dist/src/index.js

@@ -91,7 +91,7 @@ import {
   evaluateCommandProtection,
   extractToolCommand,
   reportCommandProtectionBlock
-} from "./chunk-GICTTMFH.js";
+} from "./chunk-DTXZFR5B.js";
 import {
   ACCOUNT_RATE_LIMITS_UPDATED_METHOD,
   AGENT_MESSAGE_DELTA_METHOD,
@@ -124,20 +124,20 @@ import {
   recordCredentialFallback,
   recordExhaustedCredential,
   restartCodexAspHost
-} from "./chunk-OVQGTUPL.js";
+} from "./chunk-PRDBLI2T.js";
 import {
   ENGINE_ENV,
   IS_WARMING_MODE,
   monolithRequest,
   monolithService,
   setAgentCredentialSnapshot
-} from "./chunk-NPFAPMJ3.js";
+} from "./chunk-Z3ZWANDZ.js";
 import {
   AspClient,
   SUBPROCESS_MAX_BUFFER,
   execAsync,
   execFileAsync
-} from "./chunk-RP4LIWKF.js";
+} from "./chunk-RXBUE7A3.js";
 import {
   isRecord as isRecord2
 } from "./chunk-UZSNFLDQ.js";
@@ -252,6 +252,7 @@ import {
   detectLanguageByPath,
   excludeQueuedAcceptedEvents,
   extractErrorText,
+  extractLatestContextUsage,
   extractToolResultText,
   fetchAiGatewayModels,
   fetchModelsDevCatalog,
@@ -337,7 +338,7 @@ import {
   spawnRelaySubagentRequestSchema,
   stripAgentDiagnosticErrors,
   withTimeout
-} from "./chunk-3PHLWZEX.js";
+} from "./chunk-7WLA6QVZ.js";
 import {
   __commonJS,
   __export,
@@ -10655,7 +10656,7 @@ import { join as join37 } from "path";
 import { randomUUID as randomUUID13 } from "crypto";
 
 // ../shared/src/workspace-chat.ts
-var WORKSPACE_CHATS_QUERY_SCHEMA_VERSION = 2;
+var WORKSPACE_CHATS_QUERY_SCHEMA_VERSION = 3;
 var WORKSPACE_CHATS_QUERY_ROOT = `workspace-chats:v${WORKSPACE_CHATS_QUERY_SCHEMA_VERSION}`;
 function isAcceptedUserMessage(event) {
   return getUserMessage(event) !== null && event.payload.source === ACCEPTED_USER_MESSAGE_SOURCE;
@@ -61368,6 +61369,7 @@ function normalizePersistedChat(chat) {
   return {
     id: chat.id,
     provider: chat.provider,
+    ...typeof chat.model === "string" && chat.model ? { model: chat.model } : {},
     ...chat.provider === "relay" ? { relayBaseProvider: getChatExecutionProvider(chat) } : {},
     title: chat.title,
     createdAt: chat.createdAt,
@@ -61605,9 +61607,14 @@ var ChatService = class {
       const accepting2 = chat.acceptingSendResponses.get(idempotencyKey);
       if (accepting2) return accepting2;
     }
+    const previousAcceptance = chat.lastSendAcceptance;
     const accepting = (async () => {
       chat.acceptingMessages += 1;
       try {
+        await previousAcceptance?.catch(() => {
+        });
+        const savedModel = chat.persisted.model ?? (!request.model ? extractLatestContextUsage((await chat.provider.getHistory()).events, getChatExecutionProvider(chat.persisted))?.model : void 0);
+        request = { ...request, model: request.model ?? savedModel ?? void 0 };
         return await chat.provider.enqueueMessage(
           request,
           (result, acceptedMessage) => this.handleMessageAccepted(
@@ -61622,6 +61629,7 @@ var ChatService = class {
         chat.acceptingMessages -= 1;
       }
     })();
+    chat.lastSendAcceptance = accepting;
     if (!idempotencyKey) return accepting;
     chat.acceptingSendResponses.set(idempotencyKey, accepting);
     try {
@@ -61649,6 +61657,7 @@ var ChatService = class {
     const previousAcceptedEvent = chat.acceptedUserEvents.get(result.messageId);
     const previousPersistedResponses = chat.persisted.acceptedSendResponses;
     const previousLastMessageText = chat.persisted.lastMessageText;
+    const previousModel = chat.persisted.model;
     const previousUpdatedAt = chat.persisted.updatedAt;
     const idempotencyKey = request.idempotencyKey ?? request.messageId;
     if (idempotencyKey) {
@@ -61666,6 +61675,7 @@ var ChatService = class {
       chat.acceptedUserEvents.set(result.messageId, acceptedEvent);
     }
     chat.persisted.lastMessageText = request.message.trim().slice(0, LAST_MESSAGE_PREVIEW_MAX) || null;
+    chat.persisted.model = acceptedMessage.model ?? previousModel;
     chat.persisted.updatedAt = (/* @__PURE__ */ new Date()).toISOString();
     try {
       await this.persistAllChats();
@@ -61679,6 +61689,7 @@ var ChatService = class {
       else chat.acceptedUserEvents.delete(result.messageId);
       chat.persisted.acceptedSendResponses = previousPersistedResponses;
       chat.persisted.lastMessageText = previousLastMessageText;
+      chat.persisted.model = previousModel;
       chat.persisted.updatedAt = previousUpdatedAt;
       throw new ChatMessagePersistenceError(error);
     }
@@ -61701,6 +61712,9 @@ var ChatService = class {
         ...recordedSender ? { sender: recordedSender } : {}
       }
     });
+    if (chat.persisted.model !== previousModel) {
+      await this.publish({ type: "chat.updated", payload: { chat: this.toSummary(chat) } });
+    }
   }
   async appendSender(chatId, sender) {
     try {
@@ -62295,6 +62309,7 @@ var ChatService = class {
     return {
       id: chat.persisted.id,
       provider: chat.persisted.provider,
+      model: chat.persisted.model,
       ...chat.persisted.provider === "relay" ? { relayBaseProvider: getChatExecutionProvider(chat.persisted) } : {},
       title: chat.persisted.title,
       createdAt: chat.persisted.createdAt,

package/dist/src/post-tool-pr-hook.js

@@ -3,11 +3,11 @@ import { createRequire as __createRequire } from 'node:module';
 const require = __createRequire(import.meta.url);
 import {
   notifyPostToolUse
-} from "./chunk-OX5DTL5R.js";
+} from "./chunk-XAJWJBND.js";
 import {
   isRecord
 } from "./chunk-UZSNFLDQ.js";
-import "./chunk-3PHLWZEX.js";
+import "./chunk-7WLA6QVZ.js";
 import "./chunk-VEQXQN22.js";
 
 // src/post-tool-pr-hook.ts

package/dist/src/relay-mcp.js

@@ -4,7 +4,7 @@ const require = __createRequire(import.meta.url);
 import {
   messageRelaySubagentRequestSchema,
   spawnRelaySubagentRequestSchema
-} from "./chunk-3PHLWZEX.js";
+} from "./chunk-7WLA6QVZ.js";
 import "./chunk-VEQXQN22.js";
 
 // src/relay-mcp.ts

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "replicas-engine",
-  "version": "0.1.776",
+  "version": "0.1.778",
   "description": "Lightweight API server for Replicas workspaces",
   "type": "module",
   "main": "dist/src/index.js",