@mindstudio-ai/remy 0.1.63 → 0.1.65

package/dist/headless.js

@@ -574,7 +574,7 @@ async function* streamChat(params) {
 var MAX_RETRIES = 3;
 var INITIAL_BACKOFF_MS = 1e3;
 function isRetryableError(error) {
-  return /Network error/i.test(error) || /HTTP 5\d\d/i.test(error) || /Stream stalled/i.test(error);
+  return /Network error/i.test(error) || /HTTP 5\d\d/i.test(error) || /Stream stalled/i.test(error) || /overloaded/i.test(error);
 }
 function sleep(ms) {
   return new Promise((resolve) => setTimeout(resolve, ms));
@@ -1892,7 +1892,7 @@ function formatOccurrenceError(count, lines, filePath) {
 var editFileTool = {
   definition: {
     name: "editFile",
-    description: "Replace a string in a file. old_string must appear exactly once (minor indentation differences are handled automatically). Set replace_all to true to replace every occurrence at once. For bulk mechanical substitutions (renaming a variable, swapping colors), prefer replace_all. Always read the file first so you know the exact text to match.",
+    description: "Replace a string in a file. old_string must appear exactly once (minor indentation differences are handled automatically). Set replace_all to true to replace every occurrence at once. For bulk mechanical substitutions (renaming a variable, swapping colors), prefer replace_all. Always read the file first so you know the exact text to match. When editing nested structures (objects, function bodies, arrays, template literals), always include the full enclosing structure in old_string rather than just an inner fragment. Replacing a partial slice from the middle of nested code is the most common source of syntax errors.",
     inputSchema: {
       type: "object",
       properties: {
@@ -3345,11 +3345,37 @@ var definition5 = {
       path: {
         type: "string",
         description: 'Navigate to this path before capturing (e.g. "/settings"). If omitted, screenshots the current page.'
+      },
+      instructions: {
+        type: "string",
+        description: "If the screenshot you need requires interaction first (dismissing a modal, clicking a tab, filling out a form, navigating a flow), describe the steps to get there. A browser automation agent will follow these instructions before capturing the screenshot."
       }
     }
   }
 };
-async function execute5(input, onLog) {
+async function execute5(input, onLog, context) {
+  if (input.instructions && context) {
+    try {
+      const task = input.path ? `Navigate to "${input.path}", then: ${input.instructions}. After completing these steps, take a full-page screenshot.` : `${input.instructions}. After completing these steps, take a full-page screenshot.`;
+      const result = await browserAutomationTool.execute({ task }, context);
+      const urlMatch = result.match(
+        /https:\/\/[^\s"')]+\.(?:png|jpg|jpeg|webp)/i
+      );
+      if (!urlMatch) {
+        return `Error: browser navigation completed but no screenshot URL was returned. Agent output: ${result}`;
+      }
+      const url = urlMatch[0];
+      const analysisPrompt = input.prompt || SCREENSHOT_ANALYSIS_PROMPT;
+      const analysis = await analyzeImage({
+        prompt: analysisPrompt,
+        imageUrl: url,
+        onLog
+      });
+      return JSON.stringify({ url, analysis });
+    } catch (err) {
+      return `Error taking interactive screenshot: ${err.message}`;
+    }
+  }
   try {
     return await captureAndAnalyzeScreenshot({
       prompt: input.prompt,
@@ -3618,7 +3644,7 @@ async function executeDesignExpertTool(name, input, context, toolCallId, onLog)
   if (!tool) {
     return `Error: unknown tool "${name}"`;
   }
-  return tool.execute(input, onLog);
+  return tool.execute(input, onLog, context);
 }
 
