replicas-engine 0.1.373 → 0.1.375

package/dist/src/index.js

@@ -295,7 +295,7 @@ 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-2026-06-30-v2";
+var E2B_TEMPLATE_NAME = "replicas-sandbox-2026-07-01-v1";
 
 // ../shared/src/runtime-env.ts
 function parsePosixEnvFile(content) {
@@ -2254,26 +2254,39 @@ var DEFAULT_CHAT_TITLES = {
 };
 var CLAUDE_OPUS_1M_MODEL = "opus[1m]";
 var LEGACY_CLAUDE_OPUS_1M_MODEL = "opus-1m";
+var CLAUDE_OPUS_4_8_MODEL = "claude-opus-4-8";
+var CLAUDE_SONNET_5_MODEL = "claude-sonnet-5";
+var CLAUDE_HAIKU_4_5_MODEL = "claude-haiku-4-5";
+var CLAUDE_OPUS_4_8_BEDROCK_MODEL = "us.anthropic.claude-opus-4-8";
+var CLAUDE_SONNET_5_BEDROCK_MODEL = "anthropic.claude-sonnet-5";
+var CLAUDE_HAIKU_4_5_BEDROCK_MODEL = "us.anthropic.claude-haiku-4-5-20251001-v1:0";
+var DEFAULT_CLAUDE_MODEL = CLAUDE_OPUS_4_8_MODEL;
 var DEFAULT_CODEX_MODEL = "gpt-5.5";
 var DEFAULT_CURSOR_MODEL = "composer-2.5";
 var DEFAULT_OPENCODE_MODEL = "z-ai/glm-5.2";
 function normalizeClaudeModel(model) {
-  if (model === LEGACY_CLAUDE_OPUS_1M_MODEL) {
-    return CLAUDE_OPUS_1M_MODEL;
+  if (model === "opus" || model === CLAUDE_OPUS_1M_MODEL || model === LEGACY_CLAUDE_OPUS_1M_MODEL) {
+    return DEFAULT_CLAUDE_MODEL;
   }
   if (model === "sonnet") {
-    return "claude-sonnet-5";
+    return CLAUDE_SONNET_5_MODEL;
+  }
+  if (model === "haiku") {
+    return CLAUDE_HAIKU_4_5_MODEL;
   }
   return model;
 }
+function canUseClaudeFastMode(model) {
+  return normalizeClaudeModel(model) === CLAUDE_OPUS_4_8_MODEL || model === "claude-opus-4-7";
+}
 var AGENT_MODELS = {
-  claude: [CLAUDE_OPUS_1M_MODEL, "sonnet", "haiku"],
+  claude: [DEFAULT_CLAUDE_MODEL, CLAUDE_SONNET_5_MODEL, CLAUDE_HAIKU_4_5_MODEL],
   codex: [DEFAULT_CODEX_MODEL, "gpt-5.4", "gpt-5.4-mini", "gpt-5.3-codex", "gpt-5.2"],
   cursor: [
-    "claude-opus-4-8",
+    CLAUDE_OPUS_4_8_MODEL,
     "gpt-5.5",
     "composer-2.5",
-    "claude-sonnet-5",
+    CLAUDE_SONNET_5_MODEL,
     "claude-opus-4-7",
     "gpt-5.4",
     "gpt-5.4-mini",
@@ -2288,11 +2301,11 @@ var AGENT_MODELS = {
     "gemini-3.1-pro",
     "gemini-3-flash",
     "grok-4.3",
-    "claude-haiku-4-5",
+    CLAUDE_HAIKU_4_5_MODEL,
     "kimi-k2.5"
   ],
   opencode: [DEFAULT_OPENCODE_MODEL, "minimax/minimax-m3", "xiaomi/mimo-v2.5-pro", "moonshotai/kimi-k2.6"],
-  relay: [CLAUDE_OPUS_1M_MODEL, "sonnet"]
+  relay: [DEFAULT_CLAUDE_MODEL, CLAUDE_SONNET_5_MODEL]
 };
 var MODEL_LABELS = {
   sonnet: "Sonnet 5",
@@ -2317,9 +2330,9 @@ var MODEL_LABELS = {
   "gpt-5.1-codex-mini": "GPT-5.1 Codex Mini",
   "gpt-5-mini": "GPT-5 Mini",
   "composer-2.5": "Composer 2.5",
-  "claude-opus-4-8": "Opus 4.8",
+  [CLAUDE_OPUS_4_8_MODEL]: "Opus 4.8",
   "claude-opus-4-7": "Opus 4.7",
-  "claude-haiku-4-5": "Haiku 4.5",
+  [CLAUDE_HAIKU_4_5_MODEL]: "Haiku 4.5",
   "gemini-3.1-pro": "Gemini 3.1 Pro",
   "gemini-3-flash": "Gemini 3 Flash",
   "grok-4.3": "Grok 4.3",
@@ -2510,6 +2523,7 @@ var CLAUDE_AUTH_ENV_KEYS = [
   "AWS_ACCESS_KEY_ID",
   "AWS_SECRET_ACCESS_KEY",
   "AWS_REGION",
+  "ANTHROPIC_SMALL_FAST_MODEL_AWS_REGION",
   "REPLICAS_CLAUDE_AUTH_METHOD"
 ];
 function claudeAuthEnvFromResponse(response) {
@@ -2524,6 +2538,7 @@ function claudeAuthEnvFromResponse(response) {
         AWS_ACCESS_KEY_ID: response.awsAccessKeyId,
         AWS_SECRET_ACCESS_KEY: response.awsSecretAccessKey,
         AWS_REGION: response.awsRegion,
+        ...response.smallFastModelRegion ? { ANTHROPIC_SMALL_FAST_MODEL_AWS_REGION: response.smallFastModelRegion } : {},
         REPLICAS_CLAUDE_AUTH_METHOD: "bedrock"
       };
   }
@@ -2537,6 +2552,7 @@ var CLAUDE_AUTH_ENV_KEYS_BY_METHOD = {
     "AWS_ACCESS_KEY_ID",
     "AWS_SECRET_ACCESS_KEY",
     "AWS_REGION",
+    "ANTHROPIC_SMALL_FAST_MODEL_AWS_REGION",
     "REPLICAS_CLAUDE_AUTH_METHOD"
   ]
 };
@@ -3001,6 +3017,7 @@ function loadEngineEnv() {
     AWS_ACCESS_KEY_ID: readEnv("AWS_ACCESS_KEY_ID"),
     AWS_SECRET_ACCESS_KEY: readEnv("AWS_SECRET_ACCESS_KEY"),
