fix(web): hide image-only bootstrap prompt

Author: eimexdev

apps/web/src/components/ChatView.tsx

@@ -183,6 +183,7 @@ import {
 import {
   appendTerminalContextsToPrompt,
   formatTerminalContextLabel,
+  IMAGE_ONLY_BOOTSTRAP_PROMPT,
   type TerminalContextDraft,
   type TerminalContextSelection,
 } from "../lib/terminalContext";
@@ -302,8 +303,6 @@ import {
 } from "../versionSkew";
 import { useAssetUrls } from "../assets/assetUrls";
 
-const IMAGE_ONLY_BOOTSTRAP_PROMPT =
-  "[User attached one or more images without additional text. Respond using the conversation context and the attached image(s).]";
 const EMPTY_ACTIVITIES: OrchestrationThreadActivity[] = [];
 const EMPTY_PROVIDERS: ServerProvider[] = [];
 const EMPTY_PROVIDER_SKILLS: ServerProvider["skills"] = [];

apps/web/src/components/chat/MessagesTimeline.tsx

@@ -930,7 +930,9 @@ function UserTimelineRow({ row }: { row: Extract<TimelineRow, { kind: "message"
   const userAudio = userAttachments.filter(
     (attachment): attachment is ChatAudioAttachment => attachment.type === "audio",
   );
-  const displayedUserMessage = deriveDisplayedUserMessageState(row.message.text);
+  const displayedUserMessage = deriveDisplayedUserMessageState(row.message.text, {
+    hasImageAttachments: userImages.length > 0,
+  });
   const terminalContexts = displayedUserMessage.contexts;
   const previewAnnotations: ParsedPreviewAnnotation[] = [];
   let visibleText = displayedUserMessage.visibleText;

apps/web/src/lib/terminalContext.test.ts

@@ -13,6 +13,7 @@ import {
   formatInlineTerminalContextLabel,
   formatTerminalContextLabel,
   hasTerminalContextText,
+  IMAGE_ONLY_BOOTSTRAP_PROMPT,
   INLINE_TERMINAL_CONTEXT_PLACEHOLDER,
   insertInlineTerminalContextPlaceholder,
   isTerminalContextExpired,
@@ -126,6 +127,27 @@ describe("terminalContext", () => {
     });
   });
 
+  it("hides the provider bootstrap prompt for image-only messages", () => {
+    expect(
+      deriveDisplayedUserMessageState(IMAGE_ONLY_BOOTSTRAP_PROMPT, {
+        hasImageAttachments: true,
+      }),
+    ).toEqual({
+      visibleText: "",
+      copyText: "",
+      contextCount: 0,
+      previewTitle: null,
+      contexts: [],
+      elementContexts: [],
+    });
+  });
+
+  it("preserves matching user text when no image is attached", () => {
+    expect(deriveDisplayedUserMessageState(IMAGE_ONLY_BOOTSTRAP_PROMPT).visibleText).toBe(
+      IMAGE_ONLY_BOOTSTRAP_PROMPT,
+    );
+  });
+
   it("preserves prompt text when no trailing terminal context block exists", () => {
     expect(extractTrailingTerminalContexts("No attached context")).toEqual({
       promptText: "No attached context",

apps/web/src/lib/terminalContext.ts

@@ -43,6 +43,8 @@ export interface ParsedTerminalContextEntry {
 }
 
 export const INLINE_TERMINAL_CONTEXT_PLACEHOLDER = "\uFFFC";
+export const IMAGE_ONLY_BOOTSTRAP_PROMPT =
+  "[User attached one or more images without additional text. Respond using the conversation context and the attached image(s).]";
 
 const TRAILING_TERMINAL_CONTEXT_BLOCK_PATTERN =
   /\n*<terminal_context>\n([\s\S]*?)\n<\/terminal_context>\s*$/;
@@ -245,15 +247,23 @@ export function extractTrailingTerminalContexts(prompt: string): ExtractedTermin
   };
 }
 
-export function deriveDisplayedUserMessageState(prompt: string): DisplayedUserMessageState {
+export function deriveDisplayedUserMessageState(
+  prompt: string,
+  options?: { readonly hasImageAttachments?: boolean },
+): DisplayedUserMessageState {
   // Order matters: send-time appends `<terminal_context>` first, then
   // `<element_context>` last. Strip element first so the (now-trailing)
   // terminal block can be matched by `extractTrailingTerminalContexts`.
   const extractedElement = extractTrailingElementContexts(prompt);
   const extractedTerminal = extractTrailingTerminalContexts(extractedElement.promptText);
+  const normalizedVisibleText = extractedTerminal.promptText.trim();
+  const hideImageOnlyBootstrapPrompt =
+    options?.hasImageAttachments === true &&
+    (normalizedVisibleText === IMAGE_ONLY_BOOTSTRAP_PROMPT ||
+      normalizedVisibleText === `Ultrathink:\n${IMAGE_ONLY_BOOTSTRAP_PROMPT}`);
   return {
-    visibleText: extractedTerminal.promptText,
-    copyText: prompt,
+    visibleText: hideImageOnlyBootstrapPrompt ? "" : extractedTerminal.promptText,
+    copyText: hideImageOnlyBootstrapPrompt ? "" : prompt,
     contextCount: extractedTerminal.contextCount,
     previewTitle: extractedTerminal.previewTitle,
     contexts: extractedTerminal.contexts,