fix(vom): add new texts and fix bugs

apps/extension/src/session-manager/__tests__/ref-store.test.ts

@@ -13,6 +13,8 @@ describe("RefStore", () => {
       backendNodeId: 42,
       tabId: 7,
       generation: 0,
+      kind: "dom",
+      capabilities: ["interact", "screenshot"],
     });
     expect(s.size()).toBe(1);
     expect(s.isEmpty()).toBe(false);
@@ -48,4 +50,41 @@ describe("RefStore", () => {
       cdpSessionId: "child-session",
     });
   });
+
+  it("preserves explicit capabilities for visual surface refs", () => {
+    const s = new RefStore();
+    s.set("e1", 42, {
+      tabId: 7,
+      kind: "surface",
+      capabilities: ["screenshot"],
+    });
+
+    expect(s.resolveEntry("e1")).toMatchObject({
+      kind: "surface",
+      capabilities: ["screenshot"],
+    });
+  });
+
+  it("never grants ordinary interaction to a surface through defaults or caller input", () => {
+    const s = new RefStore();
+    s.set("e1", 42, { tabId: 7, kind: "surface" });
+    s.set("e2", 43, {
+      tabId: 7,
+      kind: "surface",
+      capabilities: ["interact", "screenshot"],
+    });
+
+    expect(s.resolveEntry("e1")?.capabilities).toEqual(["screenshot"]);
+    expect(s.resolveEntry("e2")?.capabilities).toEqual(["screenshot"]);
+  });
+
+  it("copies caller-owned capability arrays", () => {
+    const s = new RefStore();
+    const capabilities: Array<"interact" | "screenshot"> = ["screenshot"];
+    s.set("e1", 42, { tabId: 7, capabilities });
+
+    capabilities.push("interact");
+
+    expect(s.resolveEntry("e1")?.capabilities).toEqual(["screenshot"]);
+  });
 });

apps/extension/src/session-manager/ref-store.ts

@@ -1,3 +1,5 @@
+import type { Rect } from "@browser-skill/vom";
+
 /**
  * Per-session map from `@e<N>` snapshot refs to a CDP node address.
  *
@@ -12,12 +14,17 @@
  * inside the `SessionContext`.
  */
 export type BackendNodeId = number;
+export type RefCapability = "interact" | "screenshot";
+export type RefTargetKind = "dom" | "surface";
 
 export interface RefEntry {
   backendNodeId: BackendNodeId;
   tabId: number | null;
   frameId?: string;
   cdpSessionId?: string;
+  visibleRect?: Rect;
+  kind: RefTargetKind;
+  capabilities: RefCapability[];
   generation: number;
 }
 
@@ -28,8 +35,19 @@ export type RefInput =
       tabId: number;
       frameId?: string;
       cdpSessionId?: string;
+      visibleRect?: Rect;
+      kind?: RefTargetKind;
+      capabilities?: RefCapability[];
     };
 
+function capabilitiesFor(
+  kind: RefTargetKind,
+  capabilities: RefCapability[] | undefined,
+): RefCapability[] {
+  if (kind === "surface") return ["screenshot"];
+  return capabilities ? [...capabilities] : ["interact", "screenshot"];
+}
+
 export class RefStore {
   private readonly map = new Map<string, RefEntry>();
   private generation = 0;
@@ -70,13 +88,20 @@ export class RefStore {
       tabId?: number;
       frameId?: string;
       cdpSessionId?: string;
+      visibleRect?: Rect;
+      kind?: RefTargetKind;
+      capabilities?: RefCapability[];
     } = {},
   ): void {
+    const kind = opts.kind ?? "dom";
     this.map.set(normaliseRef(ref), {
       backendNodeId: id,
       tabId: opts.tabId ?? null,
       ...(opts.frameId ? { frameId: opts.frameId } : {}),
       ...(opts.cdpSessionId ? { cdpSessionId: opts.cdpSessionId } : {}),
+      ...(opts.visibleRect ? { visibleRect: opts.visibleRect } : {}),
+      kind,
+      capabilities: capabilitiesFor(kind, opts.capabilities),
       generation: this.generation,
     });
   }
@@ -94,14 +119,20 @@ export class RefStore {
       return {
         backendNodeId: input,
         tabId: null,
+        kind: "dom",
+        capabilities: ["interact", "screenshot"],
         generation: this.generation,
       };
     }
+    const kind = input.kind ?? "dom";
     return {
       backendNodeId: input.backendNodeId,
       tabId: input.tabId,
       ...(input.frameId ? { frameId: input.frameId } : {}),
       ...(input.cdpSessionId ? { cdpSessionId: input.cdpSessionId } : {}),
+      ...(input.visibleRect ? { visibleRect: input.visibleRect } : {}),
+      kind,
+      capabilities: capabilitiesFor(kind, input.capabilities),
       generation: this.generation,
     };
   }

apps/extension/src/tools/__tests__/frame-geometry.test.ts

@@ -1,5 +1,9 @@
 import { describe, expect, it, vi } from "vitest";
-import { resolveFrameProjection, resolveNodeGeometry } from "../frame-geometry";
+import {
+  createFrameGeometryContext,
+  resolveFrameProjection,
+  resolveNodeGeometry,
+} from "../frame-geometry";
 import {
   clipPolygon,
   polygonArea,
@@ -37,6 +41,53 @@ describe("frame geometry projection", () => {
     expect(polygonBounds(projected.flat())).toEqual({ x: 224, y: 346, width: 200, height: 80 });
   });
 
+  it("projects nested same-target frame-local bounds through authoritative content quads", async () => {
+    const boxCalls: number[] = [];
+    const cdp: CdpRunner = {
+      send: vi.fn(async (_tabId, method, params) => {
+        if (method === "Page.getLayoutMetrics") {
+          return { cssLayoutViewport: { clientWidth: 1_000, clientHeight: 800 } };
+        }
+        if (method === "Page.createIsolatedWorld") return { executionContextId: 91 };
+        if (method === "Runtime.evaluate") {
+          return { result: { value: { width: 100, height: 80 } } };
+        }
+        if (method === "DOM.getBoxModel") {
+          const backendNodeId = (params as { backendNodeId?: number }).backendNodeId ?? 0;
+          boxCalls.push(backendNodeId);
+          return backendNodeId === 10
+            ? { model: { content: [30, 100.5, 330, 100.5, 330, 300.5, 30, 300.5] } }
+            : { model: { content: [45, 120.5, 145, 120.5, 145, 200.5, 45, 200.5] } };
+        }
+        throw new Error(`unexpected ${method}`);
+      }) as CdpRunner["send"],
+    };
+    const geometry = createFrameGeometryContext(cdp, {
+      rootFrameId: "main",
+      frames: [
+        { frameId: "main", target: { tabId: 4 } },
+        {
+          frameId: "child",
+          parentFrameId: "main",
+          ownerBackendNodeId: 10,
+          target: { tabId: 4 },
+        },
+        {
+          frameId: "nested",
+          parentFrameId: "child",
+          ownerBackendNodeId: 20,
+          target: { tabId: 4 },
+        },
+      ],
+    });
+
+    await expect(
+      geometry.projectFrameLocalRect("nested", { x: 12, y: 12, w: 40, h: 20 }),
+    ).resolves.toEqual({ x: 57, y: 132.5, width: 40, height: 20 });
+    await geometry.projectFrameLocalRect("nested", { x: 1, y: 1, w: 2, h: 2 });
+    expect(boxCalls.sort()).toEqual([10, 20]);
+  });
+
   it("uses a projective mapping for perspective-transformed iframe quads", () => {
     const quad: Quad = [
       { x: 0, y: 0 },

apps/extension/src/tools/__tests__/interaction.test.ts

@@ -123,6 +123,29 @@ describe("handleClick", () => {
     expect(fake.cdp.send).not.toHaveBeenCalled();
   });
 
+  it("rejects screenshot-only visual surface refs before issuing CDP calls", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    const ctx = await sm.start("aa11");
+    ctx.refStore.set("e3", 1234, {
+      tabId: 4,
+      kind: "surface",
+      capabilities: ["screenshot"],
+    });
+    const fake = makeFakeCdp({});
+
+    const res = await handleClick(
+      sm,
+      { session_id: "aa11", ref: "@e3" },
+      { cdp: fake.cdp, tabsApi: fake.tabsApi },
+    );
+
+    expect(res).toMatchObject({
+      code: "permission_denied",
+      data: { reason: "ref_capability_denied", required_capability: "interact" },
+    });
+    expect(fake.cdp.send).not.toHaveBeenCalled();
+  });
+
   it("clicks by ref, computes the quad centre, dispatches three mouse events", async () => {
     const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
     const ctx = await sm.start("aa11");

apps/extension/src/tools/__tests__/observation.test.ts

@@ -36,6 +36,14 @@ function fakeAgentWindow(ids: number[]) {
 const TINY_PNG =
   "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR4nGNgYGBgAAAABQABh6FO1AAAAABJRU5ErkJggg==";
 
+function pngWithIhdrDimensions(width: number, height: number): string {
+  const bytes = Uint8Array.from(atob(TINY_PNG), (character) => character.charCodeAt(0));
+  const view = new DataView(bytes.buffer);
+  view.setUint32(16, width);
+  view.setUint32(20, height);
+  return btoa(String.fromCharCode(...bytes));
+}
+
 function makeScreenshotDeps(
   opts: {
     cdp?: CdpRunner;
@@ -86,6 +94,7 @@ describe("stripDataUrlPrefix", () => {
 describe("parsePngDimensions", () => {
   it("parses width/height from the IHDR chunk", () => {
     expect(parsePngDimensions(TINY_PNG)).toEqual({ width: 1, height: 1 });
+    expect(parsePngDimensions(pngWithIhdrDimensions(100, 80))).toEqual({ width: 100, height: 80 });
   });
   it("returns null on non-PNG input", () => {
     expect(parsePngDimensions("not-a-png-payload-just-random-base64-text-zzzzzzzzz")).toBeNull();
@@ -120,6 +129,20 @@ describe("handleScreenshot", () => {
     expect(get).not.toHaveBeenCalled();
   });
 
+  it("fails closed when a full-tab capture is not a valid PNG", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    await sm.start("aa11");
+    const res = await handleScreenshot(
+      sm,
+      { session_id: "aa11" },
+      makeScreenshotDeps({
+        captureVisibleTab: vi.fn(async () => "data:image/png;base64,bm90LWEtcG5n"),
+      }),
+    );
+
+    expect(res).toMatchObject({ code: "cdp_failed", message: /invalid PNG data/ });
+  });
+
   it("returns not_found when Agent Window has no active tab", async () => {
     const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
     await sm.start("aa11");
@@ -300,6 +323,250 @@ describe("handleScreenshot", () => {
     expect(clip).toMatchObject({ x: 10, y: 20, width: 100, height: 40 });
   });
 
