replicas-engine 0.1.240 → 0.1.241

package/dist/src/index.js

@@ -151,6 +151,9 @@ function percentage(tokens, maxTokens) {
   if (!maxTokens || maxTokens <= 0) return 0;
   return clampPercentage(tokens / maxTokens * 100);
 }
+function autoCompactThresholdField(autoCompactThreshold) {
+  return typeof autoCompactThreshold === "number" && Number.isFinite(autoCompactThreshold) ? { autoCompactThreshold } : {};
+}
 function clampTokensToWindow(rawTokens, maxTokens) {
   if (!maxTokens || maxTokens <= 0) return { totalTokens: rawTokens };
   if (rawTokens <= maxTokens) return { totalTokens: rawTokens };
@@ -180,6 +183,7 @@ function buildCodexTokenUsageContextUsagePayload(usage) {
     rawMaxTokens: maxTokens,
     percentage: percentage(totalTokens, maxTokens),
     compactsAutomatically: true,
+    ...autoCompactThresholdField(usage.autoCompactThreshold),
     categories: compactCategories([
       {
         name: "Input context",
@@ -286,7 +290,7 @@ var WORKSPACE_SIZES = ["small", "large"];
 var INVALID_WORKSPACE_SIZE_ERROR = `Invalid size: must be one of ${WORKSPACE_SIZES.join(", ")}`;
 
 // ../shared/src/e2b.ts
-var E2B_TEMPLATE_NAME = "replicas-sandbox-2026-05-30-v1";
+var E2B_TEMPLATE_NAME = "replicas-sandbox-2026-05-30-v2";
 
 // ../shared/src/runtime-env.ts
 function parsePosixEnvFile(content) {
@@ -1785,6 +1789,7 @@ var DEFAULT_CHAT_TITLES = {
 };
 var CLAUDE_OPUS_1M_MODEL = "opus[1m]";
 var LEGACY_CLAUDE_OPUS_1M_MODEL = "opus-1m";
+var DEFAULT_CODEX_MODEL = "gpt-5.5";
 function normalizeClaudeModel(model) {
   if (model === LEGACY_CLAUDE_OPUS_1M_MODEL) {
     return CLAUDE_OPUS_1M_MODEL;
@@ -1797,7 +1802,7 @@ var MODEL_LABELS = {
   [CLAUDE_OPUS_1M_MODEL]: "Opus 4.8 (1M)",
   [LEGACY_CLAUDE_OPUS_1M_MODEL]: "Opus 4.8 (1M)",
   haiku: "Haiku 4.5",
-  "gpt-5.5": "GPT-5.5",
+  [DEFAULT_CODEX_MODEL]: "GPT-5.5",
   "gpt-5.4": "GPT-5.4",
   "gpt-5.4-mini": "GPT-5.4 Mini",
   "gpt-5.3-codex": "GPT-5.3 Codex",
@@ -5783,7 +5788,7 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
       rawMaxTokens: Number.isFinite(usage.rawMaxTokens) ? usage.rawMaxTokens : maxTokens,
       percentage: clampPercentage(rawPercentage),
       compactsAutomatically: usage.isAutoCompactEnabled === true,
-      ...typeof usage.autoCompactThreshold === "number" && Number.isFinite(usage.autoCompactThreshold) ? { autoCompactThreshold: usage.autoCompactThreshold } : {},
+      ...autoCompactThresholdField(usage.autoCompactThreshold),
       categories: usage.categories.map((category) => ({
         name: category.name,
         tokens: category.tokens,
@@ -6081,7 +6086,7 @@ var AspClient = class {
 // src/managers/codex-asp/app-server-process.ts
 var DEFAULT_CODEX_BINARY = "codex";
 var DEFAULT_CODEX_ARGS = ["app-server", "--listen", "stdio://"];
-var ENGINE_PACKAGE_VERSION = "0.1.240";
+var ENGINE_PACKAGE_VERSION = "0.1.241";
 var INITIALIZE_METHOD = "initialize";
 var INITIALIZED_NOTIFICATION = "initialized";
 var ACCOUNT_LOGIN_START_METHOD = "account/login/start";
@@ -6348,7 +6353,7 @@ function isCodexAuthError(error) {
 }
 
