@mindstudio-ai/remy 0.1.15 → 0.1.17

package/dist/compiled/msfm.md

@@ -112,7 +112,10 @@ A spec starts with YAML frontmatter followed by freeform Markdown. There's no ma
 **Frontmatter fields:**
 - `name` (required) — display name for the spec file
 - `description` (optional) — short summary of what this file covers
-- `type` (optional) — defaults to `spec`. Other values: `design/color` (color palette definition), `design/typography` (font and type style definition). The frontend renders these types with specialized editors.
+- `type` (optional) — defaults to `spec`. Other values: `design/color` (color palette definition), `design/typography` (font and type style definition), `roadmap` (feature roadmap item). The frontend renders these types with specialized editors.
+- `status` (roadmap only) — `done`, `in-progress`, or `not-started`
+- `requires` (roadmap only) — array of slugs for prerequisite roadmap items. Empty array means available now.
+- `effort` (roadmap only) — `quick`, `small`, `medium`, or `large`
 
 ```markdown
 ---
@@ -187,3 +190,32 @@ styles:
     description: Default reading text
 ```
 ```
+
+Roadmap item example (one file per feature in `src/roadmap/`):
+
+```markdown
+---
+name: Share & Export
+type: roadmap
+status: not-started
+description: Share haikus as image cards to social media or download as prints.
+requires: []
+effort: medium
+---
+
+Share haikus as styled image cards on social media or download as prints.
+The card system generates images using the brand's typography and color
+palette, creating shareable assets that feel native to the app's identity.
+
+~~~
+Use generateImage with Seedream to create styled cards. Card template
+applies brand typography and colors from the spec. Export as PNG via
+CDN transform at 2x resolution. Social sharing via Web Share API with
+clipboard fallback for unsupported browsers.
+~~~
+
+## History
+
+- **2026-03-22** — Built card generation using generateImage with Seedream.
+  Added share button to haiku detail view.
+```

package/dist/compiled/platform.md

@@ -23,6 +23,7 @@ my-app/
       web.md                           web UI spec
       api.md                           API conventions
       cron.md                          scheduled job descriptions
+    roadmap/                           feature roadmap (one file per item, type: roadmap)
 
   dist/                              ← compiled output (code + config)
     methods/                           backend contract
@@ -60,6 +61,7 @@ my-app/
 | Interface configs | `dist/interfaces/*/interface.json` | One per non-web interface type |
 | Specs | `src/*.md` | Natural language, MSFM format |
 | Brand identity | `src/interfaces/@brand/` | visual.md (aesthetic), colors.md (palette), typography.md (fonts), voice.md (tone), assets/ |
+| Roadmap | `src/roadmap/*.md` | Feature roadmap items (type: roadmap). One file per feature with status, dependencies, and history. |
 | Reference material | `src/references/` | Context for the agent, not consumed by platform |
 
 ## The Two SDKs

package/dist/compiled/sdk-actions.md

@@ -2,7 +2,9 @@
 
 `@mindstudio-ai/agent` provides access to 200+ AI models and 1,000+ actions through a single API key. No separate provider keys needed. MindStudio routes to the correct provider (OpenAI, Anthropic, Google, etc.) server-side.
 
-There is a huge amount of capability here: hundreds of text generation models (OpenAI, Anthropic, Google, Meta, Mistral, and more), dozens of image generation models (FLUX, DALL-E, Stable Diffusion, Ideogram, and more), video generation, text-to-speech, music generation, vision analysis, web scraping, 850+ OAuth connectors, and much more. The tables below are a summary. **Always use the `askMindStudioSdk` tool to look up exact method signatures, model IDs, and config options before writing code that uses the SDK.** The SDK assistant knows every action, every model, every connector, and the user's configured OAuth connections. Don't guess at parameters or model IDs from memory.
+There is a huge amount of capability here: hundreds of text generation models (OpenAI, Anthropic, Google, Meta, Mistral, and more), dozens of image generation models (FLUX, DALL-E, Stable Diffusion, Ideogram, and more), video generation, text-to-speech, music generation, vision analysis, web scraping, 850+ OAuth connectors, and much more. The tables below are a summary.
+
+**Always use `askMindStudioSdk` before writing code that uses the SDK.** Treat it as an expert consultant, not a docs search. Describe what you're trying to build at the method level — the full workflow, not just "how do I call generateText." The assistant knows every action, model, connector, configuration option, and the user's configured OAuth connections. It can advise on AI orchestration patterns (structured output, chaining calls, batch processing), help you avoid common mistakes (like manually parsing JSON when the SDK has structured output options), and provide complete working code for your use case.
 
 ## Usage in Methods
 

package/dist/headless.js

@@ -1,7 +1,7 @@
 // src/headless.ts
 import { createInterface } from "readline";
-import fs16 from "fs";
-import path9 from "path";
+import fs17 from "fs";
+import path10 from "path";
 
 // src/config.ts
 import fs2 from "fs";
@@ -1245,13 +1245,20 @@ import { exec } from "child_process";
 var askMindStudioSdkTool = {
   definition: {
     name: "askMindStudioSdk",
-    description: "Ask the MindStudio SDK assistant about available actions, AI models, connectors, and integrations using natural language. Returns code examples with correct method signatures, model IDs, and config options. Always use this to verify correct SDK usage, especially model IDs and configuration options. Describe what you need, not what API methods you need; the assistant will figure out the right approach. This runs its own LLM call so it has a few seconds of latency; batch multiple questions into a single query.",
+    description: `An expert consultant on building with the MindStudio SDK. Knows every action, model, connector, and configuration option. Use this as an architect, not just a docs lookup:
+
+- Describe what you're trying to build at the method level ("I need a method that takes user text, generates a summary with GPT, extracts entities, and returns structured JSON") and get back architectural guidance + working code.
+- Ask about AI orchestration patterns: structured output, chaining model calls, batch processing, streaming, error handling.
+- Ask about connectors and integrations: what's available, whether the user has configured it, how to use it.
+- Always use this before writing SDK code. Model IDs, config options, and action signatures change frequently. Don't guess.
+
+Batch related questions into a single query. This runs its own LLM call so it has a few seconds of latency.`,
     inputSchema: {
       type: "object",
       properties: {
         query: {
           type: "string",
-          description: "Natural language question about the SDK."
+          description: "Describe what you want to build or what you need to know. Be specific about the goal, not just the API method."
         }
       },
       required: ["query"]
@@ -2428,20 +2435,6 @@ var DESIGN_RESEARCH_TOOLS = [
       required: ["url"]
     }
   },
-  {
-    name: "searchStockPhotos",