     AWS_REGION: readEnv("AWS_REGION"),
+    ANTHROPIC_SMALL_FAST_MODEL_AWS_REGION: readEnv("ANTHROPIC_SMALL_FAST_MODEL_AWS_REGION"),
     REPLICAS_CLAUDE_AUTH_METHOD: parseClaudeAuthMethod(readEnv("REPLICAS_CLAUDE_AUTH_METHOD")),
     REPLICAS_CODEX_AUTH_METHOD: parseCodexAuthMethod(readEnv("REPLICAS_CODEX_AUTH_METHOD")),
     REPLICAS_DISABLE_GH_PR_MERGE: readEnv("REPLICAS_DISABLE_GH_PR_MERGE")?.toLowerCase() === "true",
@@ -6060,9 +6077,17 @@ function buildClaudeAgentEnv(overrides) {
   if (overrides) {
     Object.assign(env, overrides);
   }
-  if (shouldStripAnthropicApiKey()) {
+  const isBedrock = isClaudeBedrockEnv(env);
+  if (shouldStripAnthropicApiKey(isBedrock)) {
     env.ANTHROPIC_API_KEY = void 0;
   }
+  if (isBedrock) {
+    env.CLAUDE_CODE_USE_BEDROCK = "1";
+    env.CLAUDE_CODE_USE_MANTLE = "1";
+    env.ANTHROPIC_DEFAULT_OPUS_MODEL = CLAUDE_OPUS_4_8_BEDROCK_MODEL;
+    env.ANTHROPIC_DEFAULT_SONNET_MODEL = CLAUDE_SONNET_5_BEDROCK_MODEL;
+    env.ANTHROPIC_DEFAULT_HAIKU_MODEL = CLAUDE_HAIKU_4_5_BEDROCK_MODEL;
+  }
   env.GH_TOKEN = void 0;
   env.GITHUB_TOKEN = void 0;
   env.GH_CONFIG_DIR = void 0;
@@ -6084,9 +6109,12 @@ function buildCodexAgentEnv() {
   delete env.GH_CONFIG_DIR;
   return env;
 }
-function shouldStripAnthropicApiKey() {
+function shouldStripAnthropicApiKey(isBedrock) {
   const method = ENGINE_ENV.REPLICAS_CLAUDE_AUTH_METHOD;
-  return method === "oauth" || method === "bedrock";
+  return method === "oauth" || method === "bedrock" || isBedrock;
+}
+function isClaudeBedrockEnv(env) {
+  return ENGINE_ENV.REPLICAS_CLAUDE_AUTH_METHOD === "bedrock" || env.REPLICAS_CLAUDE_AUTH_METHOD === "bedrock" || env.CLAUDE_CODE_USE_BEDROCK === "1";
 }
 function shouldStripOpenAIApiKey() {
   return ENGINE_ENV.REPLICAS_CODEX_AUTH_METHOD === "oauth";
@@ -6571,8 +6599,23 @@ var COMMAND_PROTECTION_SAFE_TOOLS = /* @__PURE__ */ new Set([
 ]);
 var CLAUDE_PARTIAL_MESSAGE_FLUSH_MS = 500;
 function supportsClaudeThinkingDisplay(model) {
-  const normalized = model.toLowerCase();
-  return AGENT_MODELS.claude.includes(normalized) || normalized === "opus" || /^claude-(?:opus|sonnet|haiku)-[4-9]/.test(normalized);
+  const normalized = (normalizeClaudeModel(model) ?? model).toLowerCase();
+  return AGENT_MODELS.claude.includes(normalized) || /^claude-(?:opus|sonnet|haiku)-[4-9]/.test(normalized);
+}
+function toClaudeCodeModel(model) {
+  if (ENGINE_ENV.REPLICAS_CLAUDE_AUTH_METHOD !== "bedrock" && ENGINE_ENV.CLAUDE_CODE_USE_BEDROCK !== "1") {
+    return model;
+  }
+  switch (normalizeClaudeModel(model)) {
+    case CLAUDE_OPUS_4_8_MODEL:
+      return "opus";
+    case CLAUDE_SONNET_5_MODEL:
+      return "sonnet";
+    case CLAUDE_HAIKU_4_5_MODEL:
+      return "haiku";
+    default:
+      return model;
+  }
 }
 var TOOL_INPUT_HANDLERS = {
   ExitPlanMode: {
@@ -7022,8 +7065,10 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
       await this.recordEvent(userMessage);
     }
     const combinedInstructions = this.buildCombinedInstructions(customInstructions);
-    const resolvedModel = normalizeClaudeModel(model) || CLAUDE_OPUS_1M_MODEL;
+    const resolvedModel = normalizeClaudeModel(model) || DEFAULT_CLAUDE_MODEL;
+    const claudeCodeModel = toClaudeCodeModel(resolvedModel);
     const resolvedPermissionMode = planMode ? "plan" : "bypassPermissions";
+    const resolvedFastMode = Boolean(fastMode && canUseClaudeFastMode(resolvedModel));
     const signature = {
       combinedInstructions,
       thinkingLevel,
@@ -7031,12 +7076,13 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
       // tracks the effective flag so a mode change can hot-swap via setPermissionMode
       // without restarting.
       enableInteractiveTools: Boolean(enableInteractiveTools),
-      fastMode: Boolean(fastMode)
+      fastMode: resolvedFastMode
     };
     await this.ensureSession({
       signature,
       combinedInstructions,