 // src/managers/codex-asp/mappers.ts
-var DEFAULT_MODEL = "gpt-5.5";
+var DEFAULT_MODEL = DEFAULT_CODEX_MODEL;
 var THREAD_START_METHOD = "thread/start";
 var THREAD_RESUME_METHOD = "thread/resume";
 var THREAD_READ_METHOD = "thread/read";
@@ -6359,6 +6364,24 @@ var TURN_START_METHOD = "turn/start";
 var TURN_INTERRUPT_METHOD = "turn/interrupt";
 var ACCOUNT_RATE_LIMITS_READ_METHOD = "account/rateLimits/read";
 var MAX_CODEX_ASP_TRANSCRIPT_OUTPUT_CHARS = DEFAULT_HOOK_OUTPUT_PREVIEW_CHARS;
+var CODEX_ASP_AUTO_COMPACT_RATIO = 0.85;
+var CODEX_ASP_MODEL_CONTEXT_WINDOWS = {
+  [DEFAULT_MODEL]: 258400
+};
+function codexAspAutoCompactTokenLimitForModel(model) {
+  const contextWindow = CODEX_ASP_MODEL_CONTEXT_WINDOWS[model ?? DEFAULT_MODEL];
+  return contextWindow === void 0 ? void 0 : Math.floor(contextWindow * CODEX_ASP_AUTO_COMPACT_RATIO);
+}
+function codexAspConfigForModel(model) {
+  const tokenLimit = codexAspAutoCompactTokenLimitForModel(model);
+  return {
+    web_search: "live",
+    ...tokenLimit !== void 0 ? {
+      model_auto_compact_token_limit: tokenLimit,
+      model_auto_compact_token_limit_scope: "total"
+    } : {}
+  };
+}
 function toReasoningEffort(thinkingLevel) {
   return codexReasoningEffortForThinkingLevel(thinkingLevel);
 }
@@ -6731,27 +6754,29 @@ function mergeCodexAspTranscripts(primary, supplemental) {
 }
 async function buildThreadStartParams(workingDirectory, request, developerInstructions) {
   const additionalDirectories = await getAgentAdditionalDirectories();
+  const model = request.model ?? DEFAULT_MODEL;
   return {
-    model: request.model ?? DEFAULT_MODEL,
+    model,
     cwd: workingDirectory,
     runtimeWorkspaceRoots: additionalDirectories,
     sandbox: "danger-full-access",
     developerInstructions: developerInstructions ?? null,
-    config: { web_search: "live" },
+    config: codexAspConfigForModel(model),
     experimentalRawEvents: false,
     persistExtendedHistory: false
   };
 }
 async function buildThreadResumeParams(workingDirectory, threadId, request, developerInstructions) {
   const additionalDirectories = await getAgentAdditionalDirectories();
+  const model = request.model ?? DEFAULT_MODEL;
   return {
     threadId,
-    model: request.model ?? DEFAULT_MODEL,
+    model,
     cwd: workingDirectory,
     runtimeWorkspaceRoots: additionalDirectories,
     sandbox: "danger-full-access",
     developerInstructions: developerInstructions ?? null,
-    config: { web_search: "live" },
+    config: codexAspConfigForModel(model),
     excludeTurns: false,
     persistExtendedHistory: false
   };
@@ -7574,6 +7599,7 @@ var CodexAspManager = class extends CodingAgentManager {
     const payload = buildCodexTokenUsageContextUsagePayload({
       model,
       modelContextWindow: tokenUsage.modelContextWindow,
+      autoCompactThreshold: codexAspAutoCompactTokenLimitForModel(model),
       last: {
         inputTokens: tokenUsage.last.inputTokens,
         outputTokens: tokenUsage.last.outputTokens,
@@ -7598,7 +7624,6 @@ import { existsSync as existsSync6 } from "fs";
 import { join as join14 } from "path";
 import { homedir as homedir12 } from "os";
 import { parse as parseToml, stringify as stringifyToml } from "smol-toml";
-var DEFAULT_MODEL2 = "gpt-5.5";
 var CODEX_CONFIG_PATH = join14(homedir12(), ".codex", "config.toml");
 function isJsonlEvent2(value) {
   if (!isRecord4(value)) {
@@ -7758,7 +7783,7 @@ var CodexManager = class extends CodingAgentManager {
         workingDirectory: this.workingDirectory,
         skipGitRepoCheck: true,
         sandboxMode,
-        model: model || DEFAULT_MODEL2,
+        model: model || DEFAULT_CODEX_MODEL,
         webSearchMode,
         additionalDirectories,
         ...codexReasoningEffort ? { modelReasoningEffort: codexReasoningEffort } : {}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "replicas-engine",
-  "version": "0.1.240",
+  "version": "0.1.241",
   "description": "Lightweight API server for Replicas workspaces",
   "type": "module",
   "main": "dist/src/index.js",