 // src/subagents/designExpert/prompt.ts
@@ -4856,6 +4882,9 @@ async function runTurn(params) {
   let turnCacheCreation = 0;
   let turnCacheRead = 0;
   let turnLlmCalls = 0;
+  let lastCallInputTokens = 0;
+  let lastCallCacheCreation = 0;
+  let lastCallCacheRead = 0;
   while (true) {
     let getOrCreateAccumulator2 = function(id, name) {
       let acc = toolInputAccumulators.get(id);
@@ -5051,10 +5080,13 @@ async function runTurn(params) {
           case "done":
             stopReason = event.stopReason;
             turnLlmCalls++;
-            turnInputTokens += event.usage.inputTokens;
+            lastCallInputTokens = event.usage.inputTokens;
+            lastCallCacheCreation = event.usage.cacheCreationTokens ?? 0;
+            lastCallCacheRead = event.usage.cacheReadTokens ?? 0;
+            turnInputTokens += lastCallInputTokens;
             turnOutputTokens += event.usage.outputTokens;
-            turnCacheCreation += event.usage.cacheCreationTokens ?? 0;
-            turnCacheRead += event.usage.cacheReadTokens ?? 0;
+            turnCacheCreation += lastCallCacheCreation;
+            turnCacheRead += lastCallCacheRead;
             break;
           case "error":
             onEvent({ type: "error", error: friendlyError(event.error) });
@@ -5099,7 +5131,10 @@ async function runTurn(params) {
           outputTokens: turnOutputTokens,
           cacheCreationTokens: turnCacheCreation || void 0,
           cacheReadTokens: turnCacheRead || void 0,
-          llmCalls: turnLlmCalls
+          llmCalls: turnLlmCalls,
+          lastCallInputTokens,
+          lastCallCacheCreation: lastCallCacheCreation || void 0,
+          lastCallCacheRead: lastCallCacheRead || void 0
         }
       });
       return;
@@ -5550,7 +5585,7 @@ ${xmlParts}
           sessionStats.totalOutputTokens += e.stats.outputTokens;
           sessionStats.totalCacheCreationTokens += e.stats.cacheCreationTokens ?? 0;
           sessionStats.totalCacheReadTokens += e.stats.cacheReadTokens ?? 0;
-          sessionStats.lastContextSize = e.stats.inputTokens;
+          sessionStats.lastContextSize = e.stats.lastCallInputTokens ?? e.stats.inputTokens;
         }
         sessionStats.messageCount = state.messages.length;
         sessionStats.updatedAt = Date.now();