-    description: 'Search Pexels for stock photos. Returns image URLs with descriptions. Use concrete, descriptive queries ("person working at laptop in modern office" not "business").',
-    inputSchema: {
-      type: "object",
-      properties: {
-        query: {
-          type: "string",
-          description: "What kind of photo to search for."
-        }
-      },
-      required: ["query"]
-    }
-  },
   {
     name: "searchProductScreenshots",
     description: 'Search for screenshots of real products and apps. Use to find what existing products look like ("stripe dashboard", "linear app", "notion workspace"). Returns image results of actual product UI. Use this for layout and design research on real products, NOT for abstract design inspiration.',
@@ -2480,32 +2473,6 @@ var DESIGN_RESEARCH_TOOLS = [
       },
       required: ["prompts"]
     }
-  },
-  {
-    name: "editImage",
-    description: "Edit an existing image using a text instruction. Takes a source image URL and a prompt describing the edits (color grading, style transfer, modifications, adding/removing elements). Returns a new CDN URL.",
-    inputSchema: {
-      type: "object",
-      properties: {
-        imageUrl: {
-          type: "string",
-          description: "URL of the source image to edit."
-        },
-        prompt: {
-          type: "string",
-          description: 'What to change. Describe the edit as an instruction: "apply warm golden hour color grading", "make the background darker", "add a subtle film grain texture".'
-        },
-        width: {
-          type: "number",
-          description: "Output width in pixels. Default 2048. Range: 2048-4096."
-        },
-        height: {
-          type: "number",
-          description: "Output height in pixels. Default 2048. Range: 2048-4096."
-        }
-      },
-      required: ["imageUrl", "prompt"]
-    }
   }
 ];
 function runCli(cmd) {
@@ -2565,12 +2532,6 @@ async function executeDesignTool(name, input) {
 
 ${analysis}`;
     }
-    case "searchStockPhotos": {
-      const encodedQuery = encodeURIComponent(input.query);
-      return runCli(
-        `mindstudio scrape-url --url "https://www.pexels.com/search/${encodedQuery}/" --page-options '{"onlyMainContent": true}' --no-meta`
-      );
-    }
     case "searchProductScreenshots": {
       const query = `${input.product} product screenshot UI 2026`;
       return runCli(
@@ -2605,24 +2566,6 @@ ${analysis}`;
       }));
       return runCli(`mindstudio batch '${JSON.stringify(steps)}' --no-meta`);
     }
-    case "editImage": {
-      const width = input.width || 2048;
-      const height = input.height || 2048;
-      const step = JSON.stringify({
-        prompt: input.prompt,
-        imageModelOverride: {
-          model: "seedream-4.5",
-          config: {
-            images: [input.imageUrl],
-            width,
-            height
-          }
-        }
-      });
-      return runCli(
-        `mindstudio generate-image '${step}' --output-key imageUrl --no-meta`
-      );
-    }
     default:
       return `Error: unknown tool "${name}"`;
   }
@@ -2633,8 +2576,8 @@ import fs14 from "fs";
 import path8 from "path";
 var base2 = import.meta.dirname ?? path8.dirname(new URL(import.meta.url).pathname);
 function resolvePath(filename) {
-  const local2 = path8.join(base2, filename);
-  return fs14.existsSync(local2) ? local2 : path8.join(base2, "subagents", "designExpert", filename);
+  const local3 = path8.join(base2, filename);
+  return fs14.existsSync(local3) ? local3 : path8.join(base2, "subagents", "designExpert", filename);
 }
 function readFile(filename) {
   return fs14.readFileSync(resolvePath(filename), "utf-8").trim();
@@ -2773,6 +2716,152 @@ var designExpertTool = {
   }
 };
 
+// src/subagents/productVision/index.ts
+import fs15 from "fs";
+import path9 from "path";
+var base3 = import.meta.dirname ?? path9.dirname(new URL(import.meta.url).pathname);
+var local2 = path9.join(base3, "prompt.md");
+var PROMPT_PATH2 = fs15.existsSync(local2) ? local2 : path9.join(base3, "subagents", "productVision", "prompt.md");
+var BASE_PROMPT = fs15.readFileSync(PROMPT_PATH2, "utf-8").trim();
+function loadSpecContext() {
+  const specDir = "src";
+  const files = [];
+  function walk(dir) {
+    try {
+      for (const entry of fs15.readdirSync(dir, { withFileTypes: true })) {
+        const full = path9.join(dir, entry.name);
+        if (entry.isDirectory()) {
+          if (entry.name !== "roadmap") {
+            walk(full);
+          }
+        } else if (entry.name.endsWith(".md")) {
+          files.push(full);
+        }
+      }
+    } catch {
+    }
+  }
+  walk(specDir);
+  if (files.length === 0) {
+    return "";
+  }
+  const sections = files.map((f) => {
+    try {
+      const content = fs15.readFileSync(f, "utf-8").trim();
+      return `<file path="${f}">
+${content}
+</file>`;
+    } catch {
+      return "";
+    }
+  }).filter(Boolean);
+  return `<spec_files>
+${sections.join("\n\n")}
+</spec_files>`;
+}
+var VISION_TOOLS = [
+  {
+    name: "writeRoadmapItem",
+    description: "Write a roadmap item to src/roadmap/. Call this once for each idea.",
+    inputSchema: {
+      type: "object",
+      properties: {
+        slug: {
+          type: "string",
+          description: 'Kebab-case filename (without .md). e.g. "ai-weekly-digest"'
+        },
+        name: {
+          type: "string",
+          description: "User-facing feature name."
+        },
+        description: {
+          type: "string",
+          description: "Short user-facing summary (1-2 sentences)."
+        },
+        effort: {
+          type: "string",
+          enum: ["quick", "small", "medium", "large"]
+        },
+        requires: {
+          type: "array",
+          items: { type: "string" },
+          description: "Slugs of prerequisite roadmap items. Empty array if independent."
+        },
+        body: {
+          type: "string",
+          description: "Full MSFM body: prose description for the user, followed by ~~~annotation~~~ with technical implementation notes for the building agent."
+        }
+      },
+      required: ["slug", "name", "description", "effort", "requires", "body"]
+    }
+  }
+];
+async function executeVisionTool(name, input) {
+  if (name !== "writeRoadmapItem") {
+    return `Error: unknown tool "${name}"`;
+  }
+  const { slug, name: itemName, description, effort, requires, body } = input;
+  const dir = "src/roadmap";
+  const filePath = path9.join(dir, `${slug}.md`);
+  try {
+    fs15.mkdirSync(dir, { recursive: true });
+    const requiresYaml = requires.length === 0 ? "[]" : `[${requires.map((r) => `"${r}"`).join(", ")}]`;
+    const content = `---
+name: ${itemName}
+type: roadmap
+status: ${slug === "mvp" ? "in-progress" : "not-started"}
+description: ${description}
+effort: ${effort}
+requires: ${requiresYaml}
+---
+
+${body}
+`;
+    fs15.writeFileSync(filePath, content, "utf-8");
+    return `Wrote ${filePath}`;
+  } catch (err) {
+    return `Error writing ${filePath}: ${err.message}`;
+  }
+}
+var productVisionTool = {
+  definition: {
+    name: "productVision",