       resolvedModel,
+      claudeCodeModel,
       resolvedPermissionMode,
       enableInteractiveTools: signature.enableInteractiveTools,
       thinkingLevel
@@ -7060,15 +7106,16 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
       signature,
       combinedInstructions,
       resolvedModel,
+      claudeCodeModel,
       resolvedPermissionMode,
       enableInteractiveTools,
       thinkingLevel
     } = args;
     if (this.activeQuery && this.activeSessionSignature && this.sessionSignaturesMatch(this.activeSessionSignature, signature)) {
-      if (this.activeSessionModel !== resolvedModel) {
+      if (this.activeSessionModel !== claudeCodeModel) {
         try {
-          await this.activeQuery.setModel(resolvedModel);
-          this.activeSessionModel = resolvedModel;
+          await this.activeQuery.setModel(claudeCodeModel);
+          this.activeSessionModel = claudeCodeModel;
         } catch (err) {
           console.warn("[ClaudeManager] setModel failed; recreating session:", err);
           await this.tearDownSession();
@@ -7090,6 +7137,7 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
     await this.startSession({
       combinedInstructions,
       resolvedModel,
+      claudeCodeModel,
       resolvedPermissionMode,
       enableInteractiveTools,
       thinkingLevel,
@@ -7103,6 +7151,7 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
     const {
       combinedInstructions,
       resolvedModel,
+      claudeCodeModel,
       resolvedPermissionMode,
       enableInteractiveTools,
       thinkingLevel,
@@ -7153,7 +7202,7 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
         ...registryPlugins.length > 0 ? { plugins: registryPlugins } : {},
         ...enableRegistrySkills ? { skills: "all" } : {},
         env: queryEnv,
-        model: resolvedModel,
+        model: claudeCodeModel,
         settings: { fastMode: signature.fastMode },
         includePartialMessages: true,
         ...supportsClaudeThinkingDisplay(resolvedModel) ? { thinking: { type: "adaptive", display: "summarized" } } : {},
@@ -7173,7 +7222,7 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
     this.activeQuery = response;
     this.activePromptStream = promptStream;
     this.activeSessionSignature = signature;
-    this.activeSessionModel = resolvedModel;
+    this.activeSessionModel = claudeCodeModel;
     this.activeSessionPermissionMode = resolvedPermissionMode;
     this.sessionLinearForwarder = new LinearEventForwarder(ENGINE_ENV.LINEAR_SESSION_ID);
     this.activeBackgroundTasks.clear();
@@ -7734,7 +7783,7 @@ 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 = "0.1.373";
+var ENGINE_PACKAGE_VERSION = "0.1.375";
 var INITIALIZE_METHOD = "initialize";
 var INITIALIZED_NOTIFICATION = "initialized";
 var ACCOUNT_LOGIN_START_METHOD = "account/login/start";
@@ -10436,7 +10485,7 @@ You will also receive the chatId so you can send follow-up messages or clean up
       provider: providerEnum.describe(providerDesc),
       prompt: z.string().describe("The full prompt/instructions for the subagent. Be detailed - it has no context from your conversation."),
       model: z.string().optional().describe([
-        `Model override. Claude: ${AGENT_MODELS.claude.join(", ")} (opus[1m] is the default, 1M context, use for very large codebases or huge context tasks).`,
+        `Model override. Claude: ${AGENT_MODELS.claude.join(", ")} (opus is the default; sonnet is faster).`,
         codexAvailable ? "Codex: gpt-5.5, gpt-5.4, gpt-5.3-codex, etc." : null,
         cursorAvailable ? `Cursor: ${AGENT_MODELS.cursor.join(", ")}.` : null,
         opencodeAvailable ? `Opencode: ${AGENT_MODELS.opencode.join(", ")}.` : null
@@ -10685,7 +10734,7 @@ function getDelegationSection(codexAvailable, cursorAvailable, opencodeAvailable
   ].filter(Boolean);
   const agentSelectionLines = extraAgentLines.length > 0 ? `${extraAgentLines.join("\n\n")}
 
