@mindstudio-ai/remy 0.1.255 → 0.1.256

package/dist/headless.js

@@ -2764,6 +2764,7 @@ async function sidecarRequest(endpoint, body = {}, options) {
     throw new Error("Sidecar not available");
   }
   const url = `${baseUrl}${endpoint}`;
+  let data;
   try {
     const res = await fetch(url, {
       method: "POST",
@@ -2775,12 +2776,7 @@ async function sidecarRequest(endpoint, body = {}, options) {
       log4.error("Sidecar error", { endpoint, status: res.status });
       throw new Error(`Sidecar error: ${res.status}`);
     }
-    const data = await res.json();
-    if (data?.success === false) {
-      const code = data.errorCode ? ` [${data.errorCode}]` : "";
-      throw new Error(`${data.error || "Unknown error"}${code}`);
-    }
-    return data;
+    data = await res.json();
   } catch (err) {
     if (err.message.startsWith("Sidecar error")) {
       throw err;
@@ -2788,6 +2784,16 @@ async function sidecarRequest(endpoint, body = {}, options) {
     log4.error("Sidecar connection error", { endpoint, error: err.message });
     throw new Error(`Sidecar connection error: ${err.message}`);
   }
+  if (data?.success === false) {
+    const code = data.errorCode ? ` [${data.errorCode}]` : "";
+    log4.error("Sidecar command failed", {
+      endpoint,
+      error: data.error,
+      errorCode: data.errorCode
+    });
+    throw new Error(`${data.error || "Unknown error"}${code}`);
+  }
+  return data;
 }
 
 // src/tools/_helpers/lsp.ts
@@ -2973,6 +2979,8 @@ async function analyzeImage(params) {
 }
 
 // src/tools/_helpers/screenshot.ts
+var VIEWPORT_CAPTURE_TIMEOUT_MS = 45e3;
+var FULLPAGE_CAPTURE_TIMEOUT_MS = 135e3;
 var SCREENSHOT_ANALYSIS_PROMPT = `Describe everything visible on screen from top to bottom \u2014 every element, its position, its size relative to the viewport, its colors, its content. Be comprehensive, thorough, and spatial. After the inventory, note anything that looks visually broken (overlapping elements, clipped text, misaligned components).`;
 var ANALYSIS_RESPONSE_FORMAT = `Respond only with your analysis as Markdown and absolutely no other text. Do not use emojis - use unicode if you need symbols.`;
 function buildScreenshotAnalysisPrompt(opts) {
@@ -3048,7 +3056,9 @@ async function captureAndAnalyzeScreenshot(promptOrOptions) {
         ...height != null ? { height } : {},
         ...format ? { format } : {}
       },
-      { timeout: fullPage ? 12e4 : 3e4 }
+      {
+        timeout: fullPage ? FULLPAGE_CAPTURE_TIMEOUT_MS : VIEWPORT_CAPTURE_TIMEOUT_MS
+      }
     );
     url = ssResult?.url || ssResult?.screenshotUrl;
     if (!url) {
@@ -3250,6 +3260,9 @@ function findLastSummaryCheckpoint(messages, name) {
   }
   return -1;
 }
+function portableToolCallId(id) {
+  return id.replace(/[^a-zA-Z0-9_-]/g, "_");
+}
 function fixOrphanedToolCalls(messages) {
   const toolResultIds = /* @__PURE__ */ new Set();
   for (const msg of messages) {
@@ -3341,14 +3354,30 @@ ${summaryBlock.text}
 
 ${content}` : attachmentHeader;
       }
-      return { ...rest, content };
+      return {
+        ...rest,
+        content,
+        ...msg.toolCallId && {
+          toolCallId: portableToolCallId(msg.toolCallId)
+        }
+      };
     }
     if (!Array.isArray(msg.content)) {
       return msg;
     }
     const blocks = msg.content;
     const text = blocks.filter((b) => b.type === "text").map((b) => b.text).join("");
-    const toolCalls = blocks.filter((b) => b.type === "tool").map((b) => ({ id: b.id, name: b.name, input: b.input }));
+    const toolBlocks = blocks.filter(
+      (b) => b.type === "tool"
+    );
+    const toolCalls = toolBlocks.map((b) => ({
+      id: portableToolCallId(b.id),
+      name: b.name,
+      input: b.input
+    }));
+    const rewroteToolIds = toolBlocks.some(
+      (b) => portableToolCallId(b.id) !== b.id
+    );
     const cleaned2 = {
       role: msg.role,
       content: text
@@ -3356,7 +3385,7 @@ ${content}` : attachmentHeader;
     if (toolCalls.length > 0) {
       cleaned2.toolCalls = toolCalls;
     }
-    if (msg.providerMetadata) {
+    if (msg.providerMetadata && !rewroteToolIds) {
       cleaned2.providerMetadata = msg.providerMetadata;
     }
     if (msg.hidden) {
@@ -3874,7 +3903,7 @@ var BROWSER_TOOLS = [
                   "screenshotViewport",
                   "setViewport"
                 ],
-                description: 'snapshot: accessibility tree of the page (waits for network to settle). click: click an element (animated cursor, full event sequence). type: type text into input (one char at a time, works with React/Vue/Svelte). select: select a dropdown option by text. wait: wait for an element to appear (polls 100ms, waits for network). navigate: navigate to a URL within the app (waits for load, subsequent steps run on new page). evaluate: run JS in the page. styles: read computed CSS styles from elements (pass properties array with camelCase names, or omit for defaults). screenshotFullPage: full-page viewport-stitched screenshot (returns CDN url with dimensions). screenshotViewport: screenshot of just the visible viewport \u2014 pass `scrollToSelector` (or `scrollY`) on this step to scroll a section into view and capture it in one atomic step (no separate scroll needed). setViewport: switch the browser between desktop and mobile rendering (pass `mode`: "desktop" or "mobile"). Reloads the page so responsive layouts, media queries, and matchMedia re-evaluate \u2014 use it to QA mobile/responsive views.'
+                description: 'snapshot: accessibility tree of the page (waits for network to settle). click: click an element (animated cursor, full event sequence). type: type text into input (one char at a time, works with React/Vue/Svelte). select: select a dropdown option by text. wait: wait for an element to appear (polls 100ms, waits for network). navigate: navigate to a URL within the app (waits for load, subsequent steps run on new page). evaluate: run JS in the page. styles: read computed CSS styles from elements (pass properties array with camelCase names, or omit for defaults). screenshotFullPage: screenshot of the whole page top-to-bottom (returns a CDN url with dimensions and a written analysis). screenshotViewport: screenshot of just the visible viewport \u2014 pass `scrollToSelector` (or `scrollY`) on this step to scroll a section into view and capture it in one atomic step (no separate scroll needed). setViewport: switch the browser between desktop and mobile rendering (pass `mode`: "desktop" or "mobile"). Reloads the page so responsive layouts, media queries, and matchMedia re-evaluate \u2014 use it to QA mobile/responsive views.'
               },
               ref: {
                 type: "string",
@@ -3942,20 +3971,11 @@ var BROWSER_TOOLS = [
       required: ["steps"]
     }
   },
-  {
-    clearable: true,
-    name: "screenshotFullPage",
-    description: "Capture a full-height screenshot of the current page. Returns a CDN URL with full text analysis and description.",
-    inputSchema: {
-      type: "object",
-      properties: {
-        path: {
-          type: "string",
-          description: 'Navigate to this path before capturing (e.g. "/settings"). If omitted, screenshots the current page.'
-        }
-      }
-    }
-  },
+  // Captures are `browserCommand` steps only — there is deliberately no
+  // standalone screenshot tool here. Both used to exist for full-page, with
+  // different budgets, different result plumbing, and analysis on only one of
+  // them, so which door you picked changed what you got back.
+  //
   // Read tools so the QA agent can pull full spec detail on demand — the spec
   // context in its prompt is a lightweight index (see prompt.ts) that points
   // here. Routed to the global executeTool in index.ts, mirroring specSync.
@@ -4149,6 +4169,7 @@ function getBrowserAutomationPrompt() {
 