+  it("captures a bounded visible crop for a visual surface without scrolling it", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    const ctx = await sm.start("aa11");
+    ctx.refStore.set("e5", 999, {
+      tabId: 7,
+      kind: "surface",
+      capabilities: ["screenshot"],
+      visibleRect: { x: 0, y: 0, w: 4096, h: 4096 },
+    });
+    const { cdp, sent } = makeFakeCdp({
+      "Page.getLayoutMetrics": () => ({
+        cssLayoutViewport: { clientWidth: 4096, clientHeight: 4096 },
+      }),
+      "DOM.getContentQuads": () => ({ quads: [[0, 0, 4096, 0, 4096, 4096, 0, 4096]] }),
+      "Page.captureScreenshot": () => ({ data: TINY_PNG }),
+    });
+
+    const res = await handleScreenshot(
+      sm,
+      { session_id: "aa11", ref: "@e5", tab_id: 7 },
+      makeScreenshotDeps({
+        cdp,
+        get: vi.fn(async () => ({ id: 7, windowId: 100, active: false }) as chrome.tabs.Tab),
+        query: vi.fn(),
+        captureVisibleTab: vi.fn(),
+      }),
+    );
+
+    if ("code" in res) throw new Error(`unexpected error: ${JSON.stringify(res)}`);
+    expect(sent.some((call) => call.method === "DOM.scrollIntoViewIfNeeded")).toBe(false);
+    const clip = sent.find((call) => call.method === "Page.captureScreenshot")?.params as {
+      clip?: { width: number; height: number; scale: number };
+    };
+    expect(clip.clip).toMatchObject({ width: 4096, height: 4096 });
+    expect(clip.clip?.scale).toBeCloseTo(Math.sqrt(4_000_000 / (4096 * 4096)));
+  });
+
+  it("retries once when actual surface PNG dimensions exceed the pixel budget", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    const ctx = await sm.start("aa11");
+    ctx.refStore.set("e5", 999, {
+      tabId: 7,
+      kind: "surface",
+      visibleRect: { x: 0, y: 0, w: 1_000, h: 1_000 },
+    });
+    let captures = 0;
+    const { cdp, sent } = makeFakeCdp({
+      "Runtime.evaluate": () => ({ result: { value: 1 } }),
+      "Page.getLayoutMetrics": () => ({
+        cssLayoutViewport: { clientWidth: 1_000, clientHeight: 1_000 },
+      }),
+      "DOM.getContentQuads": () => ({
+        quads: [[0, 0, 1_000, 0, 1_000, 1_000, 0, 1_000]],
+      }),
+      "Page.captureScreenshot": () => ({
+        data:
+          ++captures === 1
+            ? pngWithIhdrDimensions(4_000, 4_000)
+            : pngWithIhdrDimensions(1_998, 1_998),
+      }),
+    });
+
+    const res = await handleScreenshot(
+      sm,
+      { session_id: "aa11", ref: "@e5", tab_id: 7 },
+      makeScreenshotDeps({
+        cdp,
+        get: vi.fn(async () => ({ id: 7, windowId: 100, active: false }) as chrome.tabs.Tab),
+        query: vi.fn(),
+        captureVisibleTab: vi.fn(),
+      }),
+    );
+
+    if ("code" in res) throw new Error(`unexpected error: ${JSON.stringify(res)}`);
+    expect(res).toMatchObject({ width: 1_998, height: 1_998 });
+    const calls = sent.filter((call) => call.method === "Page.captureScreenshot");
+    expect(calls).toHaveLength(2);
+    expect((calls[1]?.params as { clip?: { scale?: number } }).clip?.scale).toBeLessThan(0.5);
+  });
+
+  it("fails closed when the retried surface PNG still exceeds the pixel budget", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    const ctx = await sm.start("aa11");
+    ctx.refStore.set("e5", 999, {
+      tabId: 7,
+      kind: "surface",
+      visibleRect: { x: 0, y: 0, w: 1_000, h: 1_000 },
+    });
+    const { cdp, sent } = makeFakeCdp({
+      "Runtime.evaluate": () => ({ result: { value: 1 } }),
+      "Page.getLayoutMetrics": () => ({
+        cssLayoutViewport: { clientWidth: 1_000, clientHeight: 1_000 },
+      }),
+      "DOM.getContentQuads": () => ({
+        quads: [[0, 0, 1_000, 0, 1_000, 1_000, 0, 1_000]],
+      }),
+      "Page.captureScreenshot": () => ({ data: pngWithIhdrDimensions(4_000, 4_000) }),
+    });
+
+    const res = await handleScreenshot(
+      sm,
+      { session_id: "aa11", ref: "@e5", tab_id: 7 },
+      makeScreenshotDeps({
+        cdp,
+        get: vi.fn(async () => ({ id: 7, windowId: 100, active: false }) as chrome.tabs.Tab),
+        query: vi.fn(),
+        captureVisibleTab: vi.fn(),
+      }),
+    );
+
+    expect(res).toEqual({
+      code: "cdp_failed",
+      message: "Page.captureScreenshot exceeded the visual surface pixel budget",
+    });
+    expect(sent.filter((call) => call.method === "Page.captureScreenshot")).toHaveLength(2);
+  });
+
+  it("crops a visual surface screenshot to its observation-time visible region", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    const ctx = await sm.start("aa11");
+    ctx.refStore.set("e5", 999, {
+      tabId: 7,
+      kind: "surface",
+      visibleRect: { x: 25, y: 30, w: 50, h: 40 },
+    });
+    const { cdp, sent } = makeFakeCdp({
+      "Page.getLayoutMetrics": () => ({
+        cssLayoutViewport: { clientWidth: 800, clientHeight: 600 },
+      }),
+      "DOM.getContentQuads": () => ({ quads: [[0, 0, 200, 0, 200, 100, 0, 100]] }),
+      "Page.captureScreenshot": () => ({ data: pngWithIhdrDimensions(100, 80) }),
+    });
+
+    const res = await handleScreenshot(
+      sm,
+      { session_id: "aa11", ref: "@e5", tab_id: 7 },
+      makeScreenshotDeps({
+        cdp,
+        get: vi.fn(async () => ({ id: 7, windowId: 100, active: false }) as chrome.tabs.Tab),
+        query: vi.fn(),
+        captureVisibleTab: vi.fn(),
+      }),
+    );
+
+    if ("code" in res) throw new Error(`unexpected error: ${JSON.stringify(res)}`);
+    const clip = sent.find((call) => call.method === "Page.captureScreenshot")?.params as {
+      clip?: { x: number; y: number; width: number; height: number };
+    };
+    expect(clip.clip).toMatchObject({ x: 25, y: 30, width: 50, height: 40 });
+    expect(res).toMatchObject({ width: 100, height: 80 });
+  });
+
+  it("routes OOPIF surface geometry through its child session and captures top coordinates", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    const ctx = await sm.start("aa11");
+    ctx.refStore.set("e5", 999, {
+      tabId: 7,
+      frameId: "child-frame",
+      cdpSessionId: "child-session",
+      kind: "surface",
+      visibleRect: { x: 224, y: 346, w: 200, h: 80 },
+    });
+    const { cdp, sent } = makeFakeCdp({
+      "DOM.getBoxModel": () => ({
+        model: { content: [204, 306, 604, 306, 604, 506, 204, 506] },
+      }),
+      "Page.getLayoutMetrics": () => ({
+        cssLayoutViewport: { clientWidth: 1000, clientHeight: 800 },
+      }),
+      "Page.captureScreenshot": () => ({ data: TINY_PNG }),
+    });
+    cdp.getFrameGraph = vi.fn(async () => ({
+      rootFrameId: "main",
+      frames: [
+        { frameId: "main", target: { tabId: 7 } },
+        {
+          frameId: "child-frame",
+          parentFrameId: "main",
+          ownerBackendNodeId: 99,
+          target: { tabId: 7, sessionId: "child-session" },
+        },
+      ],
+    }));
+    const targetCalls: Array<{ sessionId?: string; method: string }> = [];
+    cdp.sendToTarget = vi.fn(async (target, method) => {
+      targetCalls.push({ sessionId: target.sessionId, method });
+      if (method === "DOM.getContentQuads") {
+        return { quads: [[10, 20, 110, 20, 110, 60, 10, 60]] };
+      }
+      if (method === "Page.getLayoutMetrics") {
+        return { cssLayoutViewport: { clientWidth: 200, clientHeight: 100 } };
+      }
+      throw new Error(`unexpected child CDP call ${method}`);
+    }) as CdpRunner["sendToTarget"];
+
+    const res = await handleScreenshot(
+      sm,
+      { session_id: "aa11", ref: "@e5", tab_id: 7 },
+      makeScreenshotDeps({
+        cdp,
+        get: vi.fn(async () => ({ id: 7, windowId: 100, active: false }) as chrome.tabs.Tab),
+        query: vi.fn(),
+        captureVisibleTab: vi.fn(),
+      }),
+    );
+
+    if ("code" in res) throw new Error(`unexpected error: ${JSON.stringify(res)}`);
+    expect(targetCalls).toEqual([
+      { sessionId: "child-session", method: "DOM.getContentQuads" },
+      { sessionId: "child-session", method: "Page.getLayoutMetrics" },
+    ]);
+    const clip = sent.find((call) => call.method === "Page.captureScreenshot")?.params as {
+      clip?: { x: number; y: number; width: number; height: number };
+    };
+    expect(clip.clip).toMatchObject({ x: 224, y: 346, width: 200, height: 80 });
+  });
+
+  it("fails closed when a ref capture is not a valid PNG", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    const ctx = await sm.start("aa11");
+    ctx.refStore.set("e5", 999, {
+      tabId: 7,
+      kind: "surface",
+      visibleRect: { x: 0, y: 0, w: 50, h: 50 },
+    });
+    const { cdp } = makeFakeCdp({
+      "DOM.getContentQuads": () => ({ quads: [[0, 0, 50, 0, 50, 50, 0, 50]] }),
+      "Page.captureScreenshot": () => ({ data: "bm90LWEtcG5n" }),
+    });
+
+    const res = await handleScreenshot(
+      sm,
+      { session_id: "aa11", ref: "@e5", tab_id: 7 },
+      makeScreenshotDeps({
+        cdp,
+        get: vi.fn(async () => ({ id: 7, windowId: 100, active: false }) as chrome.tabs.Tab),
+        query: vi.fn(),
+        captureVisibleTab: vi.fn(),
+      }),
+    );
+
+    expect(res).toMatchObject({ code: "cdp_failed", message: /invalid PNG data/ });
+  });
+
   it("returns not_found for unknown ref", async () => {
     const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
     await sm.start("aa11");
@@ -2833,6 +3100,229 @@ describe("handleSnapshot", () => {
     );
   });
 
+  it("discovers rendered canvases only in observe and stores a screenshot-only ref", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    const ctx = await sm.start("aa11");
+    const root: CdpAxNode = {
+      nodeId: "1",
+      role: { type: "role", value: "RootWebArea" },
+      name: { type: "computedString", value: "Canvas app" },
+      backendDOMNodeId: 100,
+    };
+    const strings = ["body", "canvas", "static", "auto"];
+    const i = (value: string) => strings.indexOf(value);
+    const send = vi.fn(async (_tabId: number, method: string) => {
+      if (method === "Accessibility.enable" || method === "DOMSnapshot.enable") return {};
+      if (method === "Accessibility.getFullAXTree") return { nodes: [root] };
+      if (method === "Page.getLayoutMetrics") {
+        return { cssLayoutViewport: { clientWidth: 1000, clientHeight: 800, pageX: 0, pageY: 0 } };
+      }
+      if (method === "DOMSnapshot.captureSnapshot") {
+        return {
+          strings,
+          documents: [
+            {
+              nodes: {
+                parentIndex: [-1, 0],
+                nodeName: [i("body"), i("canvas")],
+                backendNodeId: [100, 200],
+                attributes: [[], []],
+              },
+              layout: {
+                nodeIndex: [0, 1],
+                styles: [
+                  [i("static"), i("auto"), i("auto")],
+                  [i("static"), i("auto"), i("auto")],
+                ],
+                bounds: [
+                  [0, 0, 1000, 800],
+                  [20, 30, 800, 500],
+                ],
+                paintOrders: [0, 1],
+              },
+            },
+          ],
+        };
+      }
+      throw new Error(`unexpected CDP method ${method}`);
+    });
+    const deps = {
+      cdp: {
+        send: send as unknown as <T = unknown>(
+          tabId: number,
+          method: string,
+          params?: object,
+        ) => Promise<T>,
+        trackSessionTab: vi.fn(),
+      },
+      tabsApi: {
+        get: vi.fn(
+          async (tabId: number) => ({ id: tabId, windowId: 100, active: true }) as chrome.tabs.Tab,
+        ),
+        query: vi.fn(async () => [{ id: 4, windowId: 100, active: true } as chrome.tabs.Tab]),
+      },
+      conditionalSurfaceProbe: false,
+    };
+
+    const observed = await handleObserve(sm, { session_id: "aa11" }, deps);
+    if ("code" in observed) throw new Error(`unexpected error: ${JSON.stringify(observed)}`);
+    expect(observed.text).toContain('@e1 surface "canvas visual surface"');
+    expect(ctx.refStore.resolveEntry("e1")).toMatchObject({
+      backendNodeId: 200,
+      kind: "surface",
+      capabilities: ["screenshot"],
+    });
+
+    const snapshot = await handleSnapshot(sm, { session_id: "aa11" }, deps);
+    if ("code" in snapshot) throw new Error(`unexpected error: ${JSON.stringify(snapshot)}`);
+    expect(snapshot.text).not.toContain("visual surface");
+    expect(snapshot.ref_count).toBe(0);
+  });
+
+  it("preserves an overflow-clipped surface region through observation and screenshot", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    const ctx = await sm.start("aa11");
+    const root: CdpAxNode = {
+      nodeId: "1",
+      role: { type: "role", value: "RootWebArea" },
+      name: { type: "computedString", value: "Clipped canvas" },
+      backendDOMNodeId: 100,
+    };
+    const strings = [
+      "body",
+      "div",
+      "canvas",
+      "aria-label",
+      "clipped canvas",
+      "static",
+      "auto",
+      "visible",
+      "hidden",
+      "1",
+    ];
+    const i = (value: string) => strings.indexOf(value);
+    const send = vi.fn(async (_tabId: number, method: string, _params?: object) => {
+      if (method === "Accessibility.enable" || method === "DOMSnapshot.enable") return {};
+      if (method === "Accessibility.getFullAXTree") return { nodes: [root] };
+      if (method === "Page.getLayoutMetrics") {
+        return {
+          cssLayoutViewport: { clientWidth: 1000, clientHeight: 800, pageX: 0, pageY: 0 },
+          layoutViewport: { clientWidth: 1000, clientHeight: 800 },
+        };
+      }
+      if (method === "DOMSnapshot.captureSnapshot") {
+        return {
+          strings,
+          documents: [
+            {
+              nodes: {
+                parentIndex: [-1, 0, 1],
+                nodeName: [i("body"), i("div"), i("canvas")],
+                backendNodeId: [100, 150, 200],
+                attributes: [[], [], [i("aria-label"), i("clipped canvas")]],
+              },
+              layout: {
+                nodeIndex: [0, 1, 2],
+                styles: [
+                  [
+                    i("static"),
+                    i("auto"),
+                    i("auto"),
+                    i("visible"),
+                    i("1"),
+                    i("visible"),
+                    i("visible"),
+                  ],
+                  [
+                    i("static"),
+                    i("auto"),
+                    i("auto"),
+                    i("visible"),
+                    i("1"),
+                    i("hidden"),
+                    i("hidden"),
+                  ],
+                  [
+                    i("static"),
+                    i("auto"),
+                    i("auto"),
+                    i("visible"),
+                    i("1"),
+                    i("visible"),
+                    i("visible"),
+                  ],
+                ],
+                bounds: [
+                  [0, 0, 1000, 800],
+                  [10, 20, 100, 70],
+                  [55, 30, 140, 60],
+                ],
+                paintOrders: [0, 1, 2],
+              },
+            },
+          ],
+        };
+      }
+      if (method === "DOM.getContentQuads") {
+        return { quads: [[55, 30, 195, 30, 195, 90, 55, 90]] };
+      }
+      if (method === "Runtime.evaluate") return { result: { value: 1 } };
+      if (method === "Page.captureScreenshot") {
+        return { data: pngWithIhdrDimensions(55, 60) };
+      }
+      throw new Error(`unexpected CDP method ${method}`);
+    });
+    const cdp = {
+      send: send as unknown as CdpRunner["send"],
+      trackSessionTab: vi.fn(),
+      getFrameGraph: vi.fn(async () => ({
+        rootFrameId: "root",
+        frames: [{ frameId: "root", target: { tabId: 4 } }],
+      })),
+    } as CdpRunner;
+    const deps = {
+      cdp,
+      tabsApi: {
+        get: vi.fn(
+          async (tabId: number) => ({ id: tabId, windowId: 100, active: true }) as chrome.tabs.Tab,
+        ),
+        query: vi.fn(async () => [{ id: 4, windowId: 100, active: true } as chrome.tabs.Tab]),
+      },
+      conditionalSurfaceProbe: false,
+    };
+
+    const observed = await handleObserve(sm, { session_id: "aa11" }, deps);
+    if ("code" in observed) throw new Error(`unexpected error: ${JSON.stringify(observed)}`);
+    expect(observed.text).toContain(
+      '@e1 surface "clipped canvas" [bounds=55,30,55,60; rendering=canvas;',
+    );
+    expect(ctx.refStore.resolveEntry("e1")).toMatchObject({
+      backendNodeId: 200,
+      kind: "surface",
+      visibleRect: { x: 55, y: 30, w: 55, h: 60 },
+    });
+
+    const screenshot = await handleScreenshot(
+      sm,
+      { session_id: "aa11", ref: "@e1", tab_id: 4 },
+      makeScreenshotDeps({
+        cdp,
+        get: vi.fn(async () => ({ id: 4, windowId: 100, active: true }) as chrome.tabs.Tab),
+      }),
+    );
+    if ("code" in screenshot) {
+      throw new Error(`unexpected error: ${JSON.stringify(screenshot)}`);
+    }
+    const captureCall = send.mock.calls.find(([, method]) => method === "Page.captureScreenshot");
+    expect((captureCall?.[2] as { clip?: object }).clip).toMatchObject({
+      x: 55,
+      y: 30,
+      width: 55,
+      height: 60,
+    });
+    expect(screenshot).toMatchObject({ width: 55, height: 60 });
+  });
+
   it("allows passive snapshots of explicit user-window tabs", async () => {
     const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
     await sm.start("aa11");

apps/extension/src/tools/__tests__/snapshot-ref.test.ts

@@ -24,10 +24,14 @@ describe("lookupSnapshotRef", () => {
     expect(lookupSnapshotRef(ctx, "@e3", 4)).toEqual({
       backendNodeId: 1234,
       refKey: "e3",
+      kind: "dom",
+      capabilities: ["interact", "screenshot"],
     });
     expect(lookupSnapshotRef(ctx, "e3", 4)).toEqual({
       backendNodeId: 1234,
       refKey: "e3",
+      kind: "dom",
+      capabilities: ["interact", "screenshot"],
     });
   });
 