-Use provider 'claude' for codebase exploration, code review, planning, complex debugging, and tasks requiring nuanced architectural understanding. Suggested models: ${claudeModelList} (opus[1m] is the default, 1M context window, use for very large codebases; sonnet is faster).` : `Use provider 'claude' for all tasks including code writing, codebase exploration, code review, planning, complex debugging, and tasks requiring nuanced architectural understanding. Suggested models: ${claudeModelList} (opus[1m] is the default, 1M context window, use for very large codebases; sonnet is faster).`;
+Use provider 'claude' for codebase exploration, code review, planning, complex debugging, and tasks requiring nuanced architectural understanding. Suggested models: ${claudeModelList} (opus is the default; sonnet is faster).` : `Use provider 'claude' for all tasks including code writing, codebase exploration, code review, planning, complex debugging, and tasks requiring nuanced architectural understanding. Suggested models: ${claudeModelList} (opus is the default; sonnet is faster).`;
   return `# Delegation
 
 You have three subagent tools for spawning and managing agents that run in separate context windows:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "replicas-engine",
-  "version": "0.1.373",
+  "version": "0.1.375",
   "description": "Lightweight API server for Replicas workspaces",
   "type": "module",
   "main": "dist/src/index.js",

package/scripts/engine-watchdog.sh

@@ -11,6 +11,11 @@ BACKOFF=1
 RESTART_COUNT=0
 MAX_RESTARTS=20
 STABLE_THRESHOLD=300
+BUN_INSTALL="${BUN_INSTALL:-/usr/local/bun}"
+if [ -d "$BUN_INSTALL/bin" ]; then
+  export BUN_INSTALL
+  export PATH="$BUN_INSTALL/bin:$PATH"
+fi
 
 log() {
   echo "[$(date -Iseconds)] [engine-watchdog] $*"