 // src/subagents/browserAutomation/index.ts
 var log7 = createLogger("browser-automation");
+var CAPTURE_COMMANDS = /* @__PURE__ */ new Set(["screenshotViewport", "screenshotFullPage"]);
 async function runBrowserAutomation(task, context, opts) {
   const release = await acquireBrowserLock();
   try {
@@ -4160,7 +4181,7 @@ async function runBrowserAutomation(task, context, opts) {
       );
     } catch {
     }
-    let lastBrowserCommandViewport;
+    let lastCapture = {};
     const result = await runSubAgent({
       system: getBrowserAutomationPrompt(),
       task,
@@ -4182,22 +4203,6 @@ async function runBrowserAutomation(task, context, opts) {
             return `Error setting up browser: ${err.message}`;
           }
         }
-        if (name === "screenshotFullPage") {
-          try {
-            return await captureAndAnalyzeScreenshot({
-              path: _input.path,
-              fullPage: true,
-              onLog,
-              model: resolveModel(
-                "imageAnalysis",
-                context.models,
-                context.model
-              )
-            });
-          } catch (err) {
-            return `Error taking screenshot: ${err.message}`;
-          }
-        }
         if (COMMON_READ_TOOL_NAMES.has(name) || name === readSpecTool.definition.name) {
           return executeTool(name, _input, context);
         }
@@ -4219,14 +4224,16 @@ async function runBrowserAutomation(task, context, opts) {
           try {
             const parsed = JSON.parse(result2);
             const screenshotSteps = (parsed.steps || []).filter(
-              (s) => s.command === "screenshotViewport" && s.result?.url
+              (s) => CAPTURE_COMMANDS.has(s.command) && s.result?.url
             );
             if (screenshotSteps.length > 0) {
-              const lastStep = screenshotSteps[screenshotSteps.length - 1];
-              lastBrowserCommandViewport = {
-                url: lastStep.result.url,
-                styleMap: lastStep.result.styleMap
-              };
+              for (const step of screenshotSteps) {
+                const kind = step.command === "screenshotFullPage" ? "fullPage" : "viewport";
+                lastCapture[kind] = {
+                  url: step.result.url,
+                  styleMap: step.result.styleMap
+                };
+              }
               const visionOverride = {
                 model: resolveModel(
                   "imageAnalysis",
@@ -4250,13 +4257,12 @@ async function runBrowserAutomation(task, context, opts) {
               );
               try {
                 const analyses = JSON.parse(batchResult);
-                let ai = 0;
-                for (const step of parsed.steps) {
-                  if (step.command === "screenshotViewport" && step.result?.url && ai < analyses.length) {
-                    step.result.analysis = analyses[ai]?.output?.analysis || analyses[ai]?.output || "";
-                    ai++;
+                screenshotSteps.forEach((step, i) => {
+                  if (i >= analyses.length) {
+                    return;
                   }
-                }
+                  step.result.analysis = analyses[i]?.output?.analysis || analyses[i]?.output || "";
+                });
               } catch {
                 log7.debug("Failed to parse batch analysis result", {
                   batchResult
@@ -4269,13 +4275,10 @@ async function runBrowserAutomation(task, context, opts) {
         }
         return result2;
       },
-      toolRegistry: context.toolRegistry,
-      captureArtifacts: ["screenshotFullPage"]
+      toolRegistry: context.toolRegistry
     });
     context.subAgentMessages?.set(context.toolCallId, result.messages);
-    const fullPage = result.artifacts?.screenshotFullPage;
-    const viewport = lastBrowserCommandViewport;
-    const preferred = opts?.capture === "viewport" ? viewport ?? fullPage : fullPage ?? viewport;
+    const preferred = opts?.capture === "viewport" ? lastCapture.viewport ?? lastCapture.fullPage : lastCapture.fullPage ?? lastCapture.viewport;
     return {
       text: result.text,
       ...preferred?.url ? { screenshot: { url: preferred.url, styleMap: preferred.styleMap } } : {}
@@ -4315,98 +4318,102 @@ var browserAutomationTool = {
 };
 