@@ -53,6 +57,8 @@ describe("resolveSnapshotRef", () => {
       tabId: 4,
       frameId: "child-frame",
       cdpSessionId: "child-session",
+      kind: "dom",
+      capabilities: ["interact", "screenshot"],
     });
 
     const expected = {
@@ -60,6 +66,8 @@ describe("resolveSnapshotRef", () => {
       refKey: "e3",
       frameId: "child-frame",
       cdpSessionId: "child-session",
+      kind: "dom" as const,
+      capabilities: ["interact", "screenshot"] as const,
     };
     expect(lookupSnapshotRef(ctx, "@e3", 4)).toEqual(expected);
     expect(resolveSnapshotRef(ctx, "@e3", 4)).toEqual(expected);
@@ -105,6 +113,28 @@ describe("resolveSnapshotRef", () => {
     expect(resolveSnapshotRef(ctx, "@e3", 4)).toEqual({
       backendNodeId: 1234,
       refKey: "e3",
+      kind: "dom",
+      capabilities: ["interact", "screenshot"],
+    });
+  });
+
+  it("rejects an operation outside the ref's declared capabilities", async () => {
+    const sm = new SessionManager({ agentWindow: fakeAgentWindow([100]) });
+    const ctx = await sm.start("aa11");
+    ctx.refStore.set("e3", 1234, {
+      tabId: 4,
+      kind: "surface",
+      capabilities: ["screenshot"],
+    });
+
+    expect(resolveSnapshotRef(ctx, "@e3", 4, "interact")).toMatchObject({
+      code: "permission_denied",
+      message: "ref @e3 does not support interact",
+      data: {
+        reason: "ref_capability_denied",
+        required_capability: "interact",
+        kind: "surface",
+      },
     });
   });
 });

apps/extension/src/tools/__tests__/visual-screenshot-plan.test.ts

@@ -0,0 +1,37 @@
+import { describe, expect, it } from "vitest";
+import {
+  correctedVisualScreenshotScale,
+  planVisualScreenshot,
+  screenshotFitsPixelBudget,
+} from "../visual-screenshot-plan";
+
+const budget = { maxEdge: 2_048, maxPixels: 4_000_000 };
+
+describe("visual screenshot pixel planning", () => {
+  it.each([1, 1.25, 1.5, 2, 3])("accounts for raster scale %s", (rasterScaleEstimate) => {
+    const plan = planVisualScreenshot({
+      cssRect: { width: 2_000, height: 1_000 },
+      rasterScaleEstimate,
+      budget,
+    });
+    const width = 2_000 * rasterScaleEstimate * plan.clipScale;
+    const height = 1_000 * rasterScaleEstimate * plan.clipScale;
+
+    expect(Math.max(width, height)).toBeLessThanOrEqual(budget.maxEdge);
+    expect(width * height).toBeLessThanOrEqual(budget.maxPixels);
+  });
+
+  it("corrects a scale from actual PNG dimensions", () => {
+    const corrected = correctedVisualScreenshotScale({
+      previousClipScale: 1,
+      actualWidth: 4_000,
+      actualHeight: 2_000,
+      budget,
+    });
+
+    expect(corrected).not.toBeNull();
+    expect(corrected as number).toBeLessThan(2_048 / 4_000);
+    expect(screenshotFitsPixelBudget({ width: 2_000, height: 1_000 }, budget)).toBe(true);
+    expect(screenshotFitsPixelBudget({ width: 4_000, height: 2_000 }, budget)).toBe(false);
+  });
+});

apps/extension/src/tools/frame-geometry.ts

@@ -1,4 +1,9 @@
-import { type CdpFrame, type CdpFrameGraph, type CdpTarget } from "@/browser-driver/frame-graph";
+import {
+  type CdpFrame,
+  type CdpFrameGraph,
+  type CdpTarget,
+  cdpTargetKey,
+} from "@/browser-driver/frame-graph";
 import type { RpcError } from "@/transport/types";
 import { nodeContentRegion, scrollNodeIntoView } from "./element-geometry";
 import {
@@ -10,6 +15,7 @@ import {
   parseCdpQuad,
   polygonArea,
   polygonCentroid,
+  projectRectToViewport,
   projectRegionToViewport,
   type Quad,
   type Region,
@@ -84,6 +90,41 @@ async function targetViewport(cdp: CdpRunner, target: CdpTarget): Promise<Size |
   return width > 0 && height > 0 ? { width, height } : null;
 }
 
+async function sameTargetFrameViewport(cdp: CdpRunner, frame: CdpFrame): Promise<Size | null> {
+  const world = await sendToCdpTarget<{ executionContextId?: number }>(
+    cdp,
+    frame.target,
+    "Page.createIsolatedWorld",
+    {
+      frameId: frame.frameId,
+      worldName: "bsk-frame-geometry",
+      grantUniveralAccess: false,
+    },
+  );
+  if (world.executionContextId === undefined) return null;
+  const result = await sendToCdpTarget<{ result?: { value?: unknown } }>(
+    cdp,
+    frame.target,
+    "Runtime.evaluate",
+    {
+      expression: "({width:window.innerWidth,height:window.innerHeight})",
+      contextId: world.executionContextId,
+      returnByValue: true,
+    },
+  );
+  const value = result.result?.value as { width?: unknown; height?: unknown } | undefined;
+  const width = value?.width;
+  const height = value?.height;
+  return typeof width === "number" &&
+    Number.isFinite(width) &&
+    width > 0 &&
+    typeof height === "number" &&
+    Number.isFinite(height) &&
+    height > 0
+    ? { width, height }
+    : null;
+}
+
 async function ownerContentQuad(
   cdp: CdpRunner,
   parent: CdpFrame,
@@ -98,26 +139,150 @@ async function ownerContentQuad(
   return parseCdpQuad(result.model?.content);
 }
 
-async function sameTargetFrameClips(
+export interface FrameGeometryContext {
+  resolveFrameProjection(frameId: string): Promise<GeometryProjection | null>;
+  projectFrameLocalRect(
+    frameId: string,
+    rect: { x: number; y: number; w: number; h: number },
+  ): Promise<ViewportRect | null>;
+}
+
+export function createFrameGeometryContext(
   cdp: CdpRunner,
-  byId: Map<string, CdpFrame>,
-  frame: CdpFrame,
-  targetRoot: CdpFrame,
-): Promise<Polygon[] | null> {
-  const clips: Polygon[] = [];
-  let current = frame;
-  const seen = new Set<string>();
-  while (current.frameId !== targetRoot.frameId) {
-    if (seen.has(current.frameId) || !current.parentFrameId) return null;
-    seen.add(current.frameId);
-    const parent = byId.get(current.parentFrameId);
-    if (!parent || current.ownerBackendNodeId === undefined) return null;
-    const clip = await ownerContentQuad(cdp, parent, current.ownerBackendNodeId);
-    if (!clip) return null;
-    clips.push(clip);
-    current = parent;
-  }
-  return clips;
+  graph: CdpFrameGraph,
+): FrameGeometryContext {
+  const byId = frameMap(graph);
+  const targetViewports = new Map<string, Promise<Size | null>>();
+  const frameViewports = new Map<string, Promise<Size | null>>();
+  const ownerQuads = new Map<string, Promise<Quad | null>>();
+  const projections = new Map<string, Promise<GeometryProjection | null>>();
+  const localProjections = new Map<string, Promise<GeometryProjection | null>>();
+
+  const cachedTargetViewport = (target: CdpTarget): Promise<Size | null> => {
+    const key = cdpTargetKey(target);
+    let value = targetViewports.get(key);
+    if (!value) {
+      value = targetViewport(cdp, target);
+      targetViewports.set(key, value);
+    }
+    return value;
+  };
+  const cachedFrameViewport = (frame: CdpFrame): Promise<Size | null> => {
+    let value = frameViewports.get(frame.frameId);
+    if (!value) {
+      const root = targetRootFrame(byId, frame);
+      value =
+        root?.frameId === frame.frameId
+          ? cachedTargetViewport(frame.target)
+          : sameTargetFrameViewport(cdp, frame);
+      frameViewports.set(frame.frameId, value);
+    }
+    return value;
+  };
+  const cachedOwnerQuad = (parent: CdpFrame, backendNodeId: number): Promise<Quad | null> => {
+    const key = `${cdpTargetKey(parent.target)}:${backendNodeId}`;
+    let value = ownerQuads.get(key);
+    if (!value) {
+      value = ownerContentQuad(cdp, parent, backendNodeId);
+      ownerQuads.set(key, value);
+    }
+    return value;
+  };
+  const cachedSameTargetClips = async (
+    frame: CdpFrame,
+    root: CdpFrame,
+  ): Promise<Polygon[] | null> => {
+    const clips: Polygon[] = [];
+    let current = frame;
+    const seen = new Set<string>();
+    while (current.frameId !== root.frameId) {
+      if (seen.has(current.frameId) || !current.parentFrameId) return null;
+      seen.add(current.frameId);
+      const parent = byId.get(current.parentFrameId);
+      if (!parent || current.ownerBackendNodeId === undefined) return null;
+      const clip = await cachedOwnerQuad(parent, current.ownerBackendNodeId);
+      if (!clip) return null;
+      clips.push(clip);
+      current = parent;
+    }
+    return clips;
+  };
+
+  const buildTargetProjection = async (frameId: string): Promise<GeometryProjection | null> => {
+    const frame = byId.get(frameId);
+    if (!frame) return null;
+    let root = targetRootFrame(byId, frame);
+    if (!root) return null;
+    const sourceViewport = await cachedTargetViewport(root.target);
+    if (!sourceViewport) return null;
+    const sourceClips = await cachedSameTargetClips(frame, root);
+    if (!sourceClips) return null;
+    const edges: ProjectiveEdge[] = [];
+    while (root.parentFrameId) {
+      const parent = byId.get(root.parentFrameId);
+      if (!parent || root.ownerBackendNodeId === undefined) return null;
+      const destinationQuad = await cachedOwnerQuad(parent, root.ownerBackendNodeId);
+      if (!destinationQuad) return null;
+      const source = await cachedTargetViewport(root.target);
+      if (!source) return null;
+      const parentRoot = targetRootFrame(byId, parent);
+      if (!parentRoot) return null;
+      const destinationClips = await cachedSameTargetClips(parent, parentRoot);
+      if (!destinationClips) return null;
+      edges.push({ sourceViewport: source, destinationQuad, destinationClips });
+      root = parentRoot;
+    }
+    const topViewport = await cachedTargetViewport(root.target);
+    return topViewport ? { sourceClips, edges, topViewport } : null;
+  };
+
+  const resolveFrameProjectionCached = (frameId: string): Promise<GeometryProjection | null> => {
+    let value = projections.get(frameId);
+    if (!value) {
+      value = buildTargetProjection(frameId);
+      projections.set(frameId, value);
+    }
+    return value;
+  };
+
+  const buildLocalProjection = async (frameId: string): Promise<GeometryProjection | null> => {
+    const frame = byId.get(frameId);
+    if (!frame) return null;
+    const root = targetRootFrame(byId, frame);
+    if (!root) return null;
+    const targetProjection = await resolveFrameProjectionCached(root.frameId);
+    if (!targetProjection) return null;
+    if (root.frameId === frame.frameId) return targetProjection;
+    if (!frame.parentFrameId || frame.ownerBackendNodeId === undefined) return null;
+    const parent = byId.get(frame.parentFrameId);
+    if (!parent) return null;
+    const sourceViewport = await cachedFrameViewport(frame);
+    const destinationQuad = await cachedOwnerQuad(parent, frame.ownerBackendNodeId);
+    const destinationClips = await cachedSameTargetClips(frame, root);
+    if (!sourceViewport || !destinationQuad || !destinationClips) return null;
+    return {
+      sourceClips: [],
+      edges: [{ sourceViewport, destinationQuad, destinationClips }, ...targetProjection.edges],
+      topViewport: targetProjection.topViewport,
+    };
+  };
+
+  const localProjection = (frameId: string): Promise<GeometryProjection | null> => {
+    let value = localProjections.get(frameId);
+    if (!value) {
+      value = buildLocalProjection(frameId);
+      localProjections.set(frameId, value);
+    }
+    return value;
+  };
+
+  return {
+    resolveFrameProjection: resolveFrameProjectionCached,
+    async projectFrameLocalRect(frameId, rect) {
+      const projection = await localProjection(frameId);
+      return projection ? projectRectToViewport(rect, projection) : null;
+    },
+  };
 }
 
 export async function resolveFrameProjection(
@@ -125,36 +290,7 @@ export async function resolveFrameProjection(
   graph: CdpFrameGraph,
   frameId: string,
 ): Promise<GeometryProjection | null> {
-  const byId = frameMap(graph);
-  const frame = byId.get(frameId);
-  if (!frame) return null;
-  let targetRoot = targetRootFrame(byId, frame);
-  if (!targetRoot) return null;
-  const sourceViewport = await targetViewport(cdp, targetRoot.target);
-  if (!sourceViewport) return null;
-  const sourceClips = await sameTargetFrameClips(cdp, byId, frame, targetRoot);
-  if (!sourceClips) return null;
-
-  const edges: ProjectiveEdge[] = [];
-  while (targetRoot.parentFrameId) {
-    const parent = byId.get(targetRoot.parentFrameId);
-    if (!parent || targetRoot.ownerBackendNodeId === undefined) return null;
-    const destinationQuad = await ownerContentQuad(cdp, parent, targetRoot.ownerBackendNodeId);
-    if (!destinationQuad) return null;
-    const source =
-      edges.length === 0 ? sourceViewport : await targetViewport(cdp, targetRoot.target);
-    if (!source) return null;
-    const parentTargetRoot = targetRootFrame(byId, parent);
-    if (!parentTargetRoot) return null;
-    const destinationClips = await sameTargetFrameClips(cdp, byId, parent, parentTargetRoot);
-    if (!destinationClips) return null;
-    edges.push({ sourceViewport: source, destinationQuad, destinationClips });
-    targetRoot = parentTargetRoot;
-  }
-
-  const topViewport =
-    edges.length === 0 ? sourceViewport : await targetViewport(cdp, targetRoot.target);
-  return topViewport ? { sourceClips, edges, topViewport } : null;
+  return createFrameGeometryContext(cdp, graph).resolveFrameProjection(frameId);
 }
 
 async function loadFrameGraph(cdp: CdpRunner, tabId: number): Promise<CdpFrameGraph | null> {

apps/extension/src/tools/geometry.ts

@@ -245,22 +245,3 @@ export function projectRectToViewport(
   if (!rect) return null;
   return regionBounds(projectRegionToViewport([rectPolygon(rect)], projection));
 }
-
-export function childFrameProjection(
-  parent: GeometryProjection,
-  ownerRectInParent: { x: number; y: number; w: number; h: number },
-): GeometryProjection {
-  const destinationQuad = rectPolygon(ownerRectInParent);
-  return {
-    sourceClips: [],
-    edges: [
-      {
-        sourceViewport: { width: ownerRectInParent.w, height: ownerRectInParent.h },
-        destinationQuad,
-        destinationClips: parent.sourceClips,
-      },
-      ...parent.edges,
-    ],
-    topViewport: parent.topViewport,
-  };
-}

apps/extension/src/tools/interaction.ts

@@ -169,7 +169,7 @@ export async function resolveBackendNode(
     };
   }
   if (hasRef) {
-    const resolved = resolveSnapshotRef(ctx, params.ref as string, target.tabId);
+    const resolved = resolveSnapshotRef(ctx, params.ref as string, target.tabId, "interact");
     if (isRpcError(resolved)) return resolved;
     return {
       backendNodeId: resolved.backendNodeId,

apps/extension/src/tools/observation.ts

@@ -8,6 +8,7 @@ import {
   applyVomInteractionRecovery,
   type CondSurface,
   isVomReferenceNode,
+  type Rect,
   renderVom,
   type VomNode,
   type VomOptions,
@@ -47,6 +48,13 @@ import {
   type ToolEffect,
 } from "./shared";
 import { resolveSnapshotRef } from "./snapshot-ref";
+import { topLevelScreenshotRasterScale } from "./viewport-metrics";
+import {
+  correctedVisualScreenshotScale,
+  planVisualScreenshot,
+  type ScreenshotPixelBudget,
+  screenshotFitsPixelBudget,
+} from "./visual-screenshot-plan";
 import {
   type CapturedNode,
   type CapturedSurfaceProbe,
@@ -62,6 +70,11 @@ import {
   type CaptureVomObservationResult,
   projectRecordSafeObservation,
 } from "./vom/record-safe-observation";
+import {
+  clusterRenderedSurfaces,
+  discoverRenderedSurfaces,
+  projectRenderedSurfaces,
+} from "./vom/rendered-surfaces";
 import {
   buildSemanticGraph,
   buildSemanticVomScene,
@@ -93,6 +106,13 @@ export const normaliseRef = sharedNormaliseRef;
 // screenshot — `tool.screenshot`
 // ---------------------------------------------------------------------------
 