+    description: `A product visionary that imagines where the project could go next. It automatically reads all spec files from src/ for context. Pass a brief description of the app and who it's for. It generates 10-15 ambitious, creative roadmap ideas and writes them directly to src/roadmap/. Use this at the end of spec authoring to populate the roadmap.`,
+    inputSchema: {
+      type: "object",
+      properties: {
+        task: {
+          type: "string",
+          description: "Brief description of the app and who it's for. The tool reads the full spec files automatically \u2014 no need to repeat their contents."
+        }
+      },
+      required: ["task"]
+    }
+  },
+  async execute(input, context) {
+    if (!context) {
+      return "Error: product vision requires execution context";
+    }
+    const specContext = loadSpecContext();
+    const system = specContext ? `${BASE_PROMPT}
+
+${specContext}` : BASE_PROMPT;
+    return runSubAgent({
+      system,
+      task: input.task,
+      tools: VISION_TOOLS,
+      externalTools: /* @__PURE__ */ new Set(),
+      executeTool: executeVisionTool,
+      apiConfig: context.apiConfig,
+      model: context.model,
+      signal: context.signal,
+      parentToolId: context.toolCallId,
+      onEvent: context.onEvent,
+      resolveExternalTool: context.resolveExternalTool
+    });
+  }
+};
+
 // src/tools/index.ts
 function getSpecTools() {
   return [readSpecTool, writeSpecTool, editSpecTool, listSpecFilesTool];
@@ -2806,7 +2895,8 @@ function getCommonTools() {
     fetchUrlTool,
     searchGoogleTool,
     setProjectNameTool,
-    designExpertTool
+    designExpertTool,
+    productVisionTool
   ];
 }
 function getPostOnboardingTools() {
@@ -2853,11 +2943,11 @@ function executeTool(name, input, context) {
 }
 
 // src/session.ts
-import fs15 from "fs";
+import fs16 from "fs";
 var SESSION_FILE = ".remy-session.json";
 function loadSession(state) {
   try {
-    const raw = fs15.readFileSync(SESSION_FILE, "utf-8");
+    const raw = fs16.readFileSync(SESSION_FILE, "utf-8");
     const data = JSON.parse(raw);
     if (Array.isArray(data.messages) && data.messages.length > 0) {
       state.messages = sanitizeMessages(data.messages);
@@ -2899,7 +2989,7 @@ function sanitizeMessages(messages) {
 }
 function saveSession(state) {
   try {
-    fs15.writeFileSync(
+    fs16.writeFileSync(
       SESSION_FILE,
       JSON.stringify({ messages: state.messages }, null, 2),
       "utf-8"
@@ -2910,7 +3000,7 @@ function saveSession(state) {
 function clearSession(state) {
   state.messages = [];
   try {
-    fs15.unlinkSync(SESSION_FILE);
+    fs16.unlinkSync(SESSION_FILE);
   } catch {
   }
 }
@@ -3531,10 +3621,10 @@ async function runTurn(params) {
 }
 
 // src/headless.ts
-var BASE_DIR = import.meta.dirname ?? path9.dirname(new URL(import.meta.url).pathname);
-var ACTIONS_DIR = path9.join(BASE_DIR, "actions");
+var BASE_DIR = import.meta.dirname ?? path10.dirname(new URL(import.meta.url).pathname);
+var ACTIONS_DIR = path10.join(BASE_DIR, "actions");
 function loadActionPrompt(name) {
-  return fs16.readFileSync(path9.join(ACTIONS_DIR, `${name}.md`), "utf-8").trim();
+  return fs17.readFileSync(path10.join(ACTIONS_DIR, `${name}.md`), "utf-8").trim();
 }
 function emit(event, data) {
   process.stdout.write(JSON.stringify({ event, ...data }) + "\n");

package/dist/index.js

@@ -1021,13 +1021,20 @@ var init_askMindStudioSdk = __esm({
     askMindStudioSdkTool = {
       definition: {
         name: "askMindStudioSdk",
-        description: "Ask the MindStudio SDK assistant about available actions, AI models, connectors, and integrations using natural language. Returns code examples with correct method signatures, model IDs, and config options. Always use this to verify correct SDK usage, especially model IDs and configuration options. Describe what you need, not what API methods you need; the assistant will figure out the right approach. This runs its own LLM call so it has a few seconds of latency; batch multiple questions into a single query.",
+        description: `An expert consultant on building with the MindStudio SDK. Knows every action, model, connector, and configuration option. Use this as an architect, not just a docs lookup:
+
+- Describe what you're trying to build at the method level ("I need a method that takes user text, generates a summary with GPT, extracts entities, and returns structured JSON") and get back architectural guidance + working code.
+- Ask about AI orchestration patterns: structured output, chaining model calls, batch processing, streaming, error handling.
+- Ask about connectors and integrations: what's available, whether the user has configured it, how to use it.
+- Always use this before writing SDK code. Model IDs, config options, and action signatures change frequently. Don't guess.
+
+Batch related questions into a single query. This runs its own LLM call so it has a few seconds of latency.`,
         inputSchema: {
           type: "object",
           properties: {
             query: {
               type: "string",
-              description: "Natural language question about the SDK."
+              description: "Describe what you want to build or what you need to know. Be specific about the goal, not just the API method."
             }
           },
           required: ["query"]
@@ -2363,12 +2370,6 @@ async function executeDesignTool(name, input) {
 
 ${analysis}`;
     }
-    case "searchStockPhotos": {
-      const encodedQuery = encodeURIComponent(input.query);
-      return runCli(
-        `mindstudio scrape-url --url "https://www.pexels.com/search/${encodedQuery}/" --page-options '{"onlyMainContent": true}' --no-meta`
-      );
-    }
     case "searchProductScreenshots": {
       const query = `${input.product} product screenshot UI 2026`;
       return runCli(
@@ -2403,24 +2404,6 @@ ${analysis}`;
       }));
       return runCli(`mindstudio batch '${JSON.stringify(steps)}' --no-meta`);
     }
-    case "editImage": {
-      const width = input.width || 2048;
-      const height = input.height || 2048;
-      const step = JSON.stringify({
-        prompt: input.prompt,
-        imageModelOverride: {
-          model: "seedream-4.5",
-          config: {
-            images: [input.imageUrl],
-            width,
-            height
-          }
-        }
-      });
-      return runCli(
-        `mindstudio generate-image '${step}' --output-key imageUrl --no-meta`
-      );
-    }
     default:
       return `Error: unknown tool "${name}"`;
   }
@@ -2519,20 +2502,6 @@ Be specific and concise.`;
           required: ["url"]
         }
       },
-      {
-        name: "searchStockPhotos",
-        description: 'Search Pexels for stock photos. Returns image URLs with descriptions. Use concrete, descriptive queries ("person working at laptop in modern office" not "business").',
-        inputSchema: {
-          type: "object",
-          properties: {
-            query: {
-              type: "string",
-              description: "What kind of photo to search for."
-            }
-          },
-          required: ["query"]
-        }
-      },
       {
         name: "searchProductScreenshots",
         description: 'Search for screenshots of real products and apps. Use to find what existing products look like ("stripe dashboard", "linear app", "notion workspace"). Returns image results of actual product UI. Use this for layout and design research on real products, NOT for abstract design inspiration.',