 // src/tools/code/screenshot.ts
-var screenshotTool = {
+var screenshotDefinition = {
   clearable: true,
-  definition: {
-    name: "screenshot",
-    description: "Capture a screenshot of the app preview and get a description of what's on screen. Choose `fullPage`: `false` captures just the visible viewport (fast \u2014 for a specific section the page is scrolled to), `true` captures the entire page top-to-bottom (slower \u2014 for overall composition or content past the fold). Captures the settled page state \u2014 it cannot catch animations, transitions, or transient state. Optionally provide specific questions about what you're looking for. Use a bulleted list to ask many questions at once. To ask additional questions about a screenshot you have already captured, pass its URL as imageUrl to skip recapture. If the screenshot requires interaction first (logging in, clicking a tab, dismissing a modal, scrolling to a section), use the instructions param to describe the steps. To render a fixed-size image such as an Open Graph share card, set `width` and `height` (e.g. 1200 \xD7 630) and `format: 'png'`: the tool navigates to `path`, clips to exactly those pixel dimensions, and returns the image URL.",
-    inputSchema: {
-      type: "object",
-      properties: {
-        fullPage: {
-          type: "boolean",
-          description: "true = full-height capture of the entire page; false = just the visible viewport. Pick based on whether you need the whole page or a specific section."
-        },
-        prompt: {
-          type: "string",
-          description: "Optional question about the screenshot. If omitted, returns a general description of what's visible."
-        },
-        imageUrl: {
-          type: "string",
-          description: "URL of an existing screenshot to analyze instead of capturing a new one. Use this for additional questions about a previous screenshot."
-        },
-        path: {
-          type: "string",
-          description: 'Navigate to this path before capturing (e.g. "/settings", "/dashboard"). If omitted, screenshots the current page.'
-        },
-        width: {
-          type: "number",
-          description: "Exact capture width in pixels. Set together with `height` to render a fixed-size image; clips to exactly this viewport instead of the default preview size."
-        },
-        height: {
-          type: "number",
-          description: "Exact capture height in pixels. Set together with `width`."
-        },
-        format: {
-          type: "string",
-          enum: ["png", "jpeg"],
-          description: "Output image format. Defaults to 'jpeg'. Use 'png' for crisp flat graphics like share cards, where JPEG artifacts show on sharp type and edges."
-        },
-        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, scrolling to a section, getting through a login/auth checkpoint), describe the steps to get there. A browser automation agent will follow these instructions, then capture per your `fullPage` choice \u2014 so with `fullPage: false` you can scroll to a section and capture just that viewport. It can bypass auth and get right to where it needs to be if you tell it to authenticate as a test user and give it the path/screen to start its test at. Never describe what names or values to use when applying the instructions - the browser automation agent must use its own values for it to work properly. If a specific auth role is required to access the content, be sure to note that - it can automatically assume it for the purpose of testing. Use only when interaction is required to *reach* the state you want to capture \u2014 log in, dismiss a modal, switch a tab, follow a route, scroll to a section. If your steps are exercising the app's functionality across multiple states (running flows, asserting behavior under interaction, multi-step QA), use `runAutomatedBrowserTest` instead."
-        }
+  name: "screenshot",
+  description: "Capture a screenshot of the app preview and get a description of what's on screen. Choose `fullPage`: `false` captures just the visible viewport (fast \u2014 for a specific section the page is scrolled to), `true` captures the entire page top-to-bottom (slower \u2014 for overall composition or content past the fold). Captures the settled page state \u2014 it cannot catch animations, transitions, or transient state. The analysis is not precise about every detail \u2014 for example it cannot reliably identify specific fonts by name, only describe what the letterforms look like. Optionally provide specific questions about what you're looking for. Use a bulleted list to ask many questions at once. To ask additional questions about a screenshot you have already captured, pass its URL as imageUrl to skip recapture. If the screenshot requires interaction first (logging in, clicking a tab, dismissing a modal, scrolling to a section), use the instructions param to describe the steps. To render a fixed-size image such as an Open Graph share card, set `width` and `height` (e.g. 1200 \xD7 630) and `format: 'png'`: the tool navigates to `path`, clips to exactly those pixel dimensions, and returns the image URL.",
+  inputSchema: {
+    type: "object",
+    properties: {
+      fullPage: {
+        type: "boolean",
+        description: "true = full-height capture of the entire page; false = just the visible viewport. Pick based on whether you need the whole page or a specific section."
       },
-      required: ["fullPage"]
-    }
-  },
-  async execute(input, context) {
-    const fullPage = input.fullPage === true;
-    const shotKind = fullPage ? "full-page" : "viewport";
-    try {
-      if (input.imageUrl) {
-        return await captureAndAnalyzeScreenshot({
-          prompt: input.prompt,
-          imageUrl: input.imageUrl,
-          onLog: context?.onLog,
-          model: resolveModel("imageAnalysis", context?.models, context?.model)
-        });
-      }
-      if (input.instructions && context) {
-        const task = input.path ? `Navigate to "${input.path}", then: ${input.instructions}. After completing these steps, take a ${shotKind} screenshot.` : `${input.instructions}. After completing these steps, take a ${shotKind} screenshot.`;
-        const result = await runBrowserAutomation(task, context, {
-          capture: fullPage ? "fullPage" : "viewport"
-        });
-        if (!result.screenshot) {
-          return result.text;
-        }
-        return await streamScreenshotAnalysis({
-          url: result.screenshot.url,
-          prompt: input.prompt,
-          styleMap: result.screenshot.styleMap,
-          onLog: context?.onLog,
-          model: resolveModel("imageAnalysis", context?.models, context?.model)
-        });
+      prompt: {
+        type: "string",
+        description: "Optional question about the screenshot. If omitted, returns a general description of what's visible."
+      },
+      imageUrl: {
+        type: "string",
+        description: "URL of an existing screenshot to analyze instead of capturing a new one. Use this for additional questions about a previous screenshot."
+      },
+      path: {
+        type: "string",
+        description: 'Navigate to this path before capturing (e.g. "/settings", "/dashboard"). If omitted, screenshots the current page.'
+      },
+      width: {
+        type: "number",
+        description: "Exact capture width in pixels. Set together with `height` to render a fixed-size image; clips to exactly this viewport instead of the default preview size."
+      },
+      height: {
+        type: "number",
+        description: "Exact capture height in pixels. Set together with `width`."
+      },
+      format: {
+        type: "string",
+        enum: ["png", "jpeg"],
+        description: "Output image format. Defaults to 'jpeg'. Use 'png' for crisp flat graphics like share cards, where JPEG artifacts show on sharp type and edges."
+      },
+      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, scrolling to a section, getting through a login/auth checkpoint), describe the steps to get there. A browser automation agent will follow these instructions, then capture per your `fullPage` choice \u2014 so with `fullPage: false` you can scroll to a section and capture just that viewport. It can bypass auth and get right to where it needs to be if you tell it to authenticate as a test user and give it the path/screen to start its test at. Never describe what names or values to use when applying the instructions - the browser automation agent must use its own values for it to work properly. If a specific auth role is required to access the content, be sure to note that - it can automatically assume it for the purpose of testing. Use only when interaction is required to *reach* the state you want to capture \u2014 log in, dismiss a modal, switch a tab, follow a route, scroll to a section. If your steps are exercising the app's functionality across multiple states (running flows, asserting behavior under interaction, multi-step QA), use `runAutomatedBrowserTest` instead."
       }
-      const release = await acquireBrowserLock();
-      try {
-        return await captureAndAnalyzeScreenshot({
-          prompt: input.prompt,
-          path: input.path,
-          fullPage,
-          width: input.width,
-          height: input.height,
-          format: input.format,
-          onLog: context?.onLog,
-          model: resolveModel("imageAnalysis", context?.models, context?.model)
-        });
-      } finally {
-        release();
+    },
+    required: ["fullPage"]
+  }
+};
+async function executeScreenshot(input, onLog, context) {
+  const fullPage = input.fullPage === true;
+  const model = resolveModel("imageAnalysis", context?.models, context?.model);
+  try {
+    if (input.imageUrl) {
+      return await captureAndAnalyzeScreenshot({
+        prompt: input.prompt,
+        imageUrl: input.imageUrl,
+        onLog,
+        model
+      });
+    }
+    if (input.instructions && context) {
+      const shotKind = fullPage ? "full-page" : "viewport";
+      const task = input.path ? `Navigate to "${input.path}", then: ${input.instructions}. After completing these steps, take a ${shotKind} screenshot.` : `${input.instructions}. After completing these steps, take a ${shotKind} screenshot.`;
+      const result = await runBrowserAutomation(task, context, {
+        capture: fullPage ? "fullPage" : "viewport"
+      });
+      if (!result.screenshot) {
+        return result.text;
       }
-    } catch (err) {
-      return `Error taking screenshot: ${err.message}`;
+      return await streamScreenshotAnalysis({
+        url: result.screenshot.url,
+        prompt: input.prompt,
+        styleMap: result.screenshot.styleMap,
+        onLog,
+        model
+      });
     }
+    const release = await acquireBrowserLock();
+    try {
+      return await captureAndAnalyzeScreenshot({
+        prompt: input.prompt,
+        path: input.path,
+        fullPage,
+        width: input.width,
+        height: input.height,
+        format: input.format,
+        onLog,
+        model
+      });
+    } finally {
+      release();
+    }
+  } catch (err) {
+    return `Error taking screenshot: ${err.message}`;
   }
+}
+var screenshotTool = {
+  clearable: true,
+  definition: screenshotDefinition,
+  execute: (input, context) => executeScreenshot(input, context?.onLog, context)
 };
 
 // src/subagents/designExpert/tools/searchGoogle.ts
@@ -4603,83 +4610,11 @@ async function execute4(input, onLog, context) {
   return JSON.stringify({ url: imageUrl, analysis });
 }
 
-// src/subagents/designExpert/tools/screenshot.ts
-var screenshot_exports = {};
-__export(screenshot_exports, {
-  definition: () => definition5,
-  execute: () => execute5
-});
-var definition5 = {
-  clearable: true,
-  name: "screenshot",
-  description: "Capture a screenshot of the current app preview and get it back with visual analysis. Choose `fullPage`: `false` captures just the visible viewport (fast \u2014 use it to review a specific section the page is scrolled to), `true` captures the entire page top-to-bottom (slower \u2014 use it to review overall composition or a layout you can't see in one screen). 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.",
-  inputSchema: {
-    type: "object",
-    properties: {
-      fullPage: {
-        type: "boolean",
-        description: "true = full-height capture of the entire page; false = just the visible viewport. Pick based on whether you need the whole page or a specific section."
-      },
-      prompt: {
-        type: "string",
-        description: "Optional specific question about the screenshot. Use a bulleted list to ask many questions at once."
-      },
-      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, scrolling to a specific section, getting through a login/auth checkpoint), describe the steps to get there. A browser automation agent will follow these instructions, then capture per your `fullPage` choice \u2014 so with `fullPage: false` you can scroll to a section and capture just that viewport. It can bypass auth and get right to where it needs to be if you tell it to authenticate as a test user and give it the path/screen to start at. Never describe what names or values to use when applying the instructions - the browser automation agent must use its own values for it to work properly. If a specific auth role is required to access the content, be sure to note that - it can automatically assume it for the purpose of testing."
-      }
-    },
-    required: ["fullPage"]
-  }
-};
-async function execute5(input, onLog, context) {
-  const fullPage = input.fullPage === true;
-  const shotKind = fullPage ? "full-page" : "viewport";
-  if (input.instructions && context) {
-    try {
-      const task = input.path ? `Navigate to "${input.path}", then: ${input.instructions}. After completing these steps, take a ${shotKind} screenshot.` : `${input.instructions}. After completing these steps, take a ${shotKind} screenshot.`;
-      const result = await runBrowserAutomation(task, context, {
-        capture: fullPage ? "fullPage" : "viewport"
-      });
-      if (!result.screenshot) {
-        return result.text;
-      }
-      return await streamScreenshotAnalysis({
-        url: result.screenshot.url,
-        prompt: input.prompt,
-        styleMap: result.screenshot.styleMap,
-        onLog,
-        model: resolveModel("imageAnalysis", context?.models, context?.model)
-      });
-    } catch (err) {
-      return `Error taking interactive screenshot: ${err.message}`;
-    }
-  }
-  const release = await acquireBrowserLock();
-  try {
-    return await captureAndAnalyzeScreenshot({
-      prompt: input.prompt,
-      path: input.path,
-      fullPage,
-      onLog,
-      model: resolveModel("imageAnalysis", context?.models, context?.model)
-    });
-  } catch (err) {
-    return `Error taking screenshot: ${err.message}`;
-  } finally {
-    release();
-  }
-}
-
 // src/subagents/designExpert/tools/images/generateImages.ts
 var generateImages_exports = {};
 __export(generateImages_exports, {
-  definition: () => definition6,
-  execute: () => execute6
+  definition: () => definition5,
+  execute: () => execute5
 });
 
 // src/subagents/designExpert/tools/images/enhancePrompt.ts
@@ -4872,7 +4807,7 @@ async function generateImageAssets(opts) {
 }
 
 // src/subagents/designExpert/tools/images/generateImages.ts
-var definition6 = {
+var definition5 = {
   clearable: false,
   name: "generateImages",
   description: "Generate images. Returns CDN URLs with a quality analysis for each image. Produces high-quality results for everything from photorealistic images and abstract/creative visuals. Pass multiple prompts to generate in parallel. No need to analyze images separately after generating \u2014 the analysis is included.",
@@ -4906,7 +4841,7 @@ var definition6 = {
     required: ["prompts"]
   }
 };
-async function execute6(input, onLog, context) {
+async function execute5(input, onLog, context) {
   return generateImageAssets({
     prompts: input.prompts,
     width: input.width,
@@ -4936,10 +4871,10 @@ async function execute6(input, onLog, context) {
 // src/subagents/designExpert/tools/images/editImages.ts
 var editImages_exports = {};
 __export(editImages_exports, {
-  definition: () => definition7,
-  execute: () => execute7
+  definition: () => definition6,
+  execute: () => execute6
 });
-var definition7 = {
+var definition6 = {
   clearable: false,
   name: "editImages",
   description: "Edit or transform existing images. Provide one or more source image URLs as reference and a prompt describing the desired edit. Use for compositing, style transfer, subject transformation, blending multiple references, or incorporating one or more references into something new. Returns CDN URLs with analysis.",
@@ -4976,7 +4911,7 @@ var definition7 = {
     required: ["prompts", "sourceImages"]
   }
 };
-async function execute7(input, onLog, context) {
+async function execute6(input, onLog, context) {
   return generateImageAssets({
     prompts: input.prompts,
     sourceImages: input.sourceImages,
@@ -5006,8 +4941,8 @@ async function execute7(input, onLog, context) {
 // src/subagents/designExpert/tools/polishCopy.ts
 var polishCopy_exports = {};
 __export(polishCopy_exports, {
-  definition: () => definition8,
-  execute: () => execute8
+  definition: () => definition7,
+  execute: () => execute7
 });
 
 // src/subagents/copyEditor/tools.ts
@@ -5064,7 +4999,7 @@ var copyEditorTool = {
 };
 
 // src/subagents/designExpert/tools/polishCopy.ts
-var definition8 = {
+var definition7 = {
   clearable: false,
   name: "polishCopy",
   description: "Hand off any user-facing copy you've written \u2014 headlines, captions, labels, body text \u2014 and get back a sharper version: better built for its audience and free of the fingerprints that make writing read as AI. It elevates how the copy communicates without inventing facts or claims you didn't give it. Give it the text plus what it's for (where it appears, the audience).",
@@ -5079,7 +5014,7 @@ var definition8 = {
     required: ["task"]
   }
 };
-async function execute8(input, _onLog, context) {
+async function execute7(input, _onLog, context) {
   return copyEditorTool.execute(input, context);
 }
 
@@ -5089,7 +5024,10 @@ var tools = {
   scrapeWebUrl: scrapeWebUrl_exports,
   analyzeDesign: analyzeDesign_exports,
   analyzeImage: analyzeImage_exports,
-  screenshot: screenshot_exports,
+  // Same tool the main agent offers, imported rather than reimplemented — the
+  // two used to be near-identical copies and had already drifted apart. Its core
+  // already takes (input, onLog, context), which is this registry's convention.
+  screenshot: { definition: screenshotDefinition, execute: executeScreenshot },
   generateImages: generateImages_exports,
   editImages: editImages_exports,
   polishCopy: polishCopy_exports

package/dist/index.js

@@ -2238,6 +2238,9 @@ function findLastSummaryCheckpoint(messages, name) {
   }
   return -1;
 }
+function portableToolCallId(id) {
+  return id.replace(/[^a-zA-Z0-9_-]/g, "_");
+}
 function fixOrphanedToolCalls(messages) {
   const toolResultIds = /* @__PURE__ */ new Set();
   for (const msg of messages) {
@@ -2329,14 +2332,30 @@ ${summaryBlock.text}
 
 ${content}` : attachmentHeader;
       }
-      return { ...rest, content };
+      return {
+        ...rest,
+        content,
+        ...msg.toolCallId && {
+          toolCallId: portableToolCallId(msg.toolCallId)
+        }
+      };
     }
     if (!Array.isArray(msg.content)) {
       return msg;
     }
     const blocks = msg.content;
     const text = blocks.filter((b) => b.type === "text").map((b) => b.text).join("");
-    const toolCalls = blocks.filter((b) => b.type === "tool").map((b) => ({ id: b.id, name: b.name, input: b.input }));
+    const toolBlocks = blocks.filter(
+      (b) => b.type === "tool"
+    );
+    const toolCalls = toolBlocks.map((b) => ({
+      id: portableToolCallId(b.id),
+      name: b.name,
+      input: b.input
+    }));
+    const rewroteToolIds = toolBlocks.some(
+      (b) => portableToolCallId(b.id) !== b.id
+    );
     const cleaned2 = {
       role: msg.role,
       content: text
@@ -2344,7 +2363,7 @@ ${content}` : attachmentHeader;
     if (toolCalls.length > 0) {
       cleaned2.toolCalls = toolCalls;
     }
-    if (msg.providerMetadata) {
+    if (msg.providerMetadata && !rewroteToolIds) {
       cleaned2.providerMetadata = msg.providerMetadata;
     }
     if (msg.hidden) {
@@ -3599,6 +3618,7 @@ async function sidecarRequest(endpoint, body = {}, options) {
     throw new Error("Sidecar not available");
   }
   const url = `${baseUrl}${endpoint}`;
+  let data;
   try {
     const res = await fetch(url, {
       method: "POST",
@@ -3610,12 +3630,7 @@ async function sidecarRequest(endpoint, body = {}, options) {
       log5.error("Sidecar error", { endpoint, status: res.status });
       throw new Error(`Sidecar error: ${res.status}`);
     }
-    const data = await res.json();
-    if (data?.success === false) {
-      const code = data.errorCode ? ` [${data.errorCode}]` : "";
-      throw new Error(`${data.error || "Unknown error"}${code}`);
-    }
-    return data;
+    data = await res.json();
   } catch (err) {
     if (err.message.startsWith("Sidecar error")) {
       throw err;
@@ -3623,6 +3638,16 @@ async function sidecarRequest(endpoint, body = {}, options) {
     log5.error("Sidecar connection error", { endpoint, error: err.message });
     throw new Error(`Sidecar connection error: ${err.message}`);
   }
+  if (data?.success === false) {
+    const code = data.errorCode ? ` [${data.errorCode}]` : "";
+    log5.error("Sidecar command failed", {
+      endpoint,
+      error: data.error,
+      errorCode: data.errorCode
+    });
+    throw new Error(`${data.error || "Unknown error"}${code}`);
+  }
+  return data;
 }
 var log5, baseUrl;
 var init_sidecar = __esm({
@@ -3935,7 +3960,9 @@ async function captureAndAnalyzeScreenshot(promptOrOptions) {
         ...height != null ? { height } : {},
         ...format ? { format } : {}
       },
-      { timeout: fullPage ? 12e4 : 3e4 }
+      {
+        timeout: fullPage ? FULLPAGE_CAPTURE_TIMEOUT_MS : VIEWPORT_CAPTURE_TIMEOUT_MS
+      }
     );
     url = ssResult?.url || ssResult?.screenshotUrl;
     if (!url) {
@@ -3961,12 +3988,14 @@ async function captureAndAnalyzeScreenshot(promptOrOptions) {
     model
   });
 }
-var SCREENSHOT_ANALYSIS_PROMPT, ANALYSIS_RESPONSE_FORMAT;
+var VIEWPORT_CAPTURE_TIMEOUT_MS, FULLPAGE_CAPTURE_TIMEOUT_MS, SCREENSHOT_ANALYSIS_PROMPT, ANALYSIS_RESPONSE_FORMAT;
 var init_screenshot = __esm({
   "src/tools/_helpers/screenshot.ts"() {
     "use strict";
     init_sidecar();
     init_analyzeImage();
+    VIEWPORT_CAPTURE_TIMEOUT_MS = 45e3;
+    FULLPAGE_CAPTURE_TIMEOUT_MS = 135e3;
     SCREENSHOT_ANALYSIS_PROMPT = `Describe everything visible on screen from top to bottom \u2014 every element, its position, its size relative to the viewport, its colors, its content. Be comprehensive, thorough, and spatial. After the inventory, note anything that looks visually broken (overlapping elements, clipped text, misaligned components).`;
     ANALYSIS_RESPONSE_FORMAT = `Respond only with your analysis as Markdown and absolutely no other text. Do not use emojis - use unicode if you need symbols.`;
   }
@@ -4673,7 +4702,7 @@ var init_tools2 = __esm({
                       "screenshotViewport",
                       "setViewport"
                     ],
-                    description: 'snapshot: accessibility tree of the page (waits for network to settle). click: click an element (animated cursor, full event sequence). type: type text into input (one char at a time, works with React/Vue/Svelte). select: select a dropdown option by text. wait: wait for an element to appear (polls 100ms, waits for network). navigate: navigate to a URL within the app (waits for load, subsequent steps run on new page). evaluate: run JS in the page. styles: read computed CSS styles from elements (pass properties array with camelCase names, or omit for defaults). screenshotFullPage: full-page viewport-stitched screenshot (returns CDN url with dimensions). screenshotViewport: screenshot of just the visible viewport \u2014 pass `scrollToSelector` (or `scrollY`) on this step to scroll a section into view and capture it in one atomic step (no separate scroll needed). setViewport: switch the browser between desktop and mobile rendering (pass `mode`: "desktop" or "mobile"). Reloads the page so responsive layouts, media queries, and matchMedia re-evaluate \u2014 use it to QA mobile/responsive views.'
+                    description: 'snapshot: accessibility tree of the page (waits for network to settle). click: click an element (animated cursor, full event sequence). type: type text into input (one char at a time, works with React/Vue/Svelte). select: select a dropdown option by text. wait: wait for an element to appear (polls 100ms, waits for network). navigate: navigate to a URL within the app (waits for load, subsequent steps run on new page). evaluate: run JS in the page. styles: read computed CSS styles from elements (pass properties array with camelCase names, or omit for defaults). screenshotFullPage: screenshot of the whole page top-to-bottom (returns a CDN url with dimensions and a written analysis). screenshotViewport: screenshot of just the visible viewport \u2014 pass `scrollToSelector` (or `scrollY`) on this step to scroll a section into view and capture it in one atomic step (no separate scroll needed). setViewport: switch the browser between desktop and mobile rendering (pass `mode`: "desktop" or "mobile"). Reloads the page so responsive layouts, media queries, and matchMedia re-evaluate \u2014 use it to QA mobile/responsive views.'
                   },
                   ref: {
                     type: "string",
@@ -4741,20 +4770,11 @@ var init_tools2 = __esm({
           required: ["steps"]
         }
       },
-      {
-        clearable: true,
-        name: "screenshotFullPage",
-        description: "Capture a full-height screenshot of the current page. Returns a CDN URL with full text analysis and description.",
-        inputSchema: {
-          type: "object",
-          properties: {
-            path: {
-              type: "string",
-              description: 'Navigate to this path before capturing (e.g. "/settings"). If omitted, screenshots the current page.'
-            }
-          }
-        }
-      },
+      // Captures are `browserCommand` steps only — there is deliberately no
+      // standalone screenshot tool here. Both used to exist for full-page, with
+      // different budgets, different result plumbing, and analysis on only one of
+      // them, so which door you picked changed what you got back.
+      //
       // Read tools so the QA agent can pull full spec detail on demand — the spec
       // context in its prompt is a lightweight index (see prompt.ts) that points
       // here. Routed to the global executeTool in index.ts, mirroring specSync.
@@ -4973,7 +4993,7 @@ async function runBrowserAutomation(task, context, opts) {
       );
     } catch {
     }
-    let lastBrowserCommandViewport;
+    let lastCapture = {};
     const result = await runSubAgent({
       system: getBrowserAutomationPrompt(),
       task,
@@ -4995,22 +5015,6 @@ async function runBrowserAutomation(task, context, opts) {
             return `Error setting up browser: ${err.message}`;
           }
         }
-        if (name === "screenshotFullPage") {
-          try {
-            return await captureAndAnalyzeScreenshot({
-              path: _input.path,
-              fullPage: true,
-              onLog,
-              model: resolveModel(
-                "imageAnalysis",
-                context.models,
-                context.model
-              )
-            });
-          } catch (err) {
-            return `Error taking screenshot: ${err.message}`;
-          }
-        }
         if (COMMON_READ_TOOL_NAMES.has(name) || name === readSpecTool.definition.name) {
           return executeTool(name, _input, context);
         }
@@ -5032,14 +5036,16 @@ async function runBrowserAutomation(task, context, opts) {
           try {
             const parsed = JSON.parse(result2);
             const screenshotSteps = (parsed.steps || []).filter(
-              (s) => s.command === "screenshotViewport" && s.result?.url
+              (s) => CAPTURE_COMMANDS.has(s.command) && s.result?.url
             );
             if (screenshotSteps.length > 0) {
-              const lastStep = screenshotSteps[screenshotSteps.length - 1];
-              lastBrowserCommandViewport = {
-                url: lastStep.result.url,
-                styleMap: lastStep.result.styleMap
-              };
+              for (const step of screenshotSteps) {
+                const kind = step.command === "screenshotFullPage" ? "fullPage" : "viewport";
+                lastCapture[kind] = {
+                  url: step.result.url,
+                  styleMap: step.result.styleMap
+                };
+              }
               const visionOverride = {
                 model: resolveModel(
                   "imageAnalysis",
@@ -5063,13 +5069,12 @@ async function runBrowserAutomation(task, context, opts) {
               );
               try {
                 const analyses = JSON.parse(batchResult);
-                let ai = 0;
-                for (const step of parsed.steps) {
-                  if (step.command === "screenshotViewport" && step.result?.url && ai < analyses.length) {
-                    step.result.analysis = analyses[ai]?.output?.analysis || analyses[ai]?.output || "";
-                    ai++;
+                screenshotSteps.forEach((step, i) => {
+                  if (i >= analyses.length) {
+                    return;
                   }
-                }
+                  step.result.analysis = analyses[i]?.output?.analysis || analyses[i]?.output || "";
+                });
               } catch {
                 log8.debug("Failed to parse batch analysis result", {
                   batchResult
@@ -5082,13 +5087,10 @@ async function runBrowserAutomation(task, context, opts) {
         }
         return result2;
       },
-      toolRegistry: context.toolRegistry,
-      captureArtifacts: ["screenshotFullPage"]
+      toolRegistry: context.toolRegistry
     });
     context.subAgentMessages?.set(context.toolCallId, result.messages);
-    const fullPage = result.artifacts?.screenshotFullPage;
-    const viewport = lastBrowserCommandViewport;
-    const preferred = opts?.capture === "viewport" ? viewport ?? fullPage : fullPage ?? viewport;
+    const preferred = opts?.capture === "viewport" ? lastCapture.viewport ?? lastCapture.fullPage : lastCapture.fullPage ?? lastCapture.viewport;
     return {
       text: result.text,
       ...preferred?.url ? { screenshot: { url: preferred.url, styleMap: preferred.styleMap } } : {}
@@ -5097,7 +5099,7 @@ async function runBrowserAutomation(task, context, opts) {
     release();
   }
 }
-var log8, browserAutomationTool;
+var log8, CAPTURE_COMMANDS, browserAutomationTool;
 var init_browserAutomation = __esm({
   "src/subagents/browserAutomation/index.ts"() {
     "use strict";
@@ -5114,6 +5116,7 @@ var init_browserAutomation = __esm({
     init_surfaces();
     init_logger();
     log8 = createLogger("browser-automation");
+    CAPTURE_COMMANDS = /* @__PURE__ */ new Set(["screenshotViewport", "screenshotFullPage"]);
     browserAutomationTool = {
       clearable: true,
       definition: {
@@ -5147,7 +5150,55 @@ var init_browserAutomation = __esm({
 });
 
 // src/tools/code/screenshot.ts
-var screenshotTool;
+async function executeScreenshot(input, onLog, context) {
+  const fullPage = input.fullPage === true;
+  const model = resolveModel("imageAnalysis", context?.models, context?.model);
+  try {
+    if (input.imageUrl) {
+      return await captureAndAnalyzeScreenshot({
+        prompt: input.prompt,
+        imageUrl: input.imageUrl,
+        onLog,
+        model
+      });
+    }
+    if (input.instructions && context) {
+      const shotKind = fullPage ? "full-page" : "viewport";
+      const task = input.path ? `Navigate to "${input.path}", then: ${input.instructions}. After completing these steps, take a ${shotKind} screenshot.` : `${input.instructions}. After completing these steps, take a ${shotKind} screenshot.`;
+      const result = await runBrowserAutomation(task, context, {
+        capture: fullPage ? "fullPage" : "viewport"
+      });
+      if (!result.screenshot) {
+        return result.text;
+      }
+      return await streamScreenshotAnalysis({
+        url: result.screenshot.url,
+        prompt: input.prompt,
+        styleMap: result.screenshot.styleMap,
+        onLog,
+        model
+      });
+    }
+    const release = await acquireBrowserLock();
+    try {
+      return await captureAndAnalyzeScreenshot({
+        prompt: input.prompt,
+        path: input.path,
+        fullPage,
+        width: input.width,
+        height: input.height,
+        format: input.format,
+        onLog,
+        model
+      });
+    } finally {
+      release();
+    }
+  } catch (err) {
+    return `Error taking screenshot: ${err.message}`;
+  }
+}
+var screenshotDefinition, screenshotTool;
 var init_screenshot2 = __esm({
   "src/tools/code/screenshot.ts"() {
     "use strict";
@@ -5155,99 +5206,55 @@ var init_screenshot2 = __esm({
     init_browserLock();
     init_browserAutomation();
     init_surfaces();
-    screenshotTool = {
+    screenshotDefinition = {
       clearable: true,
-      definition: {
-        name: "screenshot",
-        description: "Capture a screenshot of the app preview and get a description of what's on screen. Choose `fullPage`: `false` captures just the visible viewport (fast \u2014 for a specific section the page is scrolled to), `true` captures the entire page top-to-bottom (slower \u2014 for overall composition or content past the fold). Captures the settled page state \u2014 it cannot catch animations, transitions, or transient state. Optionally provide specific questions about what you're looking for. Use a bulleted list to ask many questions at once. To ask additional questions about a screenshot you have already captured, pass its URL as imageUrl to skip recapture. If the screenshot requires interaction first (logging in, clicking a tab, dismissing a modal, scrolling to a section), use the instructions param to describe the steps. To render a fixed-size image such as an Open Graph share card, set `width` and `height` (e.g. 1200 \xD7 630) and `format: 'png'`: the tool navigates to `path`, clips to exactly those pixel dimensions, and returns the image URL.",
-        inputSchema: {
-          type: "object",
-          properties: {
-            fullPage: {
-              type: "boolean",
-              description: "true = full-height capture of the entire page; false = just the visible viewport. Pick based on whether you need the whole page or a specific section."
-            },
-            prompt: {
-              type: "string",
-              description: "Optional question about the screenshot. If omitted, returns a general description of what's visible."
-            },
-            imageUrl: {
-              type: "string",
-              description: "URL of an existing screenshot to analyze instead of capturing a new one. Use this for additional questions about a previous screenshot."
-            },
-            path: {
-              type: "string",
-              description: 'Navigate to this path before capturing (e.g. "/settings", "/dashboard"). If omitted, screenshots the current page.'
-            },
-            width: {
-              type: "number",
-              description: "Exact capture width in pixels. Set together with `height` to render a fixed-size image; clips to exactly this viewport instead of the default preview size."
-            },
-            height: {
-              type: "number",
-              description: "Exact capture height in pixels. Set together with `width`."
-            },
-            format: {
-              type: "string",
-              enum: ["png", "jpeg"],
-              description: "Output image format. Defaults to 'jpeg'. Use 'png' for crisp flat graphics like share cards, where JPEG artifacts show on sharp type and edges."
-            },
-            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, scrolling to a section, getting through a login/auth checkpoint), describe the steps to get there. A browser automation agent will follow these instructions, then capture per your `fullPage` choice \u2014 so with `fullPage: false` you can scroll to a section and capture just that viewport. It can bypass auth and get right to where it needs to be if you tell it to authenticate as a test user and give it the path/screen to start its test at. Never describe what names or values to use when applying the instructions - the browser automation agent must use its own values for it to work properly. If a specific auth role is required to access the content, be sure to note that - it can automatically assume it for the purpose of testing. Use only when interaction is required to *reach* the state you want to capture \u2014 log in, dismiss a modal, switch a tab, follow a route, scroll to a section. If your steps are exercising the app's functionality across multiple states (running flows, asserting behavior under interaction, multi-step QA), use `runAutomatedBrowserTest` instead."
-            }
+      name: "screenshot",
+      description: "Capture a screenshot of the app preview and get a description of what's on screen. Choose `fullPage`: `false` captures just the visible viewport (fast \u2014 for a specific section the page is scrolled to), `true` captures the entire page top-to-bottom (slower \u2014 for overall composition or content past the fold). Captures the settled page state \u2014 it cannot catch animations, transitions, or transient state. The analysis is not precise about every detail \u2014 for example it cannot reliably identify specific fonts by name, only describe what the letterforms look like. Optionally provide specific questions about what you're looking for. Use a bulleted list to ask many questions at once. To ask additional questions about a screenshot you have already captured, pass its URL as imageUrl to skip recapture. If the screenshot requires interaction first (logging in, clicking a tab, dismissing a modal, scrolling to a section), use the instructions param to describe the steps. To render a fixed-size image such as an Open Graph share card, set `width` and `height` (e.g. 1200 \xD7 630) and `format: 'png'`: the tool navigates to `path`, clips to exactly those pixel dimensions, and returns the image URL.",
+      inputSchema: {
+        type: "object",
+        properties: {
+          fullPage: {
+            type: "boolean",
+            description: "true = full-height capture of the entire page; false = just the visible viewport. Pick based on whether you need the whole page or a specific section."
           },
-          required: ["fullPage"]
-        }
-      },
-      async execute(input, context) {
-        const fullPage = input.fullPage === true;
-        const shotKind = fullPage ? "full-page" : "viewport";
-        try {
-          if (input.imageUrl) {
-            return await captureAndAnalyzeScreenshot({
-              prompt: input.prompt,
-              imageUrl: input.imageUrl,
-              onLog: context?.onLog,
-              model: resolveModel("imageAnalysis", context?.models, context?.model)
-            });
-          }
-          if (input.instructions && context) {
-            const task = input.path ? `Navigate to "${input.path}", then: ${input.instructions}. After completing these steps, take a ${shotKind} screenshot.` : `${input.instructions}. After completing these steps, take a ${shotKind} screenshot.`;
-            const result = await runBrowserAutomation(task, context, {
-              capture: fullPage ? "fullPage" : "viewport"
-            });
-            if (!result.screenshot) {
-              return result.text;
-            }
-            return await streamScreenshotAnalysis({
-              url: result.screenshot.url,
-              prompt: input.prompt,
-              styleMap: result.screenshot.styleMap,
-              onLog: context?.onLog,
-              model: resolveModel("imageAnalysis", context?.models, context?.model)
-            });
-          }
-          const release = await acquireBrowserLock();
-          try {
-            return await captureAndAnalyzeScreenshot({
-              prompt: input.prompt,
-              path: input.path,
-              fullPage,
-              width: input.width,
-              height: input.height,
-              format: input.format,
-              onLog: context?.onLog,
-              model: resolveModel("imageAnalysis", context?.models, context?.model)
-            });
-          } finally {
-            release();
+          prompt: {
+            type: "string",
+            description: "Optional question about the screenshot. If omitted, returns a general description of what's visible."
+          },
+          imageUrl: {
+            type: "string",
+            description: "URL of an existing screenshot to analyze instead of capturing a new one. Use this for additional questions about a previous screenshot."
+          },
+          path: {
+            type: "string",
+            description: 'Navigate to this path before capturing (e.g. "/settings", "/dashboard"). If omitted, screenshots the current page.'
+          },
+          width: {
+            type: "number",
+            description: "Exact capture width in pixels. Set together with `height` to render a fixed-size image; clips to exactly this viewport instead of the default preview size."
+          },
+          height: {
+            type: "number",
+            description: "Exact capture height in pixels. Set together with `width`."
+          },
+          format: {
+            type: "string",
+            enum: ["png", "jpeg"],
+            description: "Output image format. Defaults to 'jpeg'. Use 'png' for crisp flat graphics like share cards, where JPEG artifacts show on sharp type and edges."
+          },
+          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, scrolling to a section, getting through a login/auth checkpoint), describe the steps to get there. A browser automation agent will follow these instructions, then capture per your `fullPage` choice \u2014 so with `fullPage: false` you can scroll to a section and capture just that viewport. It can bypass auth and get right to where it needs to be if you tell it to authenticate as a test user and give it the path/screen to start its test at. Never describe what names or values to use when applying the instructions - the browser automation agent must use its own values for it to work properly. If a specific auth role is required to access the content, be sure to note that - it can automatically assume it for the purpose of testing. Use only when interaction is required to *reach* the state you want to capture \u2014 log in, dismiss a modal, switch a tab, follow a route, scroll to a section. If your steps are exercising the app's functionality across multiple states (running flows, asserting behavior under interaction, multi-step QA), use `runAutomatedBrowserTest` instead."
           }
-        } catch (err) {
-          return `Error taking screenshot: ${err.message}`;
-        }
+        },
+        required: ["fullPage"]
       }
     };
+    screenshotTool = {
+      clearable: true,
+      definition: screenshotDefinition,
+      execute: (input, context) => executeScreenshot(input, context?.onLog, context)
+    };
   }
 });
 
@@ -5479,88 +5486,6 @@ var init_analyzeImage2 = __esm({
   }
 });
 
-// src/subagents/designExpert/tools/screenshot.ts
-var screenshot_exports = {};
-__export(screenshot_exports, {
-  definition: () => definition5,
-  execute: () => execute5
-});
-async function execute5(input, onLog, context) {
-  const fullPage = input.fullPage === true;
-  const shotKind = fullPage ? "full-page" : "viewport";
-  if (input.instructions && context) {
-    try {
-      const task = input.path ? `Navigate to "${input.path}", then: ${input.instructions}. After completing these steps, take a ${shotKind} screenshot.` : `${input.instructions}. After completing these steps, take a ${shotKind} screenshot.`;
-      const result = await runBrowserAutomation(task, context, {
-        capture: fullPage ? "fullPage" : "viewport"
-      });
-      if (!result.screenshot) {
-        return result.text;
-      }
-      return await streamScreenshotAnalysis({
-        url: result.screenshot.url,
-        prompt: input.prompt,
-        styleMap: result.screenshot.styleMap,
-        onLog,
-        model: resolveModel("imageAnalysis", context?.models, context?.model)
-      });
-    } catch (err) {
-      return `Error taking interactive screenshot: ${err.message}`;
-    }
-  }
-  const release = await acquireBrowserLock();
-  try {
-    return await captureAndAnalyzeScreenshot({
-      prompt: input.prompt,
-      path: input.path,
-      fullPage,
-      onLog,
-      model: resolveModel("imageAnalysis", context?.models, context?.model)
-    });
-  } catch (err) {
-    return `Error taking screenshot: ${err.message}`;
-  } finally {
-    release();
-  }
-}
-var definition5;
-var init_screenshot3 = __esm({
-  "src/subagents/designExpert/tools/screenshot.ts"() {
-    "use strict";
-    init_screenshot();
-    init_browserLock();
-    init_browserAutomation();
-    init_surfaces();
-    definition5 = {
-      clearable: true,
-      name: "screenshot",
-      description: "Capture a screenshot of the current app preview and get it back with visual analysis. Choose `fullPage`: `false` captures just the visible viewport (fast \u2014 use it to review a specific section the page is scrolled to), `true` captures the entire page top-to-bottom (slower \u2014 use it to review overall composition or a layout you can't see in one screen). 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.",
-      inputSchema: {
-        type: "object",
-        properties: {
-          fullPage: {
-            type: "boolean",
-            description: "true = full-height capture of the entire page; false = just the visible viewport. Pick based on whether you need the whole page or a specific section."
-          },
-          prompt: {
-            type: "string",
-            description: "Optional specific question about the screenshot. Use a bulleted list to ask many questions at once."
-          },
-          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, scrolling to a specific section, getting through a login/auth checkpoint), describe the steps to get there. A browser automation agent will follow these instructions, then capture per your `fullPage` choice \u2014 so with `fullPage: false` you can scroll to a section and capture just that viewport. It can bypass auth and get right to where it needs to be if you tell it to authenticate as a test user and give it the path/screen to start at. Never describe what names or values to use when applying the instructions - the browser automation agent must use its own values for it to work properly. If a specific auth role is required to access the content, be sure to note that - it can automatically assume it for the purpose of testing."
-          }
-        },
-        required: ["fullPage"]
-      }
-    };
-  }
-});
-
 // src/subagents/designExpert/tools/images/enhancePrompt.ts
 async function enhanceImagePrompt(params) {
   const {
@@ -5770,10 +5695,10 @@ var init_imageGenerator = __esm({
 // src/subagents/designExpert/tools/images/generateImages.ts
 var generateImages_exports = {};
 __export(generateImages_exports, {
-  definition: () => definition6,
-  execute: () => execute6
+  definition: () => definition5,
+  execute: () => execute5
 });
-async function execute6(input, onLog, context) {
+async function execute5(input, onLog, context) {
   return generateImageAssets({
     prompts: input.prompts,
     width: input.width,
@@ -5799,13 +5724,13 @@ async function execute6(input, onLog, context) {
     )
   });
 }
-var definition6;
+var definition5;
 var init_generateImages = __esm({
   "src/subagents/designExpert/tools/images/generateImages.ts"() {
     "use strict";
     init_imageGenerator();
     init_surfaces();
-    definition6 = {
+    definition5 = {
       clearable: false,
       name: "generateImages",
       description: "Generate images. Returns CDN URLs with a quality analysis for each image. Produces high-quality results for everything from photorealistic images and abstract/creative visuals. Pass multiple prompts to generate in parallel. No need to analyze images separately after generating \u2014 the analysis is included.",
@@ -5845,10 +5770,10 @@ var init_generateImages = __esm({
 // src/subagents/designExpert/tools/images/editImages.ts
 var editImages_exports = {};
 __export(editImages_exports, {
-  definition: () => definition7,
-  execute: () => execute7
+  definition: () => definition6,
+  execute: () => execute6
 });
-async function execute7(input, onLog, context) {
+async function execute6(input, onLog, context) {
   return generateImageAssets({
     prompts: input.prompts,
     sourceImages: input.sourceImages,
@@ -5874,13 +5799,13 @@ async function execute7(input, onLog, context) {
     )
   });
 }
-var definition7;
+var definition6;
 var init_editImages = __esm({
   "src/subagents/designExpert/tools/images/editImages.ts"() {
     "use strict";
     init_imageGenerator();
     init_surfaces();
-    definition7 = {
+    definition6 = {
       clearable: false,
       name: "editImages",
       description: "Edit or transform existing images. Provide one or more source image URLs as reference and a prompt describing the desired edit. Use for compositing, style transfer, subject transformation, blending multiple references, or incorporating one or more references into something new. Returns CDN URLs with analysis.",
@@ -5995,18 +5920,18 @@ var init_copyEditor = __esm({
 // src/subagents/designExpert/tools/polishCopy.ts
 var polishCopy_exports = {};
 __export(polishCopy_exports, {
-  definition: () => definition8,
-  execute: () => execute8
+  definition: () => definition7,
+  execute: () => execute7
 });
-async function execute8(input, _onLog, context) {
+async function execute7(input, _onLog, context) {
   return copyEditorTool.execute(input, context);
 }
-var definition8;
+var definition7;
 var init_polishCopy = __esm({
   "src/subagents/designExpert/tools/polishCopy.ts"() {
     "use strict";
     init_copyEditor();
-    definition8 = {
+    definition7 = {
       clearable: false,
       name: "polishCopy",
       description: "Hand off any user-facing copy you've written \u2014 headlines, captions, labels, body text \u2014 and get back a sharper version: better built for its audience and free of the fingerprints that make writing read as AI. It elevates how the copy communicates without inventing facts or claims you didn't give it. Give it the text plus what it's for (where it appears, the audience).",
@@ -6043,16 +5968,19 @@ var init_tools4 = __esm({
     init_scrapeWebUrl();
     init_analyzeDesign();
     init_analyzeImage2();
-    init_screenshot3();
     init_generateImages();
     init_editImages();
     init_polishCopy();
+    init_screenshot2();
     tools = {
       searchGoogle: searchGoogle_exports,
       scrapeWebUrl: scrapeWebUrl_exports,
       analyzeDesign: analyzeDesign_exports,
       analyzeImage: analyzeImage_exports,
-      screenshot: screenshot_exports,
+      // Same tool the main agent offers, imported rather than reimplemented — the
+      // two used to be near-identical copies and had already drifted apart. Its core
+      // already takes (input, onLog, context), which is this registry's convention.
+      screenshot: { definition: screenshotDefinition, execute: executeScreenshot },
       generateImages: generateImages_exports,
       editImages: editImages_exports,
       polishCopy: polishCopy_exports

package/dist/subagents/browserAutomation/prompt.md

@@ -43,6 +43,7 @@ Note: the snapshot concatenates inline text and strips whitespace. If you need t
 - `navigate`: Navigate to a new URL within the app. Waits for the new page to load before continuing with subsequent steps. Use this instead of evaluate with `window.location.href` when you need to navigate and then continue interacting with the new page. Steps after navigate execute on the new page automatically.
 - `evaluate`: Run arbitrary JavaScript in the page and return the result.
 - `styles`: Read computed CSS styles from page elements. Pass a `properties` array with camelCase CSS property names (e.g., `["backgroundColor", "borderRadius", "fontSize"]`). Omit `properties` for a default set covering colors, typography, spacing, borders, shadows, dimensions, and layout. Uses the same targeting as click/type (ref, text, role, label, selector). Omit the target to get styles for all elements from the last snapshot.
+- `screenshotFullPage`: Take a screenshot of the whole page, top to bottom. Returns CDN url with full text analysis and dimensions. Use for overall composition or content past the fold.
 - `screenshotViewport`: Take a screenshot of the visible viewport. Returns CDN url with full text analysis and dimensions. To capture a specific section, set `scrollToSelector` (a CSS selector) — or `scrollY` (an absolute offset) — on this same step; it scrolls the target into view and captures it atomically, so you do NOT need a separate scroll step. Do not use if you can get what you need with other tools - only use when you need to visually see the viewport.
 - `setViewport`: Switch the browser between desktop and mobile rendering. Set `mode` to `"desktop"` or `"mobile"`. Mobile emulates a phone (390-wide, touch, device pixel ratio 2); desktop is the standard wide viewport. This reloads the page so media queries, responsive layouts, and `matchMedia` re-evaluate — the reload clears in-page state, so switch before you set up the state you want to inspect. The mode persists across navigations within a run. Each run starts in the app's default mode, so only use this when you need to check the other one.
 
@@ -161,7 +162,7 @@ Check a count with evaluate:
 
 ### Final Screenshot
 How you take the final screenshot depends on what the task asked for:
-- **Whole page** → use the standalone `screenshotFullPage` tool. It takes a full-height screenshot of the current page and returns the URL plus a full-text description.
+- **Whole page** → use a `browserCommand` batch ending in a `screenshotFullPage` step. Returns the URL plus a full-text description.
 - **A specific section / viewport** → use a `browserCommand` batch ending in a `screenshotViewport` step with `scrollToSelector` set to the section (e.g. `{ "command": "screenshotViewport", "scrollToSelector": "#pricing" }`). This scrolls the section into view and captures it in one atomic step. Do this rather than a separate scroll step followed by a capture — capturing the viewport is only reliable when the scroll and the shot are in the same step.
 
 <rules>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mindstudio-ai/remy",
-  "version": "0.1.255",
+  "version": "0.1.256",
   "description": "Remy coding agent",
   "repository": {
     "type": "git",