package/dist/index.js

@@ -221,7 +221,7 @@ async function* streamChat(params) {
   }
 }
 function isRetryableError(error) {
-  return /Network error/i.test(error) || /HTTP 5\d\d/i.test(error) || /Stream stalled/i.test(error);
+  return /Network error/i.test(error) || /HTTP 5\d\d/i.test(error) || /Stream stalled/i.test(error) || /overloaded/i.test(error);
 }
 function sleep(ms) {
   return new Promise((resolve) => setTimeout(resolve, ms));
@@ -1477,7 +1477,7 @@ var init_editFile = __esm({
     editFileTool = {
       definition: {
         name: "editFile",
-        description: "Replace a string in a file. old_string must appear exactly once (minor indentation differences are handled automatically). Set replace_all to true to replace every occurrence at once. For bulk mechanical substitutions (renaming a variable, swapping colors), prefer replace_all. Always read the file first so you know the exact text to match.",
+        description: "Replace a string in a file. old_string must appear exactly once (minor indentation differences are handled automatically). Set replace_all to true to replace every occurrence at once. For bulk mechanical substitutions (renaming a variable, swapping colors), prefer replace_all. Always read the file first so you know the exact text to match. When editing nested structures (objects, function bodies, arrays, template literals), always include the full enclosing structure in old_string rather than just an inner fragment. Replacing a partial slice from the middle of nested code is the most common source of syntax errors.",
         inputSchema: {
           type: "object",
           properties: {
@@ -3172,7 +3172,29 @@ __export(screenshot_exports, {
   definition: () => definition5,
   execute: () => execute5
 });
-async function execute5(input, onLog) {
+async function execute5(input, onLog, context) {
+  if (input.instructions && context) {
+    try {
+      const task = input.path ? `Navigate to "${input.path}", then: ${input.instructions}. After completing these steps, take a full-page screenshot.` : `${input.instructions}. After completing these steps, take a full-page screenshot.`;
+      const result = await browserAutomationTool.execute({ task }, context);
+      const urlMatch = result.match(
+        /https:\/\/[^\s"')]+\.(?:png|jpg|jpeg|webp)/i
+      );
+      if (!urlMatch) {
+        return `Error: browser navigation completed but no screenshot URL was returned. Agent output: ${result}`;
+      }
+      const url = urlMatch[0];
+      const analysisPrompt = input.prompt || SCREENSHOT_ANALYSIS_PROMPT;
+      const analysis = await analyzeImage({
+        prompt: analysisPrompt,
+        imageUrl: url,
+        onLog
+      });
+      return JSON.stringify({ url, analysis });
+    } catch (err) {
+      return `Error taking interactive screenshot: ${err.message}`;
+    }
+  }
   try {
     return await captureAndAnalyzeScreenshot({
       prompt: input.prompt,
@@ -3188,6 +3210,8 @@ var init_screenshot3 = __esm({
   "src/subagents/designExpert/tools/screenshot.ts"() {
     "use strict";
     init_screenshot();
+    init_analyzeImage();
+    init_browserAutomation();
     definition5 = {
       name: "screenshot",
       description: "Capture a full-height screenshot of the current app preview. Returns a CDN URL along with visual analysis. Use to review the current state of the UI being built. Remember, the screenshot analysis is not overly precise - for example, it cannot reliably identify specific fonts by name \u2014 it can only describe what letterforms look like.",
@@ -3201,6 +3225,10 @@ var init_screenshot3 = __esm({
           path: {
             type: "string",
             description: 'Navigate to this path before capturing (e.g. "/settings"). If omitted, screenshots the current page.'
+          },
+          instructions: {
+            type: "string",
+            description: "If the screenshot you need requires interaction first (dismissing a modal, clicking a tab, filling out a form, navigating a flow), describe the steps to get there. A browser automation agent will follow these instructions before capturing the screenshot."
           }
         }
       }
@@ -3482,7 +3510,7 @@ async function executeDesignExpertTool(name, input, context, toolCallId, onLog)
   if (!tool) {
     return `Error: unknown tool "${name}"`;
   }
-  return tool.execute(input, onLog);
+  return tool.execute(input, onLog, context);
 }
 var tools, DESIGN_EXPERT_TOOLS;
 var init_tools2 = __esm({
@@ -4896,6 +4924,9 @@ async function runTurn(params) {
   let turnCacheCreation = 0;
   let turnCacheRead = 0;
   let turnLlmCalls = 0;
+  let lastCallInputTokens = 0;
+  let lastCallCacheCreation = 0;
+  let lastCallCacheRead = 0;
   while (true) {
     let getOrCreateAccumulator2 = function(id, name) {
       let acc = toolInputAccumulators.get(id);
@@ -5091,10 +5122,13 @@ async function runTurn(params) {
           case "done":
             stopReason = event.stopReason;
             turnLlmCalls++;
-            turnInputTokens += event.usage.inputTokens;
+            lastCallInputTokens = event.usage.inputTokens;
+            lastCallCacheCreation = event.usage.cacheCreationTokens ?? 0;
+            lastCallCacheRead = event.usage.cacheReadTokens ?? 0;
+            turnInputTokens += lastCallInputTokens;
             turnOutputTokens += event.usage.outputTokens;
-            turnCacheCreation += event.usage.cacheCreationTokens ?? 0;
-            turnCacheRead += event.usage.cacheReadTokens ?? 0;
+            turnCacheCreation += lastCallCacheCreation;
+            turnCacheRead += lastCallCacheRead;
             break;
           case "error":
             onEvent({ type: "error", error: friendlyError(event.error) });
@@ -5139,7 +5173,10 @@ async function runTurn(params) {
           outputTokens: turnOutputTokens,
           cacheCreationTokens: turnCacheCreation || void 0,
           cacheReadTokens: turnCacheRead || void 0,
-          llmCalls: turnLlmCalls
+          llmCalls: turnLlmCalls,
+          lastCallInputTokens,
+          lastCallCacheCreation: lastCallCacheCreation || void 0,
+          lastCallCacheRead: lastCallCacheRead || void 0
         }
       });
       return;
@@ -6162,7 +6199,7 @@ ${xmlParts}
           sessionStats.totalOutputTokens += e.stats.outputTokens;
           sessionStats.totalCacheCreationTokens += e.stats.cacheCreationTokens ?? 0;
           sessionStats.totalCacheReadTokens += e.stats.cacheReadTokens ?? 0;
-          sessionStats.lastContextSize = e.stats.inputTokens;
+          sessionStats.lastContextSize = e.stats.lastCallInputTokens ?? e.stats.inputTokens;
         }
         sessionStats.messageCount = state.messages.length;
         sessionStats.updatedAt = Date.now();

package/dist/prompt/static/team.md

@@ -10,7 +10,7 @@ Note: when you talk about the team to the user, refer to them by their name or a
 
 Your designer. Consult for any visual decision — choosing a color, picking fonts, proposing a layout, soucing images, reviewing whether something looks good. Not just during intake or big design moments. If you're about to write CSS and you're not sure about a color, ask. If you just built a page and want a gut check, ask the designer to take a quick look. If the user says "I don't like how this looks," ask the design expert what to change rather than guessing yourself, or if they say "I want a different image," that's the designer's problem, not yours. The design expert can also source images if you need images for placeholders in scenarios - use it for bespoke, tailor-made images suited to the scenario instead of trying to guess stock photo URLs.
 