+const VISUAL_SURFACE_SCREENSHOT_MAX_EDGE = 2_048;
+const VISUAL_SURFACE_SCREENSHOT_MAX_PIXELS = 4_000_000;
+const VISUAL_SURFACE_SCREENSHOT_BUDGET: ScreenshotPixelBudget = {
+  maxEdge: VISUAL_SURFACE_SCREENSHOT_MAX_EDGE,
+  maxPixels: VISUAL_SURFACE_SCREENSHOT_MAX_PIXELS,
+};
+
 /**
  * Strip the `data:image/...;base64,` prefix from a Chrome
  * `captureVisibleTab` dataURL and return the raw base64 payload.
@@ -107,8 +127,8 @@ export function stripDataUrlPrefix(dataUrl: string): string {
 
 /**
  * Parse a PNG's IHDR chunk and return `(width, height)`. Returns
- * `null` on any malformed input so callers fall back to `0/0` instead
- * of throwing.
+ * `null` on malformed input so callers can reject dimensions that are
+ * not authoritative instead of guessing from a CSS-space rectangle.
  *
  * PNG layout: 8-byte signature, then a 4-byte length, 4-byte type
  * ("IHDR"), then the chunk data — width is bytes 16-19 BE, height is
@@ -120,9 +140,9 @@ export function parsePngDimensions(base64: string): { width: number; height: num
     const head = base64.length > 64 ? base64.slice(0, 64) : base64;
     const bin = atob(head);
     if (bin.length < 24) return null;
-    if (bin.charCodeAt(0) !== 0x89 || bin.charCodeAt(1) !== 0x50 || bin.charCodeAt(2) !== 0x4e) {
-      return null;
-    }
+    const expectedSignature = [0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a];
+    if (expectedSignature.some((byte, index) => bin.charCodeAt(index) !== byte)) return null;
+    if (bin.slice(12, 16) !== "IHDR") return null;
     const u32 = (off: number) =>
       (bin.charCodeAt(off) << 24) |
       (bin.charCodeAt(off + 1) << 16) |
@@ -185,6 +205,8 @@ async function captureElementScreenshot(
   target: CdpTarget,
   backendNodeId: number,
   frameId?: string,
+  visualSurface = false,
+  observedVisibleRect?: Rect,
   signal?: AbortSignal,
 ): Promise<{ image_base64: string; width: number; height: number } | RpcError> {
   if (signal?.aborted) return cancelled("screenshot");
@@ -192,33 +214,74 @@ async function captureElementScreenshot(
     cdp,
     tabId,
     { target, backendNodeId, ...(frameId ? { frameId } : {}) },
-    { scrollIntoView: true },
+    { scrollIntoView: !visualSurface },
   );
   if (isRpcError(geometry)) return geometry;
   if (signal?.aborted) return cancelled("screenshot");
-  const rect = geometry.topBounds;
+  const liveRect = geometry.topBounds;
+  const rect =
+    visualSurface && observedVisibleRect
+      ? (() => {
+          const x = Math.max(liveRect.x, observedVisibleRect.x);
+          const y = Math.max(liveRect.y, observedVisibleRect.y);
+          const right = Math.min(
+            liveRect.x + liveRect.width,
+            observedVisibleRect.x + observedVisibleRect.w,
+          );
+          const bottom = Math.min(
+            liveRect.y + liveRect.height,
+            observedVisibleRect.y + observedVisibleRect.h,
+          );
+          return right > x && bottom > y ? { x, y, width: right - x, height: bottom - y } : null;
+        })()
+      : liveRect;
+  if (!rect) return { code: "permission_denied", message: "surface is no longer visible" };
+  const rasterScaleEstimate = visualSurface ? await topLevelScreenshotRasterScale(cdp, tabId) : 1;
+  let scale = visualSurface
+    ? planVisualScreenshot({
+        cssRect: rect,
+        rasterScaleEstimate,
+        budget: VISUAL_SURFACE_SCREENSHOT_BUDGET,
+      }).clipScale
+    : 1;
 
   try {
-    const shot = await cdp.send<{ data?: string }>(tabId, "Page.captureScreenshot", {
-      format: "png",
-      clip: {
-        x: rect.x,
-        y: rect.y,
-        width: rect.width,
-        height: rect.height,
-        scale: 1,
-      },
-    });
-    if (signal?.aborted) return cancelled("screenshot");
-    const image_base64 = shot.data ?? "";
-    if (!image_base64) {
-      return { code: "cdp_failed", message: "Page.captureScreenshot returned no data" };
+    for (let attempt = 0; attempt < (visualSurface ? 2 : 1); attempt += 1) {
+      const shot = await cdp.send<{ data?: string }>(tabId, "Page.captureScreenshot", {
+        format: "png",
+        clip: {
+          x: rect.x,
+          y: rect.y,
+          width: rect.width,
+          height: rect.height,
+          scale,
+        },
+      });
+      if (signal?.aborted) return cancelled("screenshot");
+      const image_base64 = shot.data ?? "";
+      if (!image_base64) {
+        return { code: "cdp_failed", message: "Page.captureScreenshot returned no data" };
+      }
+      const dims = parsePngDimensions(image_base64);
+      if (!dims) {
+        return { code: "cdp_failed", message: "Page.captureScreenshot returned invalid PNG data" };
+      }
+      if (!visualSurface || screenshotFitsPixelBudget(dims, VISUAL_SURFACE_SCREENSHOT_BUDGET)) {
+        return { image_base64, width: dims.width, height: dims.height };
+      }
+      const corrected = correctedVisualScreenshotScale({
+        previousClipScale: scale,
+        actualWidth: dims.width,
+        actualHeight: dims.height,
+        budget: VISUAL_SURFACE_SCREENSHOT_BUDGET,
+      });
+      if (attempt === 1 || corrected === null) break;
+      scale = corrected;
     }
-    const dims = parsePngDimensions(image_base64) ?? {
-      width: Math.round(rect.width),
-      height: Math.round(rect.height),
+    return {
+      code: "cdp_failed",
+      message: "Page.captureScreenshot exceeded the visual surface pixel budget",
     };
-    return { image_base64, width: dims.width, height: dims.height };
   } catch (err) {
     return {
       code: "cdp_failed",
@@ -313,8 +376,14 @@ export async function handleScreenshot(
     if (!deps.cdp) {
       return { code: "cdp_failed", message: "screenshot ref capture requires CDP" };
     }
-    const node = resolveSnapshotRef(ctx, ref, target.tabId);
+    const node = resolveSnapshotRef(ctx, ref, target.tabId, "screenshot");
     if (isRpcError(node)) return node;
+    if (node.kind === "surface" && !node.visibleRect) {
+      return {
+        code: "permission_denied",
+        message: `surface ref ${ref} has no observation-time visible bounds`,
+      };
+    }
     if (signal?.aborted) return cancelled("screenshot");
     deps.cdp.trackSessionTab?.(ctx.sessionId, target.tabId);
     await deps.cdp.ensureAttachedToUrl?.(target.tabId, target.url);
@@ -333,6 +402,8 @@ export async function handleScreenshot(
           nodeTarget,
           node.backendNodeId,
           node.frameId,
+          node.kind === "surface",
+          node.visibleRect,
           signal,
         ),
       deps.sendToTab,
@@ -364,7 +435,10 @@ export async function handleScreenshot(
   if (isRpcError(captured)) return captured;
   if (signal?.aborted) return cancelled("screenshot");
   const image_base64 = captured;
-  const dims = parsePngDimensions(image_base64) ?? { width: 0, height: 0 };
+  const dims = parsePngDimensions(image_base64);
+  if (!dims) {
+    return { code: "cdp_failed", message: "screenshot capture returned invalid PNG data" };
+  }
   return withShotDialogs({
     image_base64,
     width: dims.width,
@@ -1083,6 +1157,7 @@ async function runHoverProbes(
 
 export interface CaptureVomObservationOptions extends VomOptions {
   conditionalSurfaceProbe?: boolean;
+  renderedSurfaceDiscovery?: boolean;
   hoverProbeBypassOverlay?: (tabId: number, enabled: boolean) => Promise<void>;
   signal?: AbortSignal;
 }
@@ -1131,7 +1206,20 @@ export async function captureVomObservation(
     captured,
     hoverProbes.surfaceProbes,
   );
-  const rendered = renderVom(decoratedScene, {
+  const observedScene = options.renderedSurfaceDiscovery
+    ? projectRenderedSurfaces(
+        decoratedScene,
+        normalizedDocuments,
+        clusterRenderedSurfaces(
+          discoverRenderedSurfaces(
+            normalizedDocuments,
+            captured.viewport,
+            captured.excludedBackendNodeIds,
+          ),
+        ),
+      )
+    : decoratedScene;
+  const rendered = renderVom(observedScene, {
     maxDepth: options.maxDepth,
     maxTokens: options.maxTokens,
     redactValues: options.redactValues,
@@ -1186,6 +1274,7 @@ async function handleVomObservation(
       maxTokens: params.max_tokens,
       activeRegionPolicy: true,
       conditionalSurfaceProbe: effectiveConditionalSurfaceProbe,
+      renderedSurfaceDiscovery: toolName === "observe",
       hoverProbeBypassOverlay: deps.hoverProbeBypassOverlay,
       signal,
     });
@@ -1203,6 +1292,9 @@ async function handleVomObservation(
             tabId: target.tabId,
             ...(ref.frameId ? { frameId: ref.frameId } : {}),
             ...(refTarget?.sessionId ? { cdpSessionId: refTarget.sessionId } : {}),
+            ...(ref.visibleRect ? { visibleRect: ref.visibleRect } : {}),
+            ...(ref.kind ? { kind: ref.kind } : {}),
+            ...(ref.capabilities ? { capabilities: ref.capabilities } : {}),
           },
         ] as const;
       }),

apps/extension/src/tools/snapshot-ref.ts

@@ -2,8 +2,9 @@
 // compound CDP node identity via the session RefStore, and build stable
 // `ref_not_found` errors for hard-failure tool paths.
 
+import type { Rect } from "@browser-skill/vom";
 import type { SessionContext } from "@/session-manager/manager";
-import { normaliseRef } from "@/session-manager/ref-store";
+import { normaliseRef, type RefCapability, type RefTargetKind } from "@/session-manager/ref-store";
 import type { RpcError } from "@/transport/types";
 import { rpcError } from "./errors";
 
@@ -12,6 +13,9 @@ export interface SnapshotRefLookup {
   refKey: string;
   frameId?: string;
   cdpSessionId?: string;
+  visibleRect?: Rect;
+  kind: RefTargetKind;
+  capabilities: RefCapability[];
 }
 
 function refEntryForTab(ctx: SessionContext, refKey: string, tabId: number) {
@@ -37,6 +41,9 @@ export function lookupSnapshotRef(
     refKey,
     ...(entry.frameId ? { frameId: entry.frameId } : {}),
     ...(entry.cdpSessionId ? { cdpSessionId: entry.cdpSessionId } : {}),
+    ...(entry.visibleRect ? { visibleRect: entry.visibleRect } : {}),
+    kind: entry.kind,
+    capabilities: entry.capabilities,
   };
 }
 
@@ -49,6 +56,7 @@ export function resolveSnapshotRef(
   ctx: SessionContext,
   ref: string,
   tabId: number,
+  requiredCapability?: RefCapability,
 ): SnapshotRefLookup | RpcError {
   const looked = lookupSnapshotRef(ctx, ref, tabId);
   if (looked === null) {
@@ -58,5 +66,13 @@ export function resolveSnapshotRef(
       `ref ${ref} unknown for tab ${tabId} in session ${ctx.sessionId}`,
     );
   }
+  if (requiredCapability && !looked.capabilities.includes(requiredCapability)) {
+    return rpcError(
+      "permission_denied",
+      "ref_capability_denied",
+      `ref ${ref} does not support ${requiredCapability}`,
+      { ref: looked.refKey, required_capability: requiredCapability, kind: looked.kind },
+    );
+  }
   return looked;
 }

apps/extension/src/tools/viewport-metrics.ts

@@ -0,0 +1,18 @@
+import type { CdpRunner } from "./shared";
+
+/** Estimate the raster/CSS ratio used by top-level Page.captureScreenshot. */
+export async function topLevelScreenshotRasterScale(
+  cdp: CdpRunner,
+  tabId: number,
+): Promise<number> {
+  try {
+    const reply = await cdp.send<{ result?: { value?: unknown } }>(tabId, "Runtime.evaluate", {
+      expression: "window.devicePixelRatio",
+      returnByValue: true,
+    });
+    const value = reply.result?.value;
+    return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : 1;
+  } catch {
+    return 1;
+  }
+}

apps/extension/src/tools/visual-screenshot-plan.ts

@@ -0,0 +1,52 @@
+export interface ScreenshotPixelBudget {
+  maxEdge: number;
+  maxPixels: number;
+}
+
+export interface VisualScreenshotPlan {
+  clipScale: number;
+}
+
+export function planVisualScreenshot(input: {
+  cssRect: { width: number; height: number };
+  rasterScaleEstimate: number;
+  budget: ScreenshotPixelBudget;
+}): VisualScreenshotPlan {
+  const rasterScale =
+    Number.isFinite(input.rasterScaleEstimate) && input.rasterScaleEstimate > 0
+      ? input.rasterScaleEstimate
+      : 1;
+  const { width, height } = input.cssRect;
+  const scale = Math.min(
+    1,
+    input.budget.maxEdge / (width * rasterScale),
+    input.budget.maxEdge / (height * rasterScale),
+    Math.sqrt(input.budget.maxPixels / (width * height * rasterScale * rasterScale)),
+  );
+  return { clipScale: Math.max(Number.EPSILON, scale) };
+}
+
+export function correctedVisualScreenshotScale(input: {
+  previousClipScale: number;
+  actualWidth: number;
+  actualHeight: number;
+  budget: ScreenshotPixelBudget;
+}): number | null {
+  const correction = Math.min(
+    input.budget.maxEdge / input.actualWidth,
+    input.budget.maxEdge / input.actualHeight,
+    Math.sqrt(input.budget.maxPixels / (input.actualWidth * input.actualHeight)),
+  );
+  if (!Number.isFinite(correction) || correction >= 1 || correction <= 0) return null;
+  return Math.max(Number.EPSILON, input.previousClipScale * correction * 0.999);
+}
+
+export function screenshotFitsPixelBudget(
+  dimensions: { width: number; height: number },
+  budget: ScreenshotPixelBudget,
+): boolean {
+  return (
+    Math.max(dimensions.width, dimensions.height) <= budget.maxEdge &&
+    dimensions.width * dimensions.height <= budget.maxPixels
+  );
+}

apps/extension/src/tools/vom/__tests__/capture.test.ts

@@ -1,6 +1,7 @@
 import { describe, expect, it, vi } from "vitest";
 import { OVERLAY_HOST_MARKER_ATTR, OVERLAY_HOST_NAME } from "../../../lib/overlay-bridge";
 import { captureViewModel, collectOverlayExcludedBackendIds, probeHoverSurfaces } from "../capture";
+import { discoverRenderedSurfaces } from "../rendered-surfaces";
 