@@ -2571,32 +2540,6 @@ Be specific and concise.`;
           },
           required: ["prompts"]
         }
-      },
-      {
-        name: "editImage",
-        description: "Edit an existing image using a text instruction. Takes a source image URL and a prompt describing the edits (color grading, style transfer, modifications, adding/removing elements). Returns a new CDN URL.",
-        inputSchema: {
-          type: "object",
-          properties: {
-            imageUrl: {
-              type: "string",
-              description: "URL of the source image to edit."
-            },
-            prompt: {
-              type: "string",
-              description: 'What to change. Describe the edit as an instruction: "apply warm golden hour color grading", "make the background darker", "add a subtle film grain texture".'
-            },
-            width: {
-              type: "number",
-              description: "Output width in pixels. Default 2048. Range: 2048-4096."
-            },
-            height: {
-              type: "number",
-              description: "Output height in pixels. Default 2048. Range: 2048-4096."
-            }
-          },
-          required: ["imageUrl", "prompt"]
-        }
       }
     ];
   }
@@ -2606,8 +2549,8 @@ Be specific and concise.`;
 import fs11 from "fs";
 import path5 from "path";
 function resolvePath(filename) {
-  const local2 = path5.join(base2, filename);
-  return fs11.existsSync(local2) ? local2 : path5.join(base2, "subagents", "designExpert", filename);
+  const local3 = path5.join(base2, filename);
+  return fs11.existsSync(local3) ? local3 : path5.join(base2, "subagents", "designExpert", filename);
 }
 function readFile(filename) {
   return fs11.readFileSync(resolvePath(filename), "utf-8").trim();
@@ -2762,6 +2705,159 @@ Concrete resources: hex values, font names with CSS URLs, image URLs, layout des
   }
 });
 
+// src/subagents/productVision/index.ts
+import fs12 from "fs";
+import path6 from "path";
+function loadSpecContext() {
+  const specDir = "src";
+  const files = [];
+  function walk(dir) {
+    try {
+      for (const entry of fs12.readdirSync(dir, { withFileTypes: true })) {
+        const full = path6.join(dir, entry.name);
+        if (entry.isDirectory()) {
+          if (entry.name !== "roadmap") {
+            walk(full);
+          }
+        } else if (entry.name.endsWith(".md")) {
+          files.push(full);
+        }
+      }
+    } catch {
+    }
+  }
+  walk(specDir);
+  if (files.length === 0) {
+    return "";
+  }
+  const sections = files.map((f) => {
+    try {
+      const content = fs12.readFileSync(f, "utf-8").trim();
+      return `<file path="${f}">
+${content}
+</file>`;
+    } catch {
+      return "";
+    }
+  }).filter(Boolean);
+  return `<spec_files>
+${sections.join("\n\n")}
+</spec_files>`;
+}
+async function executeVisionTool(name, input) {
+  if (name !== "writeRoadmapItem") {
+    return `Error: unknown tool "${name}"`;
+  }
+  const { slug, name: itemName, description, effort, requires, body } = input;
+  const dir = "src/roadmap";
+  const filePath = path6.join(dir, `${slug}.md`);
+  try {
+    fs12.mkdirSync(dir, { recursive: true });
+    const requiresYaml = requires.length === 0 ? "[]" : `[${requires.map((r) => `"${r}"`).join(", ")}]`;
+    const content = `---
+name: ${itemName}
+type: roadmap
+status: ${slug === "mvp" ? "in-progress" : "not-started"}
+description: ${description}
+effort: ${effort}
+requires: ${requiresYaml}
+---
+
+${body}
+`;
+    fs12.writeFileSync(filePath, content, "utf-8");
+    return `Wrote ${filePath}`;
+  } catch (err) {
+    return `Error writing ${filePath}: ${err.message}`;
+  }
+}
+var base3, local2, PROMPT_PATH2, BASE_PROMPT, VISION_TOOLS, productVisionTool;
+var init_productVision = __esm({
+  "src/subagents/productVision/index.ts"() {
+    "use strict";
+    init_runner();
+    base3 = import.meta.dirname ?? path6.dirname(new URL(import.meta.url).pathname);
+    local2 = path6.join(base3, "prompt.md");
+    PROMPT_PATH2 = fs12.existsSync(local2) ? local2 : path6.join(base3, "subagents", "productVision", "prompt.md");
+    BASE_PROMPT = fs12.readFileSync(PROMPT_PATH2, "utf-8").trim();
+    VISION_TOOLS = [
+      {
+        name: "writeRoadmapItem",
+        description: "Write a roadmap item to src/roadmap/. Call this once for each idea.",
+        inputSchema: {
+          type: "object",
+          properties: {
+            slug: {
+              type: "string",
+              description: 'Kebab-case filename (without .md). e.g. "ai-weekly-digest"'
+            },
+            name: {
+              type: "string",
+              description: "User-facing feature name."
+            },
+            description: {
+              type: "string",
+              description: "Short user-facing summary (1-2 sentences)."
+            },
+            effort: {
+              type: "string",
+              enum: ["quick", "small", "medium", "large"]
+            },
+            requires: {
+              type: "array",
+              items: { type: "string" },
+              description: "Slugs of prerequisite roadmap items. Empty array if independent."
+            },
+            body: {
+              type: "string",
+              description: "Full MSFM body: prose description for the user, followed by ~~~annotation~~~ with technical implementation notes for the building agent."
+            }
+          },
+          required: ["slug", "name", "description", "effort", "requires", "body"]
+        }
+      }
+    ];
+    productVisionTool = {
+      definition: {
+        name: "productVision",
+        description: `A product visionary that imagines where the project could go next. It automatically reads all spec files from src/ for context. Pass a brief description of the app and who it's for. It generates 10-15 ambitious, creative roadmap ideas and writes them directly to src/roadmap/. Use this at the end of spec authoring to populate the roadmap.`,
+        inputSchema: {
+          type: "object",
+          properties: {
+            task: {
+              type: "string",
+              description: "Brief description of the app and who it's for. The tool reads the full spec files automatically \u2014 no need to repeat their contents."
+            }
+          },
+          required: ["task"]
+        }
+      },
+      async execute(input, context) {
+        if (!context) {
+          return "Error: product vision requires execution context";
+        }
+        const specContext = loadSpecContext();
+        const system = specContext ? `${BASE_PROMPT}
+
+${specContext}` : BASE_PROMPT;
+        return runSubAgent({
+          system,
+          task: input.task,
+          tools: VISION_TOOLS,
+          externalTools: /* @__PURE__ */ new Set(),
+          executeTool: executeVisionTool,
+          apiConfig: context.apiConfig,
+          model: context.model,
+          signal: context.signal,
+          parentToolId: context.toolCallId,
+          onEvent: context.onEvent,
+          resolveExternalTool: context.resolveExternalTool
+        });
+      }
+    };
+  }
+});
+
 // src/tools/index.ts
 function getSpecTools() {
   return [readSpecTool, writeSpecTool, editSpecTool, listSpecFilesTool];
