replicas-engine 0.1.104 → 0.1.106

package/dist/src/index.js

@@ -1097,7 +1097,7 @@ function parseReplicasConfigString(content, filename) {
 }
 
 // ../shared/src/engine/environment.ts
-var DAYTONA_SNAPSHOT_ID = "13-04-2026-islington-v4";
+var DAYTONA_SNAPSHOT_ID = "15-04-2026-thinking-level-v1";
 
 // ../shared/src/engine/types.ts
 var DEFAULT_CHAT_TITLES = {
@@ -2473,7 +2473,8 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
       model,
       customInstructions,
       images,
-      permissionMode
+      permissionMode,
+      thinkingLevel
     } = request;
     const linearSessionId = ENGINE_ENV.LINEAR_SESSION_ID;
     if (!message || !message.trim()) {
@@ -2533,7 +2534,8 @@ var ClaudeManager = class _ClaudeManager extends CodingAgentManager {
           systemPrompt,
           ...this.mcpServersConfig ? { mcpServers: this.mcpServersConfig } : {},
           env: queryEnv,
-          model: model || "opus"
+          model: model || "opus",
+          ...thinkingLevel ? { effort: thinkingLevel } : {}
         }
       });
       this.activeQuery = response;
@@ -2765,7 +2767,8 @@ var CodexManager = class extends CodingAgentManager {
       model,
       customInstructions,
       images,
-      permissionMode
+      permissionMode,
+      thinkingLevel
     } = request;
     const linearSessionId = ENGINE_ENV.LINEAR_SESSION_ID;
     let tempImagePaths = [];
@@ -2780,12 +2783,14 @@ var CodexManager = class extends CodingAgentManager {
       await this.updateCodexConfig(developerInstructions);
       const sandboxMode = "danger-full-access";
       const webSearchMode = "live";
+      const codexReasoningEffort = thinkingLevel ? thinkingLevel === "max" ? "xhigh" : thinkingLevel : void 0;
       const threadOptions = {
         workingDirectory: this.workingDirectory,
         skipGitRepoCheck: true,
         sandboxMode,
         model: model || DEFAULT_MODEL,
-        webSearchMode
+        webSearchMode,
+        ...codexReasoningEffort ? { modelReasoningEffort: codexReasoningEffort } : {}
       };
       abortController = new AbortController();
       this.activeAbortController = abortController;
@@ -3106,6 +3111,9 @@ 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(codexAvailable ? "Model override. Claude: opus, sonnet, haiku. Codex: gpt-5.4, gpt-5.3-codex, etc." : "Model override. Claude: opus, sonnet, haiku."),
+      thinking_level: z.enum(["low", "medium", "high", "max"]).optional().describe(
+        "Controls how much thinking/reasoning the subagent applies. low = light thinking, medium = moderate, high = deep reasoning, max = maximum effort. Defaults: Claude = high, Codex = medium."
+      ),
       title: z.string().optional().describe("Optional title for the subagent chat (for identification)."),
       timeout_minutes: z.number().positive().optional().describe("Timeout in minutes for the subagent to complete (default: 10). Set higher for large tasks to avoid losing work.")
     },
@@ -3132,6 +3140,9 @@ You will also receive the chatId so you can send follow-up messages or clean up
         if (args.model) {
           msgBody.model = args.model;
         }
+        if (args.thinking_level) {
+          msgBody.thinkingLevel = args.thinking_level;
+        }
         const sendRes = await engineFetch(`/chats/${chatId}/messages`, {
           method: "POST",
           body: JSON.stringify(msgBody)
@@ -3168,6 +3179,9 @@ The tool blocks until the subagent completes and returns its response.`,
     chatId: z.string().describe("The chat ID of the subagent (returned by spawn_agent)."),
     message: z.string().describe("The follow-up message to send."),
     model: z.string().optional().describe("Optional model override for this message."),
+    thinking_level: z.enum(["low", "medium", "high", "max"]).optional().describe(
+      "Controls how much thinking/reasoning the subagent applies. low = light thinking, medium = moderate, high = deep reasoning, max = maximum effort. Defaults: Claude = high, Codex = medium."
+    ),
     timeout_minutes: z.number().positive().optional().describe("Timeout in minutes for the subagent to complete (default: 10). Set higher for large tasks to avoid losing work.")
   },
   async (args) => {
@@ -3176,6 +3190,9 @@ The tool blocks until the subagent completes and returns its response.`,
       if (args.model) {
         msgBody.model = args.model;
       }
+      if (args.thinking_level) {
+        msgBody.thinkingLevel = args.thinking_level;
+      }
       const sendRes = await engineFetch(`/chats/${args.chatId}/messages`, {
         method: "POST",
         body: JSON.stringify(msgBody)
@@ -4287,7 +4304,8 @@ var sendMessageSchema = z2.object({
         url: z2.string().url()
       })
     ])
-  })).optional()
+  })).optional(),
+  thinkingLevel: z2.enum(["low", "medium", "high", "max"]).optional()
 });
 function jsonError(message, details) {
   return { error: message, details };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "replicas-engine",
-  "version": "0.1.104",
+  "version": "0.1.106",
   "description": "Lightweight API server for Replicas workspaces",
   "type": "module",
   "main": "dist/src/index.js",