 // Minimal but format-accurate captureSnapshot reply: a body with one
 // fixed full-screen overlay div carrying a password input.
@@ -735,6 +736,44 @@ describe("captureViewModel", () => {
     expect(nodes.find((node) => node.backendNodeId === 14)?.rendered).toBe(false);
   });
 
+  it("propagates fully transparent ancestor suppression to painted descendants", async () => {
+    const S = ["html", "body", "div", "canvas", "static", "auto", "visible", "1", "0"];
+    const i = (value: string) => S.indexOf(value);
+    const snapshot = {
+      strings: S,
+      documents: [
+        {
+          nodes: {
+            parentIndex: [-1, 0, 1, 2],
+            nodeName: [i("html"), i("body"), i("div"), i("canvas")],
+            backendNodeId: [10, 11, 12, 13],
+            attributes: [[], [], [], []],
+          },
+          layout: {
+            nodeIndex: [1, 2, 3],
+            styles: [
+              [i("static"), i("auto"), i("auto"), i("visible"), i("1")],
+              [i("static"), i("auto"), i("auto"), i("visible"), i("0")],
+              [i("static"), i("auto"), i("auto"), i("visible"), i("1")],
+            ],
+            bounds: [
+              [0, 0, 1000, 800],
+              [10, 10, 200, 100],
+              [10, 10, 200, 100],
+            ],
+            paintOrders: [0, 1, 2],
+          },
+        },
+      ],
+    };
+
+    const { nodes } = await captureViewModel(makeCdp(snapshot), 4);
+    const canvas = nodes.find((node) => node.backendNodeId === 13);
+
+    expect(canvas?.rendered).toBe(true);
+    expect(canvas?.visuallySuppressed).toBe(true);
+  });
+
   it("subtracts scroll offset to make rects viewport-relative", async () => {
     const cdp = {
       send: vi.fn(async (_t: number, method: string) => {
@@ -754,8 +793,10 @@ describe("captureViewModel", () => {
     expect(div?.rect).toMatchObject({ y: 0, h: 600 });
   });
 
-  it("normalizes device-pixel bounds by devicePixelRatio", async () => {
-    const S = ["html", "body", "div", "position", "fixed", "static", "pointer-events", "auto"];
+  it.each([
+    1, 2,
+  ])("keeps CSS-space bounds stable at devicePixelRatio=%i", async (deviceScaleFactor) => {
+    const S = ["html", "body", "canvas", "position", "fixed", "static", "pointer-events", "auto"];
     const i = (s: string) => S.indexOf(s);
     const snapshot = {
       strings: S,
@@ -763,7 +804,7 @@ describe("captureViewModel", () => {
         {
           nodes: {
             parentIndex: [-1, 0, 1],
-            nodeName: [i("html"), i("body"), i("div")],
+            nodeName: [i("html"), i("body"), i("canvas")],
             backendNodeId: [10, 11, 12],
             attributes: [[], [], []],
           },
@@ -774,8 +815,8 @@ describe("captureViewModel", () => {
               [i("fixed"), i("auto")],
             ],
             bounds: [
-              [0, 0, 2000, 3200],
-              [0, 0, 2000, 1600],
+              [0, 0, 1000 * deviceScaleFactor, 1600 * deviceScaleFactor],
+              [0, 0, 1000 * deviceScaleFactor, 800 * deviceScaleFactor],
             ],
             paintOrders: [0, 50],
           },
@@ -789,14 +830,19 @@ describe("captureViewModel", () => {
         if (method === "Page.getLayoutMetrics") {
           return {
             cssLayoutViewport: { clientWidth: 1000, clientHeight: 800, pageX: 0, pageY: 0 },
-            layoutViewport: { clientWidth: 2000, clientHeight: 1600 },
+            layoutViewport: {
+              clientWidth: 1000 * deviceScaleFactor,
+              clientHeight: 800 * deviceScaleFactor,
+            },
           };
         }
         throw new Error(method);
       }) as unknown as <T>(tabId: number, method: string, params?: object) => Promise<T>,
     };
     const { nodes } = await captureViewModel(cdp, 4);
-    expect(nodes.find((n) => n.backendNodeId === 12)?.rect).toEqual({
+    const canvas = nodes.find((node) => node.backendNodeId === 12);
+    expect(canvas?.tag).toBe("canvas");
+    expect(canvas?.rect).toEqual({
       x: 0,
       y: 0,
       w: 1000,
@@ -972,7 +1018,7 @@ describe("captureViewModel", () => {
     expect(input?.attrs.type).toBe("text");
     expect(input?.ownerFrameBackendNodeId).toBe(13);
     expect(input?.localRect).toEqual({ x: 0, y: 0, w: 200, h: 40 });
-    expect(input?.rect).toEqual({ x: 100, y: 300, w: 200, h: 40 });
+    expect(input?.rect).toBeNull();
     expect(rootFrameId).toBe("main-frame");
     expect(frameNodes?.get("child-frame")).toBe(subNodes);
     expect(frameOwnerBackendNodeIds?.get("child-frame")).toBe(13);
@@ -1144,10 +1190,95 @@ describe("captureViewModel", () => {
     const input = iframeNodes.get(23)?.find((n) => n.backendNodeId === 31);
 
     expect(nestedIframe?.localRect).toEqual({ x: 5, y: 6, w: 100, h: 80 });
-    expect(nestedIframe?.rect).toEqual({ x: 15, y: 26, w: 100, h: 80 });
+    expect(nestedIframe?.rect).toBeNull();
     expect(input?.ownerFrameBackendNodeId).toBe(23);
     expect(input?.localRect).toEqual({ x: 1, y: 2, w: 40, h: 20 });
-    expect(input?.rect).toEqual({ x: 16, y: 28, w: 40, h: 20 });
+    expect(input?.rect).toBeNull();
+  });
+
+  it("keeps same-origin iframe canvas frame-local until frame geometry is resolved", async () => {
+    const captureAtDpr = async (dpr: number) => {
+      const S = ["html", "body", "iframe", "canvas", "static", "auto"];
+      const i = (s: string) => S.indexOf(s);
+      const scale = (values: number[]) => values.map((value) => value * dpr);
+      const snapshot = {
+        strings: S,
+        documents: [
+          {
+            frameId: "main",
+            nodes: {
+              parentIndex: [-1, 0, 1],
+              nodeName: [i("html"), i("body"), i("iframe")],
+              backendNodeId: [10, 11, 13],
+              attributes: [[], [], []],
+              contentDocumentIndex: { index: [2], value: [1] },
+            },
+            layout: {
+              nodeIndex: [1, 2],
+              styles: [
+                [i("static"), i("auto")],
+                [i("static"), i("auto")],
+              ],
+              bounds: [scale([0, 0, 1000, 800]), scale([100, 150, 300, 200])],
+              paintOrders: [0, 1],
+            },
+          },
+          {
+            frameId: "child",
+            nodes: {
+              parentIndex: [-1, 0],
+              nodeName: [i("body"), i("canvas")],
+              backendNodeId: [20, 21],
+              attributes: [[], []],
+            },
+            layout: {
+              nodeIndex: [0, 1],
+              styles: [
+                [i("static"), i("auto")],
+                [i("static"), i("auto")],
+              ],
+              bounds: [scale([0, 0, 300, 200]), scale([20, 30, 120, 60])],
+              paintOrders: [0, 1],
+            },
+          },
+        ],
+      };
+      const cdp = {
+        send: vi.fn(async (_tabId: number, method: string) => {
+          if (method === "DOMSnapshot.enable") return {};
+          if (method === "DOMSnapshot.captureSnapshot") return snapshot;
+          if (method === "Page.getLayoutMetrics") {
+            return {
+              cssLayoutViewport: { clientWidth: 1000, clientHeight: 800, pageX: 0, pageY: 0 },
+              layoutViewport: { clientWidth: 1000 * dpr, clientHeight: 800 * dpr },
+            };
+          }
+          throw new Error(method);
+        }) as unknown as <T>(tabId: number, method: string, params?: object) => Promise<T>,
+      };
+      const captured = await captureViewModel(cdp, 4);
+      const childNodes = captured.frameNodes?.get("child") ?? [];
+      return discoverRenderedSurfaces(
+        [
+          {
+            frameId: "child",
+            contextScopeId: "child",
+            parentFrameId: "main",
+            ownerBackendNodeId: 13,
+            target: { tabId: 4 },
+            domNodes: childNodes,
+            axNodes: [],
+          },
+        ],
+        captured.viewport,
+      );
+    };
+
+    const atDpr1 = await captureAtDpr(1);
+    const atDpr2 = await captureAtDpr(2);
+
+    expect(atDpr1).toEqual([]);
+    expect(atDpr2).toEqual(atDpr1);
   });
 
   it("normalizes bounds then subtracts CSS scroll at dpr>1", async () => {

apps/extension/src/tools/vom/__tests__/frame-capture.test.ts

@@ -50,6 +50,105 @@ function childSnapshot(frameId: string, backendNodeId: number) {
 }
 
 describe("captureFrameData", () => {
+  it("does not request frame geometry for a root-only page", async () => {
+    const root = { ...ownerNode(1, 0), tag: "body" };
+    const captured: CapturedViewModel = {
+      nodes: [root],
+      viewport: { width: 1000, height: 800 },
+      iframeNodes: new Map(),
+      frameNodes: new Map([["main", [root]]]),
+      rootFrameId: "main",
+      excludedBackendNodeIds: new Set(),
+    };
+    const send = vi.fn(async (_tabId, method) => {
+      if (method === "Accessibility.enable") return {};
+      if (method === "Accessibility.getFullAXTree") return { nodes: [] };
+      throw new Error(`unexpected ${method}`);
+    });
+    const cdp: CdpRunner = {
+      send: send as CdpRunner["send"],
+      getFrameGraph: vi.fn(async () => ({
+        rootFrameId: "main",
+        frames: [{ frameId: "main", target: { tabId: 4 } }],
+      })),
+    };
+
+    await captureFrameData(cdp, 4, captured);
+
+    expect(send.mock.calls.some(([, method]) => method === "Page.createIsolatedWorld")).toBe(false);
+    expect(send.mock.calls.some(([, method]) => method === "DOM.getBoxModel")).toBe(false);
+    expect(send.mock.calls.some(([, method]) => method === "Page.getLayoutMetrics")).toBe(false);
+  });
+
+  it("projects same-origin frame-local rects through the iframe content box", async () => {
+    const owner = ownerNode(10, 50);
+    const canvas: CapturedNode = {
+      backendNodeId: 101,
+      parentBackendNodeId: 100,
+      frameId: "child",
+      ownerFrameBackendNodeId: 10,
+      tag: "canvas",
+      attrs: {},
+      rect: null,
+      localRect: { x: 12, y: 12, w: 224, h: 94 },
+      paintOrder: 1,
+      position: "static",
+      pointerEvents: "auto",
+    };
+    const captured: CapturedViewModel = {
+      nodes: [owner],
+      viewport: { width: 1000, height: 800 },
+      iframeNodes: new Map([[10, [canvas]]]),
+      frameNodes: new Map([
+        ["main", [owner]],
+        ["child", [canvas]],
+      ]),
+      frameOwnerBackendNodeIds: new Map([["child", 10]]),
+      frameParentIds: new Map([["child", "main"]]),
+      rootFrameId: "main",
+      excludedBackendNodeIds: new Set(),
+    };
+    const cdp: CdpRunner = {
+      send: vi.fn(async (_tabId, method) => {
+        if (method === "Accessibility.enable") return {};
+        if (method === "Accessibility.getFullAXTree") return { nodes: [] };
+        if (method === "Page.getLayoutMetrics") {
+          return { cssLayoutViewport: { clientWidth: 1000, clientHeight: 800 } };
+        }
+        if (method === "DOM.getBoxModel") {
+          return { model: { content: [52, 102, 352, 102, 352, 302, 52, 302] } };
+        }
+        if (method === "Page.createIsolatedWorld") return { executionContextId: 91 };
+        if (method === "Runtime.evaluate") {
+          return { result: { value: { width: 300, height: 200 } } };
+        }
+        throw new Error(`unexpected ${method}`);
+      }) as CdpRunner["send"],
+      getFrameGraph: vi.fn(async () => ({
+        rootFrameId: "main",
+        frames: [
+          { frameId: "main", target: { tabId: 4 } },
+          {
+            frameId: "child",
+            parentFrameId: "main",
+            ownerBackendNodeId: 10,
+            target: { tabId: 4 },
+          },
+        ],
+      })),
+    };
+
+    await captureFrameData(cdp, 4, captured);
+
+    expect(captured.frameNodes?.get("child")?.[0]?.rect).toEqual({
+      x: 64,
+      y: 114,
+      w: 224,
+      h: 94,
+    });
+    expect(captured.iframeNodes.get(10)?.[0]?.rect).toEqual({ x: 64, y: 114, w: 224, h: 94 });
+  });
+
   it("captures and positions multiple OOPIF documents missing from the root snapshot", async () => {
     const leftOwner = ownerNode(10, 50);
     const rightOwner = ownerNode(20, 500);
@@ -133,6 +232,87 @@ describe("captureFrameData", () => {
     );
   });
 