@@ -2795,7 +2891,8 @@ function getCommonTools() {
     fetchUrlTool,
     searchGoogleTool,
     setProjectNameTool,
-    designExpertTool
+    designExpertTool,
+    productVisionTool
   ];
 }
 function getPostOnboardingTools() {
@@ -2874,14 +2971,15 @@ var init_tools3 = __esm({
     init_screenshot();
     init_browserAutomation();
     init_designExpert();
+    init_productVision();
   }
 });
 
 // src/session.ts
-import fs12 from "fs";
+import fs13 from "fs";
 function loadSession(state) {
   try {
-    const raw = fs12.readFileSync(SESSION_FILE, "utf-8");
+    const raw = fs13.readFileSync(SESSION_FILE, "utf-8");
     const data = JSON.parse(raw);
     if (Array.isArray(data.messages) && data.messages.length > 0) {
       state.messages = sanitizeMessages(data.messages);
@@ -2923,7 +3021,7 @@ function sanitizeMessages(messages) {
 }
 function saveSession(state) {
   try {
-    fs12.writeFileSync(
+    fs13.writeFileSync(
       SESSION_FILE,
       JSON.stringify({ messages: state.messages }, null, 2),
       "utf-8"
@@ -2934,7 +3032,7 @@ function saveSession(state) {
 function clearSession(state) {
   state.messages = [];
   try {
-    fs12.unlinkSync(SESSION_FILE);
+    fs13.unlinkSync(SESSION_FILE);
   } catch {
   }
 }
@@ -3593,12 +3691,12 @@ var init_agent = __esm({
 });
 
 // src/prompt/static/projectContext.ts
-import fs13 from "fs";
-import path6 from "path";
+import fs14 from "fs";
+import path7 from "path";
 function loadProjectInstructions() {
   for (const file of AGENT_INSTRUCTION_FILES) {
     try {
-      const content = fs13.readFileSync(file, "utf-8").trim();
+      const content = fs14.readFileSync(file, "utf-8").trim();
       if (content) {
         return `
 ## Project Instructions (${file})
@@ -3611,7 +3709,7 @@ ${content}`;
 }
 function loadProjectManifest() {
   try {
-    const manifest = fs13.readFileSync("mindstudio.json", "utf-8");
+    const manifest = fs14.readFileSync("mindstudio.json", "utf-8");
     return `
 ## Project Manifest (mindstudio.json)
 \`\`\`json
@@ -3652,9 +3750,9 @@ ${entries.join("\n")}`;
 function walkMdFiles(dir) {
   const results = [];
   try {
-    const entries = fs13.readdirSync(dir, { withFileTypes: true });
+    const entries = fs14.readdirSync(dir, { withFileTypes: true });
     for (const entry of entries) {
-      const full = path6.join(dir, entry.name);
+      const full = path7.join(dir, entry.name);
       if (entry.isDirectory()) {
         results.push(...walkMdFiles(full));
       } else if (entry.name.endsWith(".md")) {
@@ -3667,7 +3765,7 @@ function walkMdFiles(dir) {
 }
 function parseFrontmatter(filePath) {
   try {
-    const content = fs13.readFileSync(filePath, "utf-8");
+    const content = fs14.readFileSync(filePath, "utf-8");
     const match = content.match(/^---\n([\s\S]*?)\n---/);
     if (!match) {
       return { name: "", description: "", type: "" };
@@ -3683,7 +3781,7 @@ function parseFrontmatter(filePath) {
 }
 function loadProjectFileListing() {
   try {
-    const entries = fs13.readdirSync(".", { withFileTypes: true });
+    const entries = fs14.readdirSync(".", { withFileTypes: true });
     const listing = entries.filter((e) => e.name !== ".git" && e.name !== "node_modules").sort((a, b) => {
       if (a.isDirectory() && !b.isDirectory()) {
         return -1;
@@ -3726,12 +3824,12 @@ var init_projectContext = __esm({
 });
 
 // src/prompt/index.ts
-import fs14 from "fs";
-import path7 from "path";
+import fs15 from "fs";
+import path8 from "path";
 function requireFile(filePath) {
-  const full = path7.join(PROMPT_DIR, filePath);
+  const full = path8.join(PROMPT_DIR, filePath);
   try {
-    return fs14.readFileSync(full, "utf-8").trim();
+    return fs15.readFileSync(full, "utf-8").trim();
   } catch {
     throw new Error(`Required prompt file missing: ${full}`);
   }
@@ -3856,17 +3954,17 @@ var init_prompt3 = __esm({
     "use strict";
     init_lsp();
     init_projectContext();
-    PROMPT_DIR = import.meta.dirname ?? path7.dirname(new URL(import.meta.url).pathname);
+    PROMPT_DIR = import.meta.dirname ?? path8.dirname(new URL(import.meta.url).pathname);
   }
 });
 
 // src/config.ts
-import fs15 from "fs";
-import path8 from "path";
+import fs16 from "fs";
+import path9 from "path";
 import os from "os";
 function loadConfigFile() {
   try {
-    const raw = fs15.readFileSync(CONFIG_PATH, "utf-8");
+    const raw = fs16.readFileSync(CONFIG_PATH, "utf-8");
     log.debug("Loaded config file", { path: CONFIG_PATH });
     return JSON.parse(raw);
   } catch (err) {
@@ -3902,7 +4000,7 @@ var init_config = __esm({
   "src/config.ts"() {
     "use strict";
     init_logger();
-    CONFIG_PATH = path8.join(
+    CONFIG_PATH = path9.join(
       os.homedir(),
       ".mindstudio-local-tunnel",
       "config.json"
@@ -3917,10 +4015,10 @@ __export(headless_exports, {
   startHeadless: () => startHeadless
 });
 import { createInterface } from "readline";
-import fs16 from "fs";
-import path9 from "path";
+import fs17 from "fs";
+import path10 from "path";
 function loadActionPrompt(name) {
-  return fs16.readFileSync(path9.join(ACTIONS_DIR, `${name}.md`), "utf-8").trim();
+  return fs17.readFileSync(path10.join(ACTIONS_DIR, `${name}.md`), "utf-8").trim();
 }
 function emit(event, data) {
   process.stdout.write(JSON.stringify({ event, ...data }) + "\n");
@@ -4139,16 +4237,16 @@ var init_headless = __esm({
     init_lsp();
     init_agent();
     init_session();
-    BASE_DIR = import.meta.dirname ?? path9.dirname(new URL(import.meta.url).pathname);
-    ACTIONS_DIR = path9.join(BASE_DIR, "actions");
+    BASE_DIR = import.meta.dirname ?? path10.dirname(new URL(import.meta.url).pathname);
+    ACTIONS_DIR = path10.join(BASE_DIR, "actions");
   }
 });
 
 // src/index.tsx
 import { render } from "ink";
 import os2 from "os";
-import fs17 from "fs";
-import path10 from "path";
+import fs18 from "fs";
+import path11 from "path";
 
 // src/tui/App.tsx
 import { useState as useState2, useCallback, useRef } from "react";
@@ -4465,8 +4563,8 @@ for (let i = 0; i < args.length; i++) {
 }
 function printDebugInfo(config) {
   const pkg = JSON.parse(
-    fs17.readFileSync(
-      path10.join(import.meta.dirname, "..", "package.json"),
+    fs18.readFileSync(
+      path11.join(import.meta.dirname, "..", "package.json"),
       "utf-8"
     )
   );

package/dist/prompt/compiled/msfm.md

@@ -112,7 +112,10 @@ A spec starts with YAML frontmatter followed by freeform Markdown. There's no ma
 **Frontmatter fields:**
 - `name` (required) — display name for the spec file
 - `description` (optional) — short summary of what this file covers
-- `type` (optional) — defaults to `spec`. Other values: `design/color` (color palette definition), `design/typography` (font and type style definition). The frontend renders these types with specialized editors.
+- `type` (optional) — defaults to `spec`. Other values: `design/color` (color palette definition), `design/typography` (font and type style definition), `roadmap` (feature roadmap item). The frontend renders these types with specialized editors.
+- `status` (roadmap only) — `done`, `in-progress`, or `not-started`
+- `requires` (roadmap only) — array of slugs for prerequisite roadmap items. Empty array means available now.
+- `effort` (roadmap only) — `quick`, `small`, `medium`, or `large`
 
 ```markdown
 ---
@@ -187,3 +190,32 @@ styles:
     description: Default reading text
 ```
 ```
+
+Roadmap item example (one file per feature in `src/roadmap/`):
+
+```markdown
+---
+name: Share & Export
+type: roadmap
+status: not-started
+description: Share haikus as image cards to social media or download as prints.
+requires: []
+effort: medium
+---
+
+Share haikus as styled image cards on social media or download as prints.
+The card system generates images using the brand's typography and color
+palette, creating shareable assets that feel native to the app's identity.
+
+~~~
+Use generateImage with Seedream to create styled cards. Card template
+applies brand typography and colors from the spec. Export as PNG via
+CDN transform at 2x resolution. Social sharing via Web Share API with
+clipboard fallback for unsupported browsers.
+~~~
+
+## History
+
+- **2026-03-22** — Built card generation using generateImage with Seedream.
+  Added share button to haiku detail view.
+```

package/dist/prompt/compiled/platform.md

@@ -23,6 +23,7 @@ my-app/
       web.md                           web UI spec
       api.md                           API conventions
       cron.md                          scheduled job descriptions
+    roadmap/                           feature roadmap (one file per item, type: roadmap)
 
   dist/                              ← compiled output (code + config)
     methods/                           backend contract
@@ -60,6 +61,7 @@ my-app/
 | Interface configs | `dist/interfaces/*/interface.json` | One per non-web interface type |
 | Specs | `src/*.md` | Natural language, MSFM format |
 | Brand identity | `src/interfaces/@brand/` | visual.md (aesthetic), colors.md (palette), typography.md (fonts), voice.md (tone), assets/ |
+| Roadmap | `src/roadmap/*.md` | Feature roadmap items (type: roadmap). One file per feature with status, dependencies, and history. |
 | Reference material | `src/references/` | Context for the agent, not consumed by platform |
 
 ## The Two SDKs

package/dist/prompt/compiled/sdk-actions.md

@@ -2,7 +2,9 @@
 
 `@mindstudio-ai/agent` provides access to 200+ AI models and 1,000+ actions through a single API key. No separate provider keys needed. MindStudio routes to the correct provider (OpenAI, Anthropic, Google, etc.) server-side.
 
-There is a huge amount of capability here: hundreds of text generation models (OpenAI, Anthropic, Google, Meta, Mistral, and more), dozens of image generation models (FLUX, DALL-E, Stable Diffusion, Ideogram, and more), video generation, text-to-speech, music generation, vision analysis, web scraping, 850+ OAuth connectors, and much more. The tables below are a summary. **Always use the `askMindStudioSdk` tool to look up exact method signatures, model IDs, and config options before writing code that uses the SDK.** The SDK assistant knows every action, every model, every connector, and the user's configured OAuth connections. Don't guess at parameters or model IDs from memory.
+There is a huge amount of capability here: hundreds of text generation models (OpenAI, Anthropic, Google, Meta, Mistral, and more), dozens of image generation models (FLUX, DALL-E, Stable Diffusion, Ideogram, and more), video generation, text-to-speech, music generation, vision analysis, web scraping, 850+ OAuth connectors, and much more. The tables below are a summary.
+
+**Always use `askMindStudioSdk` before writing code that uses the SDK.** Treat it as an expert consultant, not a docs search. Describe what you're trying to build at the method level — the full workflow, not just "how do I call generateText." The assistant knows every action, model, connector, configuration option, and the user's configured OAuth connections. It can advise on AI orchestration patterns (structured output, chaining calls, batch processing), help you avoid common mistakes (like manually parsing JSON when the SDK has structured output options), and provide complete working code for your use case.
 
 ## Usage in Methods
 

package/dist/prompt/static/authoring.md

@@ -17,8 +17,9 @@ The scaffold starts with these spec files that cover the full picture of the app
 - **`src/interfaces/@brand/colors.md`** (`type: design/color`) — brand color palette: 3-5 named colors with evocative names and brand-level descriptions. The design system is derived from these.
 - **`src/interfaces/@brand/typography.md`** (`type: design/typography`) — font choices with source URLs and 1-2 anchor styles (Display, Body). Additional styles are derived from these anchors.
 - **`src/interfaces/@brand/voice.md`** — voice and terminology: tone, error messages, word choices
+- **`src/roadmap/`** — feature roadmap. One file per feature (`type: roadmap`). See "Roadmap" below.
 
-Start from these four and extend as needed. Add interface specs for other interface types (`api.md`, `cron.md`, etc.) if the app uses them. Split `app.md` into multiple files if the domain is complex. The agent uses the entire `src/` folder as compilation context, so organize however serves clarity.
+Start from these and extend as needed. Add interface specs for other interface types (`api.md`, `cron.md`, etc.) if the app uses them. Split `app.md` into multiple files if the domain is complex. The agent uses the entire `src/` folder as compilation context, so organize however serves clarity.
 
 Users often care about look and feel as much as (or more than) underlying data structures. Don't treat the brand and interface specs as an afterthought — for many users, the visual identity and voice are the first things they want to get right.
 
@@ -56,6 +57,29 @@ When the user clicks "Build," you will receive a build command. Build everything
 
 Scenarios are cheap to write (same `db.push()` calls as methods) but critical for testing. An app without scenarios is not done.
 
+## Roadmap
+
+The initial build should deliver everything the user asked for. The roadmap is not a place to defer work the user requested. It's for future additions: natural extensions of the app, features the user didn't think to ask for, and ideas that would make the app even better. Think of it as "here's what you have, and here's where you could take it next."
+
+Roadmap items live in `src/roadmap/`, one MSFM file per feature with structured frontmatter:
+
+- `name` — the feature name
+- `type: roadmap`
+- `status` — `done`, `in-progress`, or `not-started`
+- `description` — short summary (used for index rendering)
+- `requires` — array of slugs for prerequisite items. Empty array means available now.
+- `effort` — `quick`, `small`, `medium`, or `large`
+
+Each roadmap item should be a meaningful chunk of work that results in a noticeably different version of the product. Not individual tasks. Bundle polish and small improvements into single items. The big items should be product pillars — think beyond the current deliverable toward the actual product the user is building. If the user asked for a landing page, the roadmap should include building the actual product the landing page is selling.
+
+Write names and descriptions for the user, not for developers. Focus on what the user gets, not how it's built. No technical jargon, no library names, no implementation details.
+
+The body is freeform MSFM: prose describing the feature for the user, annotations with technical approach and architecture notes for the agent. Append a History section as items are built.
+
+The MVP itself gets a roadmap file (`src/roadmap/mvp.md`) with `status: in-progress` that documents what the initial build covers. Update it to `done` after the build completes. Other items start as `not-started`. Some items depend on others (`requires: [share-export]`), some are independent (`requires: []`). The user picks what to build next.
+
+Write the roadmap as the final step of spec authoring, after all other spec files are written. Use the `productVision` tool to generate roadmap ideas — pass it the full context of what was built (the app domain, what it does, who it's for, the design direction) and it returns ambitious, creative ideas. Write each returned idea into its own roadmap file in `src/roadmap/`.
+
 ## Spec + Code Sync
 
 When generated code exists in `dist/`, you have both spec tools and code tools.

package/dist/static/authoring.md

@@ -17,8 +17,9 @@ The scaffold starts with these spec files that cover the full picture of the app
 - **`src/interfaces/@brand/colors.md`** (`type: design/color`) — brand color palette: 3-5 named colors with evocative names and brand-level descriptions. The design system is derived from these.
 - **`src/interfaces/@brand/typography.md`** (`type: design/typography`) — font choices with source URLs and 1-2 anchor styles (Display, Body). Additional styles are derived from these anchors.
 - **`src/interfaces/@brand/voice.md`** — voice and terminology: tone, error messages, word choices
+- **`src/roadmap/`** — feature roadmap. One file per feature (`type: roadmap`). See "Roadmap" below.
 
-Start from these four and extend as needed. Add interface specs for other interface types (`api.md`, `cron.md`, etc.) if the app uses them. Split `app.md` into multiple files if the domain is complex. The agent uses the entire `src/` folder as compilation context, so organize however serves clarity.
+Start from these and extend as needed. Add interface specs for other interface types (`api.md`, `cron.md`, etc.) if the app uses them. Split `app.md` into multiple files if the domain is complex. The agent uses the entire `src/` folder as compilation context, so organize however serves clarity.
 
 Users often care about look and feel as much as (or more than) underlying data structures. Don't treat the brand and interface specs as an afterthought — for many users, the visual identity and voice are the first things they want to get right.
 
@@ -56,6 +57,29 @@ When the user clicks "Build," you will receive a build command. Build everything
 
 Scenarios are cheap to write (same `db.push()` calls as methods) but critical for testing. An app without scenarios is not done.
 
+## Roadmap
+
+The initial build should deliver everything the user asked for. The roadmap is not a place to defer work the user requested. It's for future additions: natural extensions of the app, features the user didn't think to ask for, and ideas that would make the app even better. Think of it as "here's what you have, and here's where you could take it next."
+
+Roadmap items live in `src/roadmap/`, one MSFM file per feature with structured frontmatter:
+
+- `name` — the feature name
+- `type: roadmap`
+- `status` — `done`, `in-progress`, or `not-started`
+- `description` — short summary (used for index rendering)
+- `requires` — array of slugs for prerequisite items. Empty array means available now.
+- `effort` — `quick`, `small`, `medium`, or `large`
+
+Each roadmap item should be a meaningful chunk of work that results in a noticeably different version of the product. Not individual tasks. Bundle polish and small improvements into single items. The big items should be product pillars — think beyond the current deliverable toward the actual product the user is building. If the user asked for a landing page, the roadmap should include building the actual product the landing page is selling.
+
+Write names and descriptions for the user, not for developers. Focus on what the user gets, not how it's built. No technical jargon, no library names, no implementation details.
+
+The body is freeform MSFM: prose describing the feature for the user, annotations with technical approach and architecture notes for the agent. Append a History section as items are built.
+
+The MVP itself gets a roadmap file (`src/roadmap/mvp.md`) with `status: in-progress` that documents what the initial build covers. Update it to `done` after the build completes. Other items start as `not-started`. Some items depend on others (`requires: [share-export]`), some are independent (`requires: []`). The user picks what to build next.
+
+Write the roadmap as the final step of spec authoring, after all other spec files are written. Use the `productVision` tool to generate roadmap ideas — pass it the full context of what was built (the app domain, what it does, who it's for, the design direction) and it returns ambitious, creative ideas. Write each returned idea into its own roadmap file in `src/roadmap/`.
+
 ## Spec + Code Sync
 
 When generated code exists in `dist/`, you have both spec tools and code tools.

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

@@ -5,6 +5,7 @@
 - Spring physics for natural-feeling motion
 - Purposeful micro-interactions — scaling, color shifts, depth changes on hover/click
 - Staggered entrance reveals — content appearing sequentially as it enters view
+- Pay attention to timing, duration, speed, and layout shift - make sure animations are beautiful, especially if they involve text or elements the user is reading or interacting with.
 
 ### Libraries
   - Prefer raw CSS animations when possible. 

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

@@ -6,47 +6,26 @@ Not every interface needs images. A productivity dashboard, a finance tool, or a
 
 Do not provide images as "references" - images must be ready-to-use assets that can be included directly in the design.
 
-### Three tools
+### Image generation
 
-**AI-generated photos and images** (`generateImages`) — Seedream produces high-quality results for both photorealistic images and abstract/creative visuals. You have full control over the output: style, composition, colors, mood. When generating multiple images, batch them in a single `generateImages` call — they run in parallel. Generated images are production assets, not mockups or concepts — they are hosted on MindStudio CDN at full resolution and will be used directly in the final interface.
-
-**Image editing** (`editImage`) — takes an existing image URL and a text instruction describing what to change. Use this to adjust stock photos to match the brand: color grading, style transfer, cropping mood, adding atmosphere. Find a great stock photo, then edit it to align with the design direction.
-
-**Stock photography** (`searchStockPhotos`) — Pexels has modern, editorial-style photos. Good starting points that can be used directly or refined with `editImage`. Write specific queries: "person writing in notebook at minimalist desk, natural light" not "office."
+Use `generateImages` to create images. Seedream produces high-quality results for both photorealistic images and abstract/creative visuals. You have full control over the output: style, composition, colors, mood. When generating multiple images, batch them in a single `generateImages` call — they run in parallel. Generated images are production assets, not mockups or concepts — they are hosted on MindStudio CDN at full resolution and will be used directly in the final interface.
 
 ### Writing good generation prompts
 
-Write prompts as natural sentences describing a scene, not as comma-separated keyword lists. Describe what a camera would see, not art direction instructions.
-
-**Structure:** Subject and action first, then setting, then style and technical details. Include the intended use when relevant.
-
-- "A woman laughing while reading on a sun-drenched balcony overlooking a Mediterranean harbor. Editorial photography, shot on Kodak Portra 400, 85mm lens at f/2, soft golden hour light, shallow depth of field. For a lifestyle app hero section."
-- "An overhead view of a cluttered designer's desk with fabric swatches, sketches, and a coffee cup. Natural window light from the left, slightly desaturated tones, Canon 5D with 35mm lens. For an about page."
-- "Smooth organic shapes in deep navy and warm amber, flowing liquid forms with subtle grain texture. Abstract digital art, high contrast, editorial feel."
-
-**Photography vocabulary produces the best results.** The model responds strongly to specific references:
-- Film stocks: Kodak Portra, Fuji Superia, Cinestill 800T, expired film
-- Lenses: 85mm f/1.4, 35mm wide angle, 50mm Summilux, macro
-- Lighting: golden hour, chiaroscuro, tungsten warmth, soft diffused studio light, direct flash
-- Shot types: close-up, overhead flat lay, low angle, eye-level candid, aerial
-- Techniques: shallow depth of field, halation around highlights, film grain, motion blur
-
-**Declare the medium early.** Saying "editorial photograph" vs "watercolor painting" vs "3D render" doesn't just change style — it changes the model's entire approach to composition, color, and detail. Set this expectation in the first sentence.
+Lead with the visual style, then describe the content. This order helps the model establish the look before filling in details.
 
-**For text in images**, wrap the exact text in double quotes and specify the style: `A neon sign reading "OPEN" in cursive pink lettering against a dark brick wall.`
+**Structure:** Style/medium first, then subject, then details.
+- "Digital photography, soft natural window light, shallow depth of field. A ceramic coffee cup on a marble countertop, morning light casting long shadows, warm tones."
+- "Flat vector illustration, clean lines, limited color palette. An isometric view of a workspace with a laptop, plant, and notebook."
+- "Abstract digital art, fluid gradients, high contrast. Deep navy flowing into warm amber, organic liquid shapes, editorial feel."
 
-**Compose for the layout.** If you know the image will have text overlaid, request space for it: "negative space in the upper left for headline text" or "clean sky area above the subject." If it's a background, consider "centered subject with clean margins." The first few words of the prompt carry the most weight — lead with the medium and subject.
+**For photorealistic images:** Specify the photography style (editorial, portrait, product, aerial), lighting (natural, studio, golden hour, direct flash), and camera characteristics (close-up, wide angle, shallow depth of field, slightly grainy texture).
 
 **Avoid:**
 - Hex codes in prompts — the model renders them as visible text. Describe colors by name instead.
-- Keyword lists separated by commas — write sentences.
 - Describing positions of arms, legs, or specific limb arrangements.
 - Conflicting style instructions ("photorealistic cartoon").
 - Describing what you don't want — say "empty street" not "street with no cars."
-- Mentioning "text" or "text placement" in prompts — the model will try to render text. Request the composition you want ("negative space in the left third") without saying why.
-- Brand names (camera brands, font names, company names) can get rendered as visible text. Use technical specs ("medium format, 120mm lens") instead of brand names ("Hasselblad") when possible.
-- UI component language — "glass morphism effect", "card design", "button with hover state". Write prompts as if briefing a photographer or artist, not describing CSS.
-- Generating text that should be HTML. Headlines, body copy, CTAs, and any text the user needs to read or interact with belongs in the markup, not baked into an image. Text *within a scene* is fine — a neon sign, a logo on a t-shirt, text on a billboard in a cityscape, an app screen in a device mockup. That's part of the visual content.
 
 ### How generated images work in the UI
 

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

@@ -2,8 +2,7 @@
 
 - Use `screenshotAndAnalyze` only when you need to see the visual design of a site (layout, colors, typography in context). Do not screenshot font specimen pages, documentation, search results, or other text-heavy pages — use `fetchUrl` for those instead. Screenshots are expensive and slow; only use them when visual appearance matters.
 - Use `analyzeDesignReference` for consistent design analysis of images or screenshots. Use `analyzeImage` when you have a specific question about an image.
-- Use `searchStockPhotos` for stock imagery. Describe what you need concretely ("person working at laptop in modern office" not "business").
-- Use `searchProductScreenshots` to find screenshots of real products ("stripe dashboard", "linear app"). Use this for layout research on what real products look like. Do not use this for abstract design inspiration.
+- Use `searchProductScreenshots` to find screenshots of real products ("stripe dashboard", "linear app"). Use this for layout research on what real products look like.
 - Use `searchGoogle` for research: font pairing recommendations, "best [domain] apps 2026", design trend articles. Prioritize authoritative sources like Figma and other design leaders, avoid random blog spam.
 - Use `fetchUrl` when you need to get the text content of a site.
 - When proposing multiple options, make them genuinely different directions (dark + bold vs. light + editorial) rather than minor variations.

package/dist/subagents/productVision/prompt.md

@@ -0,0 +1,73 @@
+The role of the assistant is to act as a product visionary — the kind of person who sees a simple prototype and immediately envisions the billion-dollar company it could become. The assistant thinks like a founder pitching the next 12 months to investors who are already excited about what they see.
+
+The assistant is not a developer. It does not think in terms of implementations, libraries, or technical architecture. It thinks about what users would love, what would make them tell their friends, what would make the product indispensable. It thinks about what would make someone say "I can't believe this exists."
+
+The assistant's job is to stretch the user's imagination far beyond what they asked for. The user's stated scope is a starting point, not a ceiling. If they described a simple tool, the assistant imagines it as a platform. If they asked for one feature, the assistant sees the whole product it could be part of. The user came here because they want to be inspired — that is the actual request, even if they didn't say it. Even a wild idea that gets rejected is valuable if it sparks new thinking. The assistant makes the user's ambitions bigger, not smaller.
+
+## How to think
+
+The assistant has just been shown what version 1 looks like. It now imagines version 5. What does this product look like when it's fully realized? When it has a loyal user base? When it's the best in its category?
+
+The assistant thinks in lanes, not lists. A great product roadmap has 3-5 distinct directions the product could grow, each with depth. Like a skill tree in a game: each lane starts with a foundational feature that unlocks progressively more powerful capabilities.
+
+One lane might deepen the core experience. Another might add a social layer. Another might introduce AI capabilities that feel like magic. Another might expand beyond the web into new surfaces. Each lane has a natural progression — you can't have the advanced version without the foundation, and each step along the way results in a product that feels complete.
+
+The assistant uses the `requires` field to express these progressions. Items within a lane depend on earlier items in that lane. Items across lanes are independent. The user can choose which lane to invest in next.
+
+The assistant thinks across dimensions like:
+- The core experience: how could it be deeper, smarter, more personalized?
+- Social and community: how could users connect with each other through this?
+- AI capabilities: what could the product do automatically that feels like magic?
+- New surfaces: could this live beyond the web?
+- Insights and analytics: what could the product reveal about patterns and data?
+- Growth: what creates viral moments? What makes users invite others?
+
+Not every dimension applies to every product. But the assistant pushes itself to build real depth in at least 3 lanes rather than scattering shallow ideas across many.
+
+## Self-check
+
+Before submitting, the assistant asks itself: would a user be excited showing this roadmap to a friend? Would it make them say "holy shit, I could actually build all of this?" If not, the assistant pushes further. At least 3 items must be large effort. At least 2 lanes must extend beyond the current product scope into genuinely new territory.
+
+## What to produce
+
+First, the assistant writes an MVP item capturing what's being built right now (slug "mvp", status will be set to in-progress automatically). Then it generates 10-15 future roadmap ideas. It uses the `writeRoadmapItem` tool to write each one directly. It calls the tool once per idea — batching all calls in a single turn for efficiency.
+
+For each idea:
+- **name** — short, exciting, user-facing. No technical jargon. Something you'd see on a product launch page.
+- **description** — 1-2 sentences explaining what the user gets. Written for the user, not a developer.
+- **effort** — `quick`, `small`, `medium`, or `large`
+- **requires** — slugs of prerequisite items. Empty array if independent.
+- **body** — a structured MSFM document, not a narrative essay. Format it as:
+
+```
+[1-2 sentence elevator pitch — what is this and why does it matter]
+
+## What it looks like
+
+[Concrete description of the user experience. What do they see, what do they do, how does it feel. Use headers and bullet points to organize, not long paragraphs.]
+
+## Key details
+
+[Specific behaviors, rules, edge cases that matter for this feature.]
+
+~~~
+[Technical implementation notes for the building agent. Architecture, data model, AI prompts, integrations needed.]
+~~~
+```
+
+Keep it concise and scannable. Use markdown structure (headers, bullets, short paragraphs). The body should read like a mini spec, not a sales pitch.
+
+## Rules
+
+- Write names and descriptions for humans who have never written a line of code.
+- Be specific and concrete. "AI-Powered Weekly Digest" not "Email features."
+- Include a mix: a few quick wins for momentum, several medium features that expand the product, and a few ambitious large items that represent the full vision.
+- At least 2-3 items should make the user think "I didn't know that was even possible."
+- The ideas should form lanes with depth, not be a flat list of unrelated features. Use `requires` to build progressions.
+- Go far beyond what was asked for. The user described where they are. The assistant describes where they could be.
+- Be bold. The user can always say no. A safe, boring roadmap is worse than no roadmap at all.
+- Cap it at 15 items (plus the MVP). Quality and depth over quantity.
+
+<voice>
+No emoji. No hedging ("you could maybe consider..."). The assistant is confident and direct. It is pitching a vision, not suggesting options.
+</voice>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mindstudio-ai/remy",
-  "version": "0.1.15",
+  "version": "0.1.17",
   "description": "MindStudio coding agent",
   "repository": {
     "type": "git",