-The design expert cannot see your conversation with the user, so include relevant context and requirements in your task. It can, however, see its past conversation with you, so you don't need to re-summarize everything it already knows. Just describe what's needed now and reference prior work naturally ("the user wants the colors warmer" is enough if the designer already built the palette). It can take screenshots of the app preview on its own — just ask it to review what's been built. It has curated font catalogs and design inspiration built in — don't ask it to research generic inspiration or look up "best X apps." Only point it at specific URLs if the user references a particular site, brand, or identity to match.
+The design expert cannot see your conversation with the user, so include relevant context and requirements in your task. It can, however, see its past conversation with you, so you don't need to re-summarize everything it already knows. Just describe what's needed now and reference prior work naturally ("the user wants the colors warmer" is enough if the designer already built the palette). It can take screenshots of the app preview on its own (you need to give it paths to different pages if it needs them - it can't navigate by clicking) — just ask it to review what's been built. It has curated font catalogs and design inspiration built in — don't ask it to research generic inspiration or look up "best X apps." Only point it at specific URLs if the user references a particular site, brand, or identity to match.
 
 The designer will return concrete resources: hex values, font names with CSS URLs, image URLs, layout descriptions, as well as specific techniques, CSS properties, and other values. Even if these don't seem important, it is critical that you note them in spec annotations and rely on them while building - the user cares about design almost above all else, and it is important to be extremely precise in your work.
 
@@ -38,7 +38,7 @@ Always consult the code sanity check before writing code in initialCodegen with
 
 ### QA (`runAutomatedBrowserTest`)
 
-For verifying complex stateful interactions: multi-step form submissions, auth flows, real-time updates, flows that require specific data/role setup. This spins up a full chrome browser automation — it's heavyweight. Do not use it for basic rendering or navigation checks. If you can verify something with a screenshot or by reading the code, do that instead. Run a scenario first to seed test data and set user roles. The user is able to watch QA work on their screen via a live browser preview - the cursor will move, type, etc - so you can also use this to demo functionality to the user and help them understand how to use their app.
+For verifying complex stateful interactions: multi-step form submissions, auth flows, real-time updates, flows that require specific data/role setup. This spins up a full chrome browser automation — it's heavyweight and takes minutes to complete a full test. Do not use it for basic rendering or navigation checks. If you can verify something with a screenshot or by reading the code, do that instead. Don't run it constantly after making small changes - save it for meaningful work. Run a scenario first to seed test data and set user roles. The user is able to watch QA work on their screen via a live browser preview - the cursor will move, type, etc - so you can also use this to demo functionality to the user and help them understand how to use their app.
 
 The QA agent can see the screen. Describe what to test, not how — it will figure out what to click, what to check, and what values to use.
 

package/dist/subagents/browserAutomation/prompt.md

@@ -1,5 +1,8 @@
 You are a browser smoke test agent. You verify that features work end to end by interacting with the live preview. Focus on outcomes: does the feature work? Did the expected content appear? Just do the thing and see if it worked.
 
+## Rules to Remember
+- Don't overthink the tests - the goal is to generally make sure things work as expected, not to provide detailed QA. If something seems mostly okay, note it and move on. Don't continue exploring to try to diagnose specific issues or get specific details unless you are asked to.
+
 ## Tester Persona
 The user is watching the automation happen on their screen in real-time. When typing into forms or inputs, behave like a realistic user of this specific app. Use the app context (if provided) to understand the audience and tone. Type the way that audience would actually type — not formal, not robotic. The app developer's name is Remy, so use that and the email remy@mindstudio.ai as the basis for any testing that requires a persona.
 

package/dist/subagents/designExpert/prompts/instructions.md

@@ -10,6 +10,7 @@ Then, think about the layout and UI patterns - these are the core of the user's
 
 ## Tool Usage
 - When multiple tool calls are independent, make them all in a single turn. Searching for three different products, or fetching two reference sites: batch them instead of doing one per turn.
+- The screenshot tool supports an `instructions` parameter for taking screenshots that require interaction first. If you need to screenshot a state that's behind a modal, a specific tab, or a multi-step flow, pass `instructions` describing how to get there (e.g., "dismiss the welcome modal, then click XYZ"). A browser automation agent will follow your instructions and capture the screenshot for you.
 
 ## Voice
 - No emoji, no filler.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mindstudio-ai/remy",
-  "version": "0.1.63",
+  "version": "0.1.65",
   "description": "MindStudio coding agent",
   "repository": {
     "type": "git",