Agents: harden CLI prompt image hydration (#51373)

CHANGELOG.md

@@ -205,6 +205,7 @@ Docs: https://docs.openclaw.ai
 - Plugins/Matrix: add a new Matrix plugin backed by the official `matrix-js-sdk`. If you are upgrading from the previous public Matrix plugin, follow the migration guide: https://docs.openclaw.ai/install/migrating-matrix Thanks @gumadeiras.
 - Discord/commands: switch native command deployment to Carbon reconcile by default so Discord restarts stop churning slash commands through OpenClaw’s local deploy path. (#46597) Thanks @huntharo and @thewilloftheshadow.
 - Plugins/Matrix: durably dedupe inbound room events across gateway restarts so previously handled Matrix messages are not replayed as new, while preserving clean-restart backlog delivery for unseen events. (#50922) thanks @gumadeiras
+- BlueBubbles/CLI agents: restore inbound prompt image refs for CLI routed turns, reapply embedded runner image size guardrails, and cover both CLI image transport paths with regression tests. (#51373)
 
 ## 2026.3.13
 

src/agents/cli-runner.helpers.test.ts

@@ -0,0 +1,116 @@
+import type { ImageContent } from "@mariozechner/pi-ai";
+import { beforeEach, describe, expect, it, vi } from "vitest";
+import { MAX_IMAGE_BYTES } from "../media/constants.js";
+import type { SandboxFsBridge } from "./sandbox/fs-bridge.js";
+
+const mocks = vi.hoisted(() => ({
+  loadImageFromRef: vi.fn(),
+  sanitizeImageBlocks: vi.fn(),
+}));
+
+vi.mock("./pi-embedded-runner/run/images.js", async (importOriginal) => {
+  const actual = await importOriginal<typeof import("./pi-embedded-runner/run/images.js")>();
+  return {
+    ...actual,
+    loadImageFromRef: (...args: unknown[]) => mocks.loadImageFromRef(...args),
+  };
+});
+
+vi.mock("./tool-images.js", async (importOriginal) => {
+  const actual = await importOriginal<typeof import("./tool-images.js")>();
+  return {
+    ...actual,
+    sanitizeImageBlocks: (...args: unknown[]) => mocks.sanitizeImageBlocks(...args),
+  };
+});
+
+import { loadPromptRefImages } from "./cli-runner/helpers.js";
+
+describe("loadPromptRefImages", () => {
+  beforeEach(() => {
+    mocks.loadImageFromRef.mockReset();
+    mocks.sanitizeImageBlocks.mockReset();
+    mocks.sanitizeImageBlocks.mockImplementation(async (images: ImageContent[]) => ({
+      images,
+      dropped: 0,
+    }));
+  });
+
+  it("returns empty results when the prompt has no image refs", async () => {
+    await expect(
+      loadPromptRefImages({
+        prompt: "just text",
+        workspaceDir: "/workspace",
+      }),
+    ).resolves.toEqual([]);
+
+    expect(mocks.loadImageFromRef).not.toHaveBeenCalled();
+    expect(mocks.sanitizeImageBlocks).not.toHaveBeenCalled();
+  });
+
+  it("passes the max-byte guardrail through load and sanitize", async () => {
+    const loadedImage: ImageContent = {
+      type: "image",
+      data: "c29tZS1pbWFnZQ==",
+      mimeType: "image/png",
+    };
+    const sanitizedImage: ImageContent = {
+      type: "image",
+      data: "c2FuaXRpemVkLWltYWdl",
+      mimeType: "image/jpeg",
+    };
+    const sandbox = {
+      root: "/sandbox",
+      bridge: {} as SandboxFsBridge,
+    };
+
+    mocks.loadImageFromRef.mockResolvedValueOnce(loadedImage);
+    mocks.sanitizeImageBlocks.mockResolvedValueOnce({ images: [sanitizedImage], dropped: 0 });
+
+    const result = await loadPromptRefImages({
+      prompt: "Look at /tmp/photo.png",
+      workspaceDir: "/workspace",
+      workspaceOnly: true,
+      sandbox,
+    });
+
+    const [ref, workspaceDir, options] = mocks.loadImageFromRef.mock.calls[0] ?? [];
+    expect(ref).toMatchObject({ resolved: "/tmp/photo.png", type: "path" });
+    expect(workspaceDir).toBe("/workspace");
+    expect(options).toEqual({
+      maxBytes: MAX_IMAGE_BYTES,
+      workspaceOnly: true,
+      sandbox,
+    });
+    expect(mocks.sanitizeImageBlocks).toHaveBeenCalledWith([loadedImage], "prompt:images", {
+      maxBytes: MAX_IMAGE_BYTES,
+    });
+    expect(result).toEqual([sanitizedImage]);
+  });
+
+  it("dedupes repeated refs and skips failed loads before sanitizing", async () => {
+    const loadedImage: ImageContent = {
+      type: "image",
+      data: "b25lLWltYWdl",
+      mimeType: "image/png",
+    };
+
+    mocks.loadImageFromRef.mockResolvedValueOnce(loadedImage).mockResolvedValueOnce(null);
+
+    const result = await loadPromptRefImages({
+      prompt: "Compare /tmp/a.png with /tmp/a.png and /tmp/b.png",
+      workspaceDir: "/workspace",
+    });
+
+    expect(mocks.loadImageFromRef).toHaveBeenCalledTimes(2);
+    expect(
+      mocks.loadImageFromRef.mock.calls.map(
+        (call: unknown[]) => (call[0] as { resolved?: string } | undefined)?.resolved,
+      ),
+    ).toEqual(["/tmp/a.png", "/tmp/b.png"]);
+    expect(mocks.sanitizeImageBlocks).toHaveBeenCalledWith([loadedImage], "prompt:images", {
+      maxBytes: MAX_IMAGE_BYTES,
+    });
+    expect(result).toEqual([loadedImage]);
+  });
+});

src/agents/cli-runner.test.ts

@@ -12,6 +12,8 @@ import type { WorkspaceBootstrapFile } from "./workspace.js";
 const supervisorSpawnMock = vi.fn();
 const enqueueSystemEventMock = vi.fn();
 const requestHeartbeatNowMock = vi.fn();
+const SMALL_PNG_BASE64 =
+  "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mP8/woAAn8B9FD5fHAAAAAASUVORK5CYII=";
 const hoisted = vi.hoisted(() => {
   type BootstrapContext = {
     bootstrapFiles: WorkspaceBootstrapFile[];
@@ -206,13 +208,7 @@ describe("runCliAgent with process supervisor", () => {
       path.join(resolvePreferredOpenClawTmpDir(), "openclaw-cli-prompt-image-"),
     );
     const sourceImage = path.join(tempDir, "bb-image.png");
-    await fs.writeFile(
-      sourceImage,
-      Buffer.from(
-        "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mP8/woAAn8B9FD5fHAAAAAASUVORK5CYII=",
-        "base64",
-      ),
-    );
+    await fs.writeFile(sourceImage, Buffer.from(SMALL_PNG_BASE64, "base64"));
 
     try {
       await runCliAgent({
@@ -237,6 +233,92 @@ describe("runCliAgent with process supervisor", () => {
     expect(argv[imageArgIndex + 1]).not.toBe(sourceImage);
   });
 
+  it("appends hydrated prompt media refs to generic backend prompts", async () => {
+    supervisorSpawnMock.mockResolvedValueOnce(
+      createManagedRun({
+        reason: "exit",
+        exitCode: 0,
+        exitSignal: null,
+        durationMs: 50,
+        stdout: "ok",
+        stderr: "",
+        timedOut: false,
+        noOutputTimedOut: false,
+      }),
+    );
+
+    const tempDir = await fs.mkdtemp(
+      path.join(resolvePreferredOpenClawTmpDir(), "openclaw-cli-prompt-image-generic-"),
+    );
+    const sourceImage = path.join(tempDir, "claude-image.png");
+    await fs.writeFile(sourceImage, Buffer.from(SMALL_PNG_BASE64, "base64"));
+
+    try {
+      await runCliAgent({
+        sessionId: "s1",
+        sessionFile: "/tmp/session.jsonl",
+        workspaceDir: tempDir,
+        prompt: `[media attached: ${sourceImage} (image/png)]\n\n<media:image>`,
+        provider: "claude-cli",
+        model: "claude-opus-4-1",
+        timeoutMs: 1_000,
+        runId: "run-prompt-image-generic",
+      });
+    } finally {
+      await fs.rm(tempDir, { recursive: true, force: true });
+    }
+
+    const input = supervisorSpawnMock.mock.calls[0]?.[0] as { argv?: string[]; input?: string };
+    const argv = input.argv ?? [];
+    expect(argv).not.toContain("--image");
+    const promptCarrier = [input.input ?? "", ...argv].join("\n");
+    const appendedPath = argv.find((value) => value.includes("openclaw-cli-images-"));
+    expect(appendedPath).toBeDefined();
+    expect(appendedPath).not.toBe(sourceImage);
+    expect(promptCarrier).toContain(appendedPath ?? "");
+  });
+
+  it("prefers explicit images over prompt refs", async () => {
+    supervisorSpawnMock.mockResolvedValueOnce(
+      createManagedRun({
+        reason: "exit",
+        exitCode: 0,
+        exitSignal: null,
+        durationMs: 50,
+        stdout: "ok",
+        stderr: "",
+        timedOut: false,
+        noOutputTimedOut: false,
+      }),
+    );
+
+    const tempDir = await fs.mkdtemp(
+      path.join(resolvePreferredOpenClawTmpDir(), "openclaw-cli-explicit-images-"),
+    );
+    const sourceImage = path.join(tempDir, "ignored-prompt-image.png");
+    await fs.writeFile(sourceImage, Buffer.from(SMALL_PNG_BASE64, "base64"));
+
+    try {
+      await runCliAgent({
+        sessionId: "s1",
+        sessionFile: "/tmp/session.jsonl",
+        workspaceDir: tempDir,
+        prompt: `[media attached: ${sourceImage} (image/png)]\n\n<media:image>`,
+        images: [{ type: "image", data: SMALL_PNG_BASE64, mimeType: "image/png" }],
+        provider: "codex-cli",
+        model: "gpt-5.2-codex",
+        timeoutMs: 1_000,
+        runId: "run-explicit-image-precedence",
+      });
+    } finally {
+      await fs.rm(tempDir, { recursive: true, force: true });
+    }
+
+    const input = supervisorSpawnMock.mock.calls[0]?.[0] as { argv?: string[] };
+    const argv = input.argv ?? [];
+    expect(argv.filter((arg) => arg === "--image")).toHaveLength(1);
+  });
+
   it("fails with timeout when no-output watchdog trips", async () => {
     supervisorSpawnMock.mockResolvedValueOnce(
       createManagedRun({

src/agents/cli-runner/helpers.ts

@@ -8,6 +8,7 @@ import { KeyedAsyncQueue } from "openclaw/plugin-sdk/keyed-async-queue";
 import type { ThinkLevel } from "../../auto-reply/thinking.js";
 import type { OpenClawConfig } from "../../config/config.js";
 import type { CliBackendConfig } from "../../config/types.js";
+import { MAX_IMAGE_BYTES } from "../../media/constants.js";
 import { buildTtsSystemPromptHint } from "../../tts/tts.js";
 import { isRecord } from "../../utils.js";
 import { buildModelAliasLines } from "../model-alias-lines.js";
@@ -15,9 +16,11 @@ import { resolveDefaultModelForAgent } from "../model-selection.js";
 import { resolveOwnerDisplaySetting } from "../owner-display.js";
 import type { EmbeddedContextFile } from "../pi-embedded-helpers.js";
 import { detectImageReferences, loadImageFromRef } from "../pi-embedded-runner/run/images.js";
+import type { SandboxFsBridge } from "../sandbox/fs-bridge.js";
 import { detectRuntimeShell } from "../shell-utils.js";
 import { buildSystemPromptParams } from "../system-prompt-params.js";
 import { buildAgentSystemPrompt } from "../system-prompt.js";
+import { sanitizeImageBlocks } from "../tool-images.js";
 export { buildCliSupervisorScopeKey, resolveCliNoOutputTimeoutMs } from "./reliability.js";
 
 const CLI_RUN_QUEUE = new KeyedAsyncQueue();
@@ -328,12 +331,16 @@ export function appendImagePathsToPrompt(prompt: string, paths: string[]): strin
 export async function loadPromptRefImages(params: {
   prompt: string;
   workspaceDir: string;
+  maxBytes?: number;
+  workspaceOnly?: boolean;
+  sandbox?: { root: string; bridge: SandboxFsBridge };
 }): Promise<ImageContent[]> {
   const refs = detectImageReferences(params.prompt);
   if (refs.length === 0) {
     return [];
   }
 
+  const maxBytes = params.maxBytes ?? MAX_IMAGE_BYTES;
   const seen = new Set<string>();
   const images: ImageContent[] = [];
   for (const ref of refs) {
@@ -342,12 +349,20 @@ export async function loadPromptRefImages(params: {
       continue;
     }
     seen.add(key);
-    const image = await loadImageFromRef(ref, params.workspaceDir);
+    const image = await loadImageFromRef(ref, params.workspaceDir, {
+      maxBytes,
+      workspaceOnly: params.workspaceOnly,
+      sandbox: params.sandbox,
+    });
     if (image) {
       images.push(image);
     }
   }
-  return images;
+
+  const { images: sanitizedImages } = await sanitizeImageBlocks(images, "prompt:images", {
+    maxBytes,
+  });
+  return sanitizedImages;
 }
 
 export async function writeCliImages(