+  it("bounds concurrent same-target frame geometry resolution", async () => {
+    const owners = Array.from({ length: 8 }, (_, index) => ownerNode(10 + index, index * 40));
+    const childFrames = owners.map((owner, index) => {
+      const frameId = `child-${index}`;
+      const node: CapturedNode = {
+        backendNodeId: 100 + index,
+        parentBackendNodeId: null,
+        frameId,
+        ownerFrameBackendNodeId: owner.backendNodeId,
+        tag: "canvas",
+        attrs: {},
+        rect: null,
+        localRect: { x: 1, y: 2, w: 20, h: 10 },
+        paintOrder: 1,
+        position: "static",
+        pointerEvents: "auto",
+      };
+      return { frameId, owner, node };
+    });
+    const captured: CapturedViewModel = {
+      nodes: owners,
+      viewport: { width: 1000, height: 800 },
+      iframeNodes: new Map(childFrames.map(({ owner, node }) => [owner.backendNodeId, [node]])),
+      frameNodes: new Map([
+        ["main", owners],
+        ...childFrames.map(({ frameId, node }): [string, CapturedNode[]] => [frameId, [node]]),
+      ]),
+      frameOwnerBackendNodeIds: new Map(
+        childFrames.map(({ frameId, owner }) => [frameId, owner.backendNodeId]),
+      ),
+      rootFrameId: "main",
+      excludedBackendNodeIds: new Set(),
+    };
+    let activeWorlds = 0;
+    let maxActiveWorlds = 0;
+    const cdp: CdpRunner = {
+      send: vi.fn(async (_tabId, method, params) => {
+        if (method === "Page.getLayoutMetrics") {
+          return { cssLayoutViewport: { clientWidth: 1000, clientHeight: 800 } };
+        }
+        if (method === "DOM.getBoxModel") {
+          const backendNodeId = (params as { backendNodeId?: number }).backendNodeId ?? 10;
+          const x = (backendNodeId - 10) * 40;
+          return { model: { content: [x, 100, x + 30, 100, x + 30, 120, x, 120] } };
+        }
+        if (method === "Page.createIsolatedWorld") {
+          activeWorlds += 1;
+          maxActiveWorlds = Math.max(maxActiveWorlds, activeWorlds);
+          await new Promise((resolve) => setTimeout(resolve, 5));
+          activeWorlds -= 1;
+          return { executionContextId: 90 };
+        }
+        if (method === "Runtime.evaluate") {
+          return { result: { value: { width: 30, height: 20 } } };
+        }
+        if (method === "Accessibility.enable") return {};
+        if (method === "Accessibility.getFullAXTree") return { nodes: [] };
+        throw new Error(`unexpected ${method}`);
+      }) as CdpRunner["send"],
+      getFrameGraph: vi.fn(async () => ({
+        rootFrameId: "main",
+        frames: [
+          { frameId: "main", target: { tabId: 4 } },
+          ...childFrames.map(({ frameId, owner }) => ({
+            frameId,
+            parentFrameId: "main",
+            ownerBackendNodeId: owner.backendNodeId,
+            target: { tabId: 4 },
+          })),
+        ],
+      })),
+    };
+
+    await captureFrameData(cdp, 4, captured);
+
+    expect(maxActiveWorlds).toBe(4);
+    expect(childFrames.every(({ frameId }) => captured.frameNodes?.get(frameId)?.[0]?.rect)).toBe(
+      true,
+    );
+  });
+
   it("keeps OOPIF semantics when viewport projection is unavailable", async () => {
     const owner = { ...ownerNode(10, 50), rect: null, localRect: null };
     const captured: CapturedViewModel = {

apps/extension/src/tools/vom/__tests__/rendered-surfaces.test.ts

@@ -0,0 +1,386 @@
+import type { VomScene } from "@browser-skill/vom";
+import { describe, expect, it } from "vitest";
+import type { CapturedNode } from "../capture";
+import type { CapturedFrameDocument } from "../frame-capture";
+import {
+  clusterRenderedSurfaces,
+  discoverRenderedSurfaces,
+  projectRenderedSurfaces,
+} from "../rendered-surfaces";
+import type { SemanticAxNode } from "../semantic-graph";
+
+function domNode(backendNodeId: number, overrides: Partial<CapturedNode> = {}): CapturedNode {
+  return {
+    backendNodeId,
+    parentBackendNodeId: null,
+    frameId: "main",
+    ownerFrameBackendNodeId: null,
+    tag: "canvas",
+    attrs: {},
+    rect: { x: 10, y: 20, w: 800, h: 500 },
+    localRect: { x: 10, y: 20, w: 800, h: 500 },
+    paintOrder: 1,
+    position: "static",
+    pointerEvents: "auto",
+    rendered: true,
+    ...overrides,
+  };
+}
+
+function document(
+  domNodes: CapturedNode[],
+  axNodes: SemanticAxNode[] = [],
+): CapturedFrameDocument<SemanticAxNode> {
+  return {
+    frameId: domNodes[0]?.frameId ?? "main",
+    contextScopeId: domNodes[0]?.frameId ?? "main",
+    target: { tabId: 4 },
+    domNodes,
+    axNodes,
+  };
+}
+
+describe("rendered surface discovery", () => {
+  it("finds prominent rendered canvases and excludes hidden descendants", () => {
+    const visible = domNode(2, { parentBackendNodeId: 1 });
+    const hidden = domNode(3, { parentBackendNodeId: 4 });
+    const surfaces = discoverRenderedSurfaces(
+      [
+        document([
+          domNode(1, { tag: "main", rect: null, localRect: null }),
+          visible,
+          domNode(4, { tag: "section", attrs: { "aria-hidden": "true" }, rect: null }),
+          hidden,
+        ]),
+      ],
+      { width: 1280, height: 720 },
+    );
+
+    expect(surfaces).toHaveLength(1);
+    expect(surfaces[0]).toMatchObject({ backendNodeId: 2 });
+  });
+
+  it("keeps tiny canvases as individually addressable exact discoveries", () => {
+    const surfaces = discoverRenderedSurfaces(
+      [document([domNode(2, { rect: { x: 1, y: 1, w: 8, h: 8 } })])],
+      { width: 1280, height: 720 },
+    );
+    expect(surfaces).toEqual([expect.objectContaining({ backendNodeId: 2 })]);
+  });
+
+  it("keeps a surface when the canvas only has partial native accessibility content", () => {
+    const axNodes = [
+      {
+        nodeId: "canvas",
+        backendDOMNodeId: 2,
+        role: { type: "role", value: "presentation" },
+        childIds: ["button"],
+        frameId: "main",
+      },
+      {
+        nodeId: "button",
+        backendDOMNodeId: 3,
+        role: { type: "role", value: "button" },
+        frameId: "main",
+      },
+    ] as SemanticAxNode[];
+    expect(
+      discoverRenderedSurfaces([document([domNode(2)], axNodes)], { width: 1280, height: 720 }),
+    ).toEqual([expect.objectContaining({ backendNodeId: 2 })]);
+  });
+
+  it("discovers unnamed canvases without guessing labels from engineering attributes", () => {
+    const canvas = domNode(2, {
+      attrs: {
+        id: "must-not-be-a-label",
+        class: "must-not-be-a-label",
+        "data-testid": "must-not-be-a-label",
+      },
+    });
+
+    expect(discoverRenderedSurfaces([document([canvas])], { width: 1280, height: 720 })).toEqual([
+      expect.not.objectContaining({
+        label: expect.anything(),
+      }),
+    ]);
+  });
+
+  it("treats blank explicit Canvas names as unnamed", () => {
+    expect(
+      discoverRenderedSurfaces(
+        [document([domNode(2, { attrs: { "aria-label": "   ", title: "\n\t" } })])],
+        { width: 1280, height: 720 },
+      ),
+    ).toEqual([expect.not.objectContaining({ label: expect.anything() })]);
+  });
+
+  it("excludes canvases inside BrowserSkill-owned DOM subtrees", () => {
+    const owner = domNode(1, { tag: "div", rect: null, localRect: null });
+    const canvas = domNode(2, { parentBackendNodeId: 1 });
+
+    expect(
+      discoverRenderedSurfaces(
+        [document([owner, canvas])],
+        { width: 1280, height: 720 },
+        new Set([1]),
+      ),
+    ).toEqual([]);
+  });
+
+  it("excludes canvases suppressed by a fully transparent ancestor", () => {
+    const owner = domNode(1, {
+      tag: "div",
+      rect: { x: 0, y: 0, w: 900, h: 600 },
+      visuallySuppressed: true,
+    });
+    const canvas = domNode(2, { parentBackendNodeId: 1 });
+
+    expect(
+      discoverRenderedSurfaces([document([owner, canvas])], { width: 1280, height: 720 }),
+    ).toEqual([]);
+  });
+
+  it("clips canvases to overflow ancestors and drops fully clipped canvases", () => {
+    const owner = domNode(1, {
+      tag: "div",
+      rect: { x: 10, y: 20, w: 100, h: 70 },
+      overflowX: "hidden",
+      overflowY: "hidden",
+    });
+    const partial = domNode(2, {
+      parentBackendNodeId: 1,
+      rect: { x: 55, y: 30, w: 140, h: 60 },
+    });
+    const fullyClipped = domNode(3, {
+      parentBackendNodeId: 1,
+      rect: { x: 140, y: 30, w: 80, h: 50 },
+    });
+
+    expect(
+      discoverRenderedSurfaces([document([owner, partial, fullyClipped])], {
+        width: 1280,
+        height: 720,
+      }),
+    ).toEqual([
+      expect.objectContaining({
+        backendNodeId: 2,
+        visibleRect: { x: 55, y: 30, w: 55, h: 60 },
+      }),
+    ]);
+  });
+});
+
+describe("rendered surface clustering and projection", () => {
+  it("collapses overlapping canvas layers and attaches them to the nearest retained parent", () => {
+    const body = domNode(1, { tag: "body", rect: null, localRect: null });
+    const surfaces = discoverRenderedSurfaces(
+      [
+        document([
+          body,
+          domNode(2, { parentBackendNodeId: 1, paintOrder: 1 }),
+          domNode(3, { parentBackendNodeId: 1, paintOrder: 2 }),
+        ]),
+      ],
+      { width: 1280, height: 720 },
+    );
+    const clustered = clusterRenderedSurfaces(surfaces);
+    const scene: VomScene = {
+      viewport: { width: 1280, height: 720 },
+      nodes: [
+        {
+          id: 10,
+          parentId: null,
+          backendNodeId: 1,
+          frameId: "main",
+          tag: "body",
+          role: "RootWebArea",
+          rect: null,
+          paintOrder: 0,
+          position: "static",
+          pointerEvents: "auto",
+        },
+      ],
+    };
+
+    expect(clustered.groups).toHaveLength(1);
+    expect(clustered.groups[0]?.memberCount).toBe(2);
+    expect(clustered.groups[0]?.label).toBeUndefined();
+    expect(
+      projectRenderedSurfaces(
+        scene,
+        [document([body, ...surfaces.map((s) => domNode(s.backendNodeId))])],
+        clustered,
+      ),
+    ).toMatchObject({
+      visualSurfaces: [expect.objectContaining({ parentId: 10, backendNodeId: 3, memberCount: 2 })],
+    });
+  });
+
+  it("does not merge a contained canvas with a larger independent canvas", () => {
+    const clustered = clusterRenderedSurfaces([
+      ...discoverRenderedSurfaces(
+        [
+          document([
+            domNode(1, { tag: "main", rect: null, localRect: null }),
+            domNode(2, { parentBackendNodeId: 1 }),
+            domNode(3, {
+              parentBackendNodeId: 1,
+              rect: { x: 30, y: 40, w: 200, h: 100 },
+            }),
+          ]),
+        ],
+        { width: 1280, height: 720 },
+      ),
+    ]);
+
+    expect(clustered.groups).toHaveLength(2);
+  });
+
+  it("keeps the existing inclusive 90 percent stack threshold", () => {
+    const clustered = clusterRenderedSurfaces([
+      {
+        renderingKind: "canvas",
+        frameId: "main",
+        backendNodeId: 2,
+        parentBackendNodeId: 1,
+        visibleRect: { x: 0, y: 0, w: 100, h: 100 },
+        paintOrder: 1,
+      },
+      {
+        renderingKind: "canvas",
+        frameId: "main",
+        backendNodeId: 3,
+        parentBackendNodeId: 1,
+        visibleRect: { x: 5, y: 0, w: 90, h: 100 },
+        paintOrder: 2,
+      },
+    ]);
+
+    expect(clustered.groups).toEqual([expect.objectContaining({ memberCount: 2 })]);
+  });
+
+  it("does not merge coincident canvases owned by different containers", () => {
+    const clustered = clusterRenderedSurfaces(
+      discoverRenderedSurfaces(
+        [
+          document([
+            domNode(1, { tag: "section", rect: null, localRect: null }),
+            domNode(2, { tag: "section", rect: null, localRect: null }),
+            domNode(3, { parentBackendNodeId: 1 }),
+            domNode(4, { parentBackendNodeId: 2 }),
+          ]),
+        ],
+        { width: 1280, height: 720 },
+      ),
+    );
+
+    expect(clustered.groups).toHaveLength(2);
+  });
+
+  it("keeps adjacent map tiles and twenty sparklines as independent surfaces", () => {
+    const surfaces = Array.from({ length: 20 }, (_, index) =>
+      domNode(100 + index, {
+        parentBackendNodeId: 1,
+        rect: { x: index * 12, y: 20, w: 10, h: 8 },
+      }),
+    );
+    const clustered = clusterRenderedSurfaces(
+      discoverRenderedSurfaces(
+        [document([domNode(1, { tag: "main", rect: null, localRect: null }), ...surfaces])],
+        { width: 1280, height: 720 },
+      ),
+    );
+
+    expect(clustered.groups).toHaveLength(20);
+    expect(clustered.groups.every((group) => group.memberCount === 1)).toBe(true);
+    expect(clustered.truncated).toBe(false);
+  });
+
+  it("clusters a large coincident stack without retaining every member", () => {
+    const surfaces = Array.from({ length: 10_000 }, (_, index) => ({
+      renderingKind: "canvas" as const,
+      frameId: "main",
+      backendNodeId: index + 1,
+      parentBackendNodeId: 1,
+      visibleRect: { x: 10, y: 20, w: 800, h: 500 },
+      paintOrder: index,
+    }));
+
+    const clustered = clusterRenderedSurfaces(surfaces);
+
+    expect(clustered.groups).toEqual([
+      expect.objectContaining({ memberCount: 10_000, representative: surfaces[9_999] }),
+    ]);
+    expect(clustered.truncated).toBe(false);
+  });
+
+  it("keeps large non-overlapping surface sets linear and independent", () => {
+    const surfaces = Array.from({ length: 10_000 }, (_, index) => ({
+      renderingKind: "canvas" as const,
+      frameId: "main",
+      backendNodeId: index + 1,
+      parentBackendNodeId: 1,
+      visibleRect: { x: index * 20, y: 0, w: 10, h: 10 },
+      paintOrder: index,
+    }));
+
+    const clustered = clusterRenderedSurfaces(surfaces);
+
+    expect(clustered.groups).toHaveLength(10_000);
+    expect(clustered.groups.every((group) => group.memberCount === 1)).toBe(true);
+    expect(clustered.truncated).toBe(false);
+  });
+
+  it("keeps the highest-priority surfaces when the computation guard truncates input", () => {
+    const surfaces = [
+      {
+        renderingKind: "canvas" as const,
+        frameId: "main",
+        backendNodeId: 1,
+        parentBackendNodeId: 10,
+        visibleRect: { x: 0, y: 0, w: 10, h: 10 },
+        paintOrder: 1,
+      },
+      {
+        renderingKind: "canvas" as const,
+        frameId: "main",
+        backendNodeId: 2,
+        parentBackendNodeId: 10,
+        visibleRect: { x: 20, y: 0, w: 100, h: 100 },
+        paintOrder: 1,
+      },
+      {
+        renderingKind: "canvas" as const,
+        frameId: "main",
+        backendNodeId: 3,
+        parentBackendNodeId: 10,
+        visibleRect: { x: 140, y: 0, w: 8, h: 8 },
+        paintOrder: 1,
+        label: "Important status",
+      },
+    ];
+
+    const clustered = clusterRenderedSurfaces(surfaces, { maxSurfaces: 2 });
+
+    expect(clustered.groups.map((group) => group.representative.backendNodeId).sort()).toEqual([
+      2, 3,
+    ]);
+    expect(clustered).toMatchObject({ truncated: true, omittedCount: 1 });
+  });
+
+  it("stops at the comparison budget without dropping already completed groups", () => {
+    const surfaces = Array.from({ length: 3 }, (_, index) => ({
+      renderingKind: "canvas" as const,
+      frameId: "main",
+      backendNodeId: index + 1,
+      parentBackendNodeId: 10,
+      visibleRect: { x: 0, y: 0, w: 100, h: 100 },
+      paintOrder: index + 1,
+    }));
+
+    const clustered = clusterRenderedSurfaces(surfaces, { maxComparisons: 1 });
+
+    expect(clustered.groups).toEqual([expect.objectContaining({ memberCount: 2 })]);
+    expect(clustered).toMatchObject({ truncated: true, omittedCount: 1 });
+  });
+});

apps/extension/src/tools/vom/capture.ts

@@ -7,11 +7,19 @@
 import type { Rect, Viewport } from "@browser-skill/vom";
 import { evaluateHoverTrigger } from "@/lib/hover-trigger-policy";
 import { isOverlayHostNode, OVERLAY_HOST_SELECTOR } from "../../lib/overlay-bridge";
-import { childFrameProjection, type GeometryProjection, projectRectToViewport } from "../geometry";
+import { type GeometryProjection, projectRectToViewport } from "../geometry";
 import type { CdpRunner } from "../shared";
 import { clearHover, ProbeBudget, waitForHover } from "./hover-perception";
 
-const REQUESTED_STYLES = ["position", "pointer-events", "cursor", "visibility", "opacity"] as const;
+const REQUESTED_STYLES = [
+  "position",
+  "pointer-events",
+  "cursor",
+  "visibility",
+  "opacity",
+  "overflow-x",
+  "overflow-y",
+] as const;
 const STYLE_COL = Object.fromEntries(
   REQUESTED_STYLES.map((name, index) => [name, index]),
 ) as Record<(typeof REQUESTED_STYLES)[number], number>;
@@ -39,6 +47,10 @@ export interface CapturedNode {
    * `textContent`: the live parser always sets it, hand-built fixtures may not.
    */
   cursor?: string;
+  overflowX?: string;
+  overflowY?: string;
+  /** True when this node or one of its DOM ancestors makes the subtree fully transparent/hidden. */
+  visuallySuppressed?: boolean;
   /**
    * Whether the live DOM snapshot provides a painted, non-hidden box for this
    * node. Semantic resolution uses this only for DOM fallback nodes; AX-backed
@@ -849,6 +861,8 @@ function parseDocumentNodes(
   const cursorCol = STYLE_COL.cursor;
   const visibilityCol = STYLE_COL.visibility;
   const opacityCol = STYLE_COL.opacity;
+  const overflowXCol = STYLE_COL["overflow-x"];
+  const overflowYCol = STYLE_COL["overflow-y"];
   const inputValues = sparseIndexMap(dn.inputValue);
   const textValues = sparseIndexMap(dn.textValue);
   const checkedInputs = new Set(dn.inputChecked?.index ?? []);
@@ -874,6 +888,7 @@ function parseDocumentNodes(
 
   // Build CapturedNodes.
   const nodes: CapturedNode[] = [];
+  const visuallySuppressedByNodeIndex = new Array<boolean>(count).fill(false);
   for (let n = 0; n < count; n++) {
     if (excluded[n]) continue;
     const backendNodeId = dn.backendNodeId[n];
@@ -893,7 +908,11 @@ function parseDocumentNodes(
     let position = "static";
     let pointerEvents = "auto";
     let cursor = "auto";
+    let overflowX = "visible";
+    let overflowY = "visible";
     let rendered = false;
+    let visuallySuppressed =
+      parentIdx >= 0 ? visuallySuppressedByNodeIndex[parentIdx] === true : false;
     const li = layoutByNode.get(n);
     if (li !== undefined) {
       const b = dl?.bounds?.[li];
@@ -916,12 +935,19 @@ function parseDocumentNodes(
       cursor = str(strings, styleRow[cursorCol]) || "auto";
       const visibility = str(strings, styleRow[visibilityCol]) || "visible";
       const opacity = str(strings, styleRow[opacityCol]) || "1";
+      overflowX = str(strings, styleRow[overflowXCol]) || "visible";
+      overflowY = str(strings, styleRow[overflowYCol]) || "visible";
+      visuallySuppressed ||=
+        visibility === "hidden" ||
+        visibility === "collapse" ||
+        (Number.parseFloat(opacity) || 0) <= 0;
       rendered =
         localRect !== null &&
         visibility !== "hidden" &&
         visibility !== "collapse" &&
         (Number.parseFloat(opacity) || 0) > 0;
     }
+    visuallySuppressedByNodeIndex[n] = visuallySuppressed;
 
     // Skip non-element nodes (#text, #cdata-section, etc.) — they carry no
     // geometry and are never queried by callers.
@@ -956,6 +982,9 @@ function parseDocumentNodes(
       position,
       pointerEvents,
       cursor,
+      overflowX,
+      overflowY,
+      visuallySuppressed,
       rendered,
       textContent,
       ...(formValue !== undefined ? { formValue } : {}),
@@ -985,7 +1014,6 @@ function parseChildFrameDocuments(
   strings: string[],
   dpr: number,
   parentDocIndex: number,
-  parentNodes: CapturedNode[],
   parentContext: FrameContext,
   visited = new Set<number>(),
 ): ParseFrameDocumentsResult {
@@ -1000,19 +1028,17 @@ function parseChildFrameDocuments(
   }
 
   const parentBackendIds = parentDoc?.nodes?.backendNodeId ?? [];
-  const parentNodeByBackendId = new Map(parentNodes.map((node) => [node.backendNodeId, node]));
-
   for (const [nodeArrayIdx, childDocIndex] of cdi) {
     if (visited.has(childDocIndex)) continue;
     const childDoc = documents[childDocIndex];
     if (!childDoc) continue;
     const iframeBackendId = parentBackendIds[nodeArrayIdx];
     if (iframeBackendId === undefined) continue;
-    const iframeNode = parentNodeByBackendId.get(iframeBackendId);
-    const projection =
-      parentContext.projection && iframeNode?.localRect
-        ? childFrameProjection(parentContext.projection, iframeNode.localRect)
-        : null;
+    // DOMSnapshot reports the iframe owner's border-box bounds, but child
+    // document coordinates start at the content-box origin. Preserve the
+    // frame-local rectangles here and let frame-capture normalize them with
+    // authoritative content quads once the frame graph is available.
+    const projection = null;
 
     const childContext: FrameContext = {
       frameId: snapshotFrameId(childDoc, strings),
@@ -1036,7 +1062,6 @@ function parseChildFrameDocuments(
       strings,
       dpr,
       childDocIndex,
-      parsed.nodes,
       childContext,
       nextVisited,
     );
@@ -1114,7 +1139,7 @@ export async function captureViewModel(
   const nodes = mainParsed.nodes;
   const excludedBackendNodeIds = new Set(mainParsed.excludedBackendNodeIds);
 
-  const frameParsed = parseChildFrameDocuments(documents, strings, dpr, 0, nodes, topContext);
+  const frameParsed = parseChildFrameDocuments(documents, strings, dpr, 0, topContext);
   const iframeNodes = frameParsed.iframeNodes;
   for (const id of frameParsed.excludedBackendNodeIds) {
     excludedBackendNodeIds.add(id);

apps/extension/src/tools/vom/frame-capture.ts

@@ -1,6 +1,5 @@
 import type { CdpFrame, CdpFrameGraph } from "@/browser-driver/frame-graph";
-import { resolveFrameProjection } from "../frame-geometry";
-import { type GeometryProjection, projectRectToViewport } from "../geometry";
+import { createFrameGeometryContext } from "../frame-geometry";
 import { type CdpRunner, cdpRunnerForTarget, sendToCdpTarget } from "../shared";
 import { type CapturedNode, type CapturedViewModel, captureViewModel } from "./capture";
 import {
@@ -13,6 +12,8 @@ import {
 export type FrameAxNode = FrameOwnedAxNode;
 export type CapturedFrameDocument<T extends FrameAxNode> = FrameDocument<T>;
 
+const FRAME_GEOMETRY_CONCURRENCY = 4;
+
 function throwIfAborted(signal?: AbortSignal): void {
   if (signal?.aborted) throw new DOMException("observation aborted", "AbortError");
 }
@@ -21,6 +22,22 @@ function isAbortError(err: unknown): boolean {
   return err instanceof Error && err.name === "AbortError";
 }
 
+async function forEachWithConcurrency<T>(
+  items: readonly T[],
+  concurrency: number,
+  worker: (item: T) => Promise<void>,
+): Promise<void> {
+  let next = 0;
+  const run = async () => {
+    while (next < items.length) {
+      const index = next;
+      next += 1;
+      await worker(items[index]);
+    }
+  };
+  await Promise.all(Array.from({ length: Math.min(concurrency, items.length) }, run));
+}
+
 async function discoverFrameGraph(cdp: CdpRunner, tabId: number): Promise<CdpFrameGraph | null> {
   if (!cdp.getFrameGraph) return null;
   try {
@@ -31,21 +48,6 @@ async function discoverFrameGraph(cdp: CdpRunner, tabId: number): Promise<CdpFra
   }
 }
 
-function transformFrameNodes(
-  nodes: CapturedNode[],
-  projection: GeometryProjection | null,
-): CapturedNode[] {
-  return nodes.map((node) => ({
-    ...node,
-    rect: projection
-      ? (() => {
-          const bounds = projectRectToViewport(node.rect, projection);
-          return bounds ? { x: bounds.x, y: bounds.y, w: bounds.width, h: bounds.height } : null;
-        })()
-      : null,
-  }));
-}
-
 async function captureMissingFrameDocuments(
   cdp: CdpRunner,
   tabId: number,
@@ -63,16 +65,6 @@ async function captureMissingFrameDocuments(
       const child = await captureViewModel(cdpRunnerForTarget(cdp, frame.target), tabId, {
         signal,
       });
-      let projection: GeometryProjection | null = null;
-      try {
-        projection = await resolveFrameProjection(cdp, graph, frame.frameId);
-      } catch (err) {
-        if (isAbortError(err)) throw err;
-        console.debug("[bsk observation] child frame geometry projection failed", {
-          frameId: frame.frameId,
-          err,
-        });
-      }
       const rootFrameId = child.rootFrameId ?? frame.frameId;
       const childFrames = child.frameNodes ?? new Map<string, CapturedNode[]>();
       if (!childFrames.has(rootFrameId)) childFrames.set(rootFrameId, child.nodes);
@@ -83,7 +75,7 @@ async function captureMissingFrameDocuments(
         );
       }
       for (const [childFrameId, nodes] of childFrames) {
-        captured.frameNodes.set(childFrameId, transformFrameNodes(nodes, projection));
+        captured.frameNodes.set(childFrameId, nodes);
       }
       if (frame.ownerBackendNodeId !== undefined) {
         captured.iframeNodes.set(
@@ -109,6 +101,60 @@ async function captureMissingFrameDocuments(
   }
 }
 
+async function normalizeFrameLocalRects(
+  cdp: CdpRunner,
+  graph: CdpFrameGraph,
+  captured: CapturedViewModel,
+  signal?: AbortSignal,
+): Promise<void> {
+  if (!captured.frameNodes) return;
+  const geometry = createFrameGeometryContext(cdp, graph);
+  const frameById = new Map(graph.frames.map((frame) => [frame.frameId, frame]));
+  await forEachWithConcurrency(
+    [...captured.frameNodes],
+    FRAME_GEOMETRY_CONCURRENCY,
+    async ([frameId, nodes]) => {
+      if (frameId === graph.rootFrameId || !frameById.has(frameId)) return;
+      try {
+        const normalized: CapturedNode[] = [];
+        for (const node of nodes) {
+          if (!node.localRect) {
+            normalized.push({ ...node, rect: null });
+          } else {
+            const bounds = await geometry.projectFrameLocalRect(frameId, node.localRect);
+            normalized.push({
+              ...node,
+              rect: bounds ? { x: bounds.x, y: bounds.y, w: bounds.width, h: bounds.height } : null,
+            });
+          }
+        }
+        throwIfAborted(signal);
+        captured.frameNodes?.set(frameId, normalized);
+        const ownerBackendNodeId =
+          frameById.get(frameId)?.ownerBackendNodeId ??
+          captured.frameOwnerBackendNodeIds?.get(frameId);
+        if (ownerBackendNodeId !== undefined) {
+          captured.iframeNodes.set(ownerBackendNodeId, normalized);
+        }
+      } catch (err) {
+        if (isAbortError(err)) throw err;
+        console.debug("[bsk observation] frame-local geometry normalization failed", {
+          frameId,
+          err,
+        });
+        const unavailable = nodes.map((node) => ({ ...node, rect: null }));
+        captured.frameNodes?.set(frameId, unavailable);
+        const ownerBackendNodeId =
+          frameById.get(frameId)?.ownerBackendNodeId ??
+          captured.frameOwnerBackendNodeIds?.get(frameId);
+        if (ownerBackendNodeId !== undefined) {
+          captured.iframeNodes.set(ownerBackendNodeId, unavailable);
+        }
+      }
+    },
+  );
+}
+
 async function captureAxTrees<T extends FrameAxNode>(
   cdp: CdpRunner,
   tabId: number,
@@ -174,7 +220,10 @@ export async function captureFrameData<T extends FrameAxNode>(
 ): Promise<CapturedFrameDocument<T>[]> {
   const graph = await discoverFrameGraph(cdp, tabId);
   const frames = graph?.frames ?? [];
-  if (graph) await captureMissingFrameDocuments(cdp, tabId, graph, captured, signal);
+  if (graph) {
+    await captureMissingFrameDocuments(cdp, tabId, graph, captured, signal);
+    await normalizeFrameLocalRects(cdp, graph, captured, signal);
+  }
   throwIfAborted(signal);
   const batches = await captureAxTrees<T>(cdp, tabId, frames, captured, signal);
   return buildFrameDocuments(graph, batches, captured);

apps/extension/src/tools/vom/rendered-surfaces/cluster.ts

@@ -0,0 +1,207 @@
+import { type Rect, selectHighestPriorityVisualSurfaces } from "@browser-skill/vom";
+import type { ClusteredRenderedSurfaces, RenderedSurface, RenderedSurfaceGroup } from "./types";
+
+const STACK_IOU_RATIO = 0.9;
+const STACK_SIZE_RATIO = 0.9;
+const SIZE_BAND_BASE = 1 / STACK_SIZE_RATIO;
+const POSITION_CELL_RATIO = 0.1;
+const MAX_CLUSTER_COMPARISONS = 1_000_000;
+const MAX_CLUSTER_SURFACES = 50_000;
+
+export interface ClusterRenderedSurfaceOptions {
+  maxComparisons?: number;
+  maxSurfaces?: number;
+}
+
+interface IndexedGroup extends RenderedSurfaceGroup {
+  creationIndex: number;
+  indexKey: string;
+}
+
+function area(rect: Rect): number {
+  return Math.max(0, rect.w) * Math.max(0, rect.h);
+}
+
+function intersectionArea(a: Rect, b: Rect): number {
+  const width = Math.max(0, Math.min(a.x + a.w, b.x + b.w) - Math.max(a.x, b.x));
+  const height = Math.max(0, Math.min(a.y + a.h, b.y + b.h) - Math.max(a.y, b.y));
+  return width * height;
+}
+
+function stackScore(a: RenderedSurface, b: RenderedSurface): number | null {
+  const aArea = area(a.visibleRect);
+  const bArea = area(b.visibleRect);
+  const intersection = intersectionArea(a.visibleRect, b.visibleRect);
+  const union = aArea + bArea - intersection;
+  const widthRatio =
+    Math.min(a.visibleRect.w, b.visibleRect.w) / Math.max(a.visibleRect.w, b.visibleRect.w);
+  const heightRatio =
+    Math.min(a.visibleRect.h, b.visibleRect.h) / Math.max(a.visibleRect.h, b.visibleRect.h);
+  const iou = union > 0 ? intersection / union : 0;
+  return iou >= STACK_IOU_RATIO && widthRatio >= STACK_SIZE_RATIO && heightRatio >= STACK_SIZE_RATIO
+    ? iou
+    : null;
+}
+
+function sizeBand(value: number): number {
+  return Math.floor(Math.log(Math.max(Number.EPSILON, value)) / Math.log(SIZE_BAND_BASE));
+}
+
+function bandSize(band: number): number {
+  return SIZE_BAND_BASE ** band;
+}
+
+function spatialKey(rect: Rect): string {
+  const widthBand = sizeBand(rect.w);
+  const heightBand = sizeBand(rect.h);
+  const cellWidth = bandSize(widthBand) * POSITION_CELL_RATIO;
+  const cellHeight = bandSize(heightBand) * POSITION_CELL_RATIO;
+  const centerX = rect.x + rect.w / 2;
+  const centerY = rect.y + rect.h / 2;
+  return `${widthBand}:${heightBand}:${Math.floor(centerX / cellWidth)}:${Math.floor(centerY / cellHeight)}`;
+}
+
+function candidateKeys(rect: Rect): string[] {
+  const keys: string[] = [];
+  const centerX = rect.x + rect.w / 2;
+  const centerY = rect.y + rect.h / 2;
+  const widthBand = sizeBand(rect.w);
+  const heightBand = sizeBand(rect.h);
+  // A >=0.9 size ratio crosses at most one logarithmic band. IoU >=0.9
+  // likewise keeps centers within one 10%-of-size position cell.
+  for (let dw = -1; dw <= 1; dw += 1) {
+    for (let dh = -1; dh <= 1; dh += 1) {
+      const candidateWidthBand = widthBand + dw;
+      const candidateHeightBand = heightBand + dh;
+      const cellWidth = bandSize(candidateWidthBand) * POSITION_CELL_RATIO;
+      const cellHeight = bandSize(candidateHeightBand) * POSITION_CELL_RATIO;
+      const cellX = Math.floor(centerX / cellWidth);
+      const cellY = Math.floor(centerY / cellHeight);
+      for (let dx = -1; dx <= 1; dx += 1) {
+        for (let dy = -1; dy <= 1; dy += 1) {
+          keys.push(`${candidateWidthBand}:${candidateHeightBand}:${cellX + dx}:${cellY + dy}`);
+        }
+      }
+    }
+  }
+  return keys;
+}
+
+function preferredRepresentative(a: RenderedSurface, b: RenderedSurface): RenderedSurface {
+  const aArea = area(a.visibleRect);
+  const bArea = area(b.visibleRect);
+  return bArea > aArea || (bArea === aArea && b.paintOrder > a.paintOrder) ? b : a;
+}
+
+function partitionKey(surface: RenderedSurface): string {
+  return `${surface.frameId}\u0000${surface.parentBackendNodeId ?? "root"}`;
+}
+
+function normalizedLimit(value: number | undefined, fallback: number): number {
+  const resolved = value ?? fallback;
+  return Number.isFinite(resolved) ? Math.max(0, Math.floor(resolved)) : fallback;
+}
+
+export function clusterRenderedSurfaces(
+  surfaces: RenderedSurface[],
+  options: ClusterRenderedSurfaceOptions = {},
+): ClusteredRenderedSurfaces {
+  const maxComparisons = normalizedLimit(options.maxComparisons, MAX_CLUSTER_COMPARISONS);
+  const maxSurfaces = normalizedLimit(options.maxSurfaces, MAX_CLUSTER_SURFACES);
+  const prioritized = selectHighestPriorityVisualSurfaces(surfaces, maxSurfaces);
+  const partitions = new Map<string, Map<string, Set<IndexedGroup>>>();
+  const groups: IndexedGroup[] = [];
+  let comparisons = 0;
+  let processed = 0;
+  let truncated = prioritized.length < surfaces.length;
+  let comparisonLimitReached = false;
+
+  for (const surface of prioritized) {
+    if (comparisons >= maxComparisons) {
+      truncated = true;
+      break;
+    }
+    const key = partitionKey(surface);
+    let index = partitions.get(key);
+    if (!index) {
+      index = new Map();
+      partitions.set(key, index);
+    }
+    const candidates = new Set<IndexedGroup>();
+    for (const candidateKey of candidateKeys(surface.visibleRect)) {
+      for (const candidate of index.get(candidateKey) ?? []) candidates.add(candidate);
+    }
+
+    let best: { group: IndexedGroup; score: number; areaDelta: number } | null = null;
+    for (const group of candidates) {
+      if (comparisons >= maxComparisons) {
+        truncated = true;
+        comparisonLimitReached = true;
+        break;
+      }
+      comparisons += 1;
+      const score = stackScore(group.representative, surface);
+      if (score === null) continue;
+      const areaDelta = Math.abs(
+        area(group.representative.visibleRect) - area(surface.visibleRect),
+      );
+      if (
+        !best ||
+        score > best.score ||
+        (score === best.score && areaDelta < best.areaDelta) ||
+        (score === best.score &&
+          areaDelta === best.areaDelta &&
+          group.creationIndex < best.group.creationIndex)
+      ) {
+        best = { group, score, areaDelta };
+      }
+    }
+    if (comparisonLimitReached) break;
+
+    if (best) {
+      const group = best.group;
+      const representative = preferredRepresentative(group.representative, surface);
+      group.memberCount += 1;
+      group.label ??= surface.label;
+      if (representative !== group.representative) {
+        index.get(group.indexKey)?.delete(group);
+        group.representative = representative;
+        group.indexKey = spatialKey(representative.visibleRect);
+        const indexed = index.get(group.indexKey) ?? new Set<IndexedGroup>();
+        indexed.add(group);
+        index.set(group.indexKey, indexed);
+      }
+      processed += 1;
+      continue;
+    }
+
+    const indexKey = spatialKey(surface.visibleRect);
+    const group: IndexedGroup = {
+      frameId: surface.frameId,
+      parentBackendNodeId: surface.parentBackendNodeId,
+      representative: surface,
+      ...(surface.label ? { label: surface.label } : {}),
+      memberCount: 1,
+      creationIndex: groups.length,
+      indexKey,
+    };
+    groups.push(group);
+    const indexed = index.get(indexKey) ?? new Set<IndexedGroup>();
+    indexed.add(group);
+    index.set(indexKey, indexed);
+    processed += 1;
+  }
+
+  return {
+    groups: groups
+      .map(({ creationIndex: _creationIndex, indexKey: _indexKey, ...group }) => group)
+      .sort(
+        (a, b) =>
+          a.frameId.localeCompare(b.frameId) ||
+          a.representative.visibleRect.y - b.representative.visibleRect.y ||
+          a.representative.visibleRect.x - b.representative.visibleRect.x,
+      ),
+    truncated,
+    ...(truncated ? { omittedCount: surfaces.length - processed } : {}),
+  };
+}

apps/extension/src/tools/vom/rendered-surfaces/discovery.ts

@@ -0,0 +1,113 @@
+import type { Rect, Viewport } from "@browser-skill/vom";
+import type { CapturedNode } from "../capture";
+import type { CapturedFrameDocument } from "../frame-capture";
+import type { SemanticAxNode } from "../semantic-graph";
+import type { RenderedSurface } from "./types";
+
+function clean(value: string | undefined): string | undefined {
+  const normalized = value?.replace(/\s+/g, " ").trim();
+  return normalized || undefined;
+}
+
+function intersectViewport(rect: Rect, viewport: Viewport): Rect | null {
+  return intersectRect(rect, { x: 0, y: 0, w: viewport.width, h: viewport.height });
+}
+
+function intersectRect(a: Rect, b: Rect): Rect | null {
+  const x = Math.max(a.x, b.x);
+  const y = Math.max(a.y, b.y);
+  const right = Math.min(a.x + a.w, b.x + b.w);
+  const bottom = Math.min(a.y + a.h, b.y + b.h);
+  if (right <= x || bottom <= y) return null;
+  return { x, y, w: right - x, h: bottom - y };
+}
+
+const CLIPPING_OVERFLOW = new Set(["auto", "clip", "hidden", "scroll"]);
+
+function clipToOverflowAncestors(
+  node: CapturedNode,
+  rect: Rect,
+  nodesByBackend: ReadonlyMap<number, CapturedNode>,
+): Rect | null {
+  let clipped: Rect | null = rect;
+  let parentBackendNodeId = node.parentBackendNodeId;
+  let guard = 0;
+  while (clipped && parentBackendNodeId !== null && guard <= nodesByBackend.size) {
+    const parent = nodesByBackend.get(parentBackendNodeId);
+    if (!parent) break;
+    if (parent.rect) {
+      const clipX = CLIPPING_OVERFLOW.has(parent.overflowX ?? "visible");
+      const clipY = CLIPPING_OVERFLOW.has(parent.overflowY ?? "visible");
+      if (clipX || clipY) {
+        const clipRect = {
+          x: clipX ? parent.rect.x : clipped.x,
+          y: clipY ? parent.rect.y : clipped.y,
+          w: clipX ? parent.rect.w : clipped.w,
+          h: clipY ? parent.rect.h : clipped.h,
+        };
+        clipped = intersectRect(clipped, clipRect);
+      }
+    }
+    parentBackendNodeId = parent.parentBackendNodeId;
+    guard += 1;
+  }
+  return clipped;
+}
+
+function excludedByDomPolicy(
+  node: CapturedNode,
+  nodesByBackend: Map<number, CapturedNode>,
+  excludedBackendNodeIds: ReadonlySet<number>,
+): boolean {
+  let current: CapturedNode | undefined = node;
+  let guard = 0;
+  while (current && guard <= nodesByBackend.size) {
+    if (excludedBackendNodeIds.has(current.backendNodeId)) return true;
+    if (current.visuallySuppressed === true) return true;
+    const attrs = current.attrs;
+    if (
+      Object.prototype.hasOwnProperty.call(attrs, "hidden") ||
+      Object.prototype.hasOwnProperty.call(attrs, "inert") ||
+      (attrs["aria-hidden"] ?? "").toLowerCase() === "true"
+    ) {
+      return true;
+    }
+    current =
+      current.parentBackendNodeId === null
+        ? undefined
+        : nodesByBackend.get(current.parentBackendNodeId);
+    guard += 1;
+  }
+  return false;
+}
+
+export function discoverRenderedSurfaces(
+  documents: CapturedFrameDocument<SemanticAxNode>[],
+  viewport: Viewport,
+  excludedBackendNodeIds: ReadonlySet<number> = new Set(),
+): RenderedSurface[] {
+  const surfaces: RenderedSurface[] = [];
+  for (const document of documents) {
+    const nodesByBackend = new Map(document.domNodes.map((node) => [node.backendNodeId, node]));
+    for (const node of document.domNodes) {
+      if (node.tag.toLowerCase() !== "canvas" || node.rendered !== true || !node.rect) continue;
+      if (excludedByDomPolicy(node, nodesByBackend, excludedBackendNodeIds)) continue;
+      const viewportRect = intersectViewport(node.rect, viewport);
+      const visibleRect = viewportRect
+        ? clipToOverflowAncestors(node, viewportRect, nodesByBackend)
+        : null;
+      if (!visibleRect) continue;
+      const label = clean(node.attrs["aria-label"] ?? node.attrs.title);
+      surfaces.push({
+        renderingKind: "canvas",
+        frameId: document.frameId,
+        backendNodeId: node.backendNodeId,
+        parentBackendNodeId: node.parentBackendNodeId,
+        visibleRect,
+        paintOrder: node.paintOrder,
+        ...(label ? { label } : {}),
+      });
+    }
+  }
+  return surfaces;
+}

apps/extension/src/tools/vom/rendered-surfaces/index.ts

@@ -0,0 +1,8 @@
+export { type ClusterRenderedSurfaceOptions, clusterRenderedSurfaces } from "./cluster";
+export { discoverRenderedSurfaces } from "./discovery";
+export { projectRenderedSurfaces } from "./project";
+export type {
+  ClusteredRenderedSurfaces,
+  RenderedSurface,
+  RenderedSurfaceGroup,
+} from "./types";

apps/extension/src/tools/vom/rendered-surfaces/project.ts

@@ -0,0 +1,108 @@
+import type { VomScene, VomVisualSurface } from "@browser-skill/vom";
+import type { CapturedFrameDocument } from "../frame-capture";
+import type { SemanticAxNode } from "../semantic-graph";
+import type { ClusteredRenderedSurfaces, RenderedSurfaceGroup } from "./types";
+
+function frameBackendKey(frameId: string, backendNodeId: number): string {
+  return `${frameId}\u0000${backendNodeId}`;
+}
+
+interface ProjectionIndex {
+  sceneByBackend: ