diff --git a/src/agentMode/session/AgentSession.test.ts b/src/agentMode/session/AgentSession.test.ts
index d36164d4..fee779d6 100644
--- a/src/agentMode/session/AgentSession.test.ts
+++ b/src/agentMode/session/AgentSession.test.ts
@@ -1,6 +1,11 @@
 import { AI_SENDER, USER_SENDER } from "@/constants";
 import type { TFile } from "obsidian";
-import { AgentSession, buildPromptBlocks, tryReadExitPlanModeCall } from "./AgentSession";
+import {
+  AgentSession,
+  buildPromptBlocks,
+  buildUserDisplayContent,
+  tryReadExitPlanModeCall,
+} from "./AgentSession";
 import { AuthRequiredError, MethodUnsupportedError } from "./errors";
 import type {
   AgentToolCallOutput,
@@ -318,6 +323,33 @@ describe("AgentSession.restoreLabel", () => {
   });
 });
 
+describe("buildUserDisplayContent", () => {
+  it("returns undefined when there are no images", () => {
+    expect(buildUserDisplayContent("hi")).toBeUndefined();
+    expect(buildUserDisplayContent("hi", [])).toBeUndefined();
+    expect(buildUserDisplayContent("hi", [{ type: "text", text: "x" }])).toBeUndefined();
+  });
+
+  it("puts the prompt text first, then an image_url entry per image", () => {
+    expect(
+      buildUserDisplayContent("describe these", [
+        { type: "image", mimeType: "image/png", data: "AAA=" },
+        { type: "image", mimeType: "image/jpeg", data: "BBB=" },
+      ])
+    ).toEqual([
+      { type: "text", text: "describe these" },
+      { type: "image_url", image_url: { url: "data:image/png;base64,AAA=" } },
+      { type: "image_url", image_url: { url: "data:image/jpeg;base64,BBB=" } },
+    ]);
+  });
+
+  it("omits the text entry for an image-only message", () => {
+    expect(
+      buildUserDisplayContent("   ", [{ type: "image", mimeType: "image/png", data: "AAA=" }])
+    ).toEqual([{ type: "image_url", image_url: { url: "data:image/png;base64,AAA=" } }]);
+  });
+});
+
 describe("AgentSession.sendPrompt", () => {
   it("appends user + placeholder synchronously and resolves on stopReason", async () => {
     const mock = makeMockBackend();
@@ -361,7 +393,13 @@ describe("AgentSession.sendPrompt", () => {
     ]).turn;
     const messages = session.store.getDisplayMessages();
     expect(messages[0].message).toBe("describe");
-    expect(messages[0].content).toBeUndefined();
+    // The posted user bubble carries the prompt text plus the image as a
+    // renderable data-URL entry, while the backend still receives the original
+    // base64 image block.
+    expect(messages[0].content).toEqual([
+      { type: "text", text: "describe" },
+      { type: "image_url", image_url: { url: "data:image/png;base64,aGVsbG8=" } },
+    ]);
     expect(mock.prompt).toHaveBeenCalledWith({
       sessionId: "acp-1",
       prompt: [
@@ -371,6 +409,18 @@ describe("AgentSession.sendPrompt", () => {
     });
   });
 
+  it("leaves the user message content unset when no images are attached", () => {
+    const mock = makeMockBackend();
+    const session = new AgentSession({
+      backend: mock.asBackend,
+      backendSessionId: "acp-1",
+      internalId: "internal-1",
+      backendId: "opencode",
+    });
+    session.sendPrompt("just text");
+    expect(session.store.getDisplayMessages()[0].content).toBeUndefined();
+  });
+
   it("rejects if a turn is already in flight", () => {
     const mock = makeMockBackend();
     const session = new AgentSession({
diff --git a/src/agentMode/session/AgentSession.ts b/src/agentMode/session/AgentSession.ts
index 47e5bf1b..3c8e7d06 100644
--- a/src/agentMode/session/AgentSession.ts
+++ b/src/agentMode/session/AgentSession.ts
@@ -750,6 +750,9 @@ export class AgentSession {
       timestamp: formatDateTime(new Date()),
       isVisible: true,
       context,
+      // Surface attached images in the posted bubble. The backend consumes the
+      // original `promptContent` image blocks; this is a display-only projection.
+      content: buildUserDisplayContent(displayText, promptContent),
     };
     const userMessageId = this.store.addMessage(userMessage);
 
@@ -1506,6 +1509,42 @@ function tryParseJsonObject(s: string): Record<string, unknown> | null {
   }
 }
 
+type DisplayContentItem =
+  | { type: "text"; text: string }
+  | { type: "image_url"; image_url: { url: string } };
+
+/**
+ * Project a user prompt's text + outgoing image blocks into the display
+ * `content` shape the chat renderer understands: `ChatSingleMessage` renders a
+ * `content` array of `text` and `image_url` entries (text first, then images,
+ * mirroring the legacy chat composer). The backend still consumes the original
+ * `PromptContent` image blocks (base64); this is a display-only projection so
+ * attached images show in the posted user bubble.
+ *
+ * Returns undefined when there are no images — text-only messages render via the
+ * renderer's plain `message` fallback and don't need a `content` array. When
+ * images are present, the text entry is included only if there is prompt text,
+ * so an image-only message doesn't render an empty text line.
+ */
+export function buildUserDisplayContent(
+  displayText: string,
+  promptContent?: PromptContent[]
+): DisplayContentItem[] | undefined {
+  const images = (promptContent ?? []).filter(
+    (p): p is Extract<PromptContent, { type: "image" }> => p.type === "image"
+  );
+  if (images.length === 0) return undefined;
+  const content: DisplayContentItem[] = [];
+  if (displayText.trim()) content.push({ type: "text", text: displayText });
+  for (const img of images) {
+    content.push({
+      type: "image_url",
+      image_url: { url: `data:${img.mimeType};base64,${img.data}` },
+    });
+  }
+  return content;
+}
+
 export function buildPromptBlocks(
   displayText: string,
   context?: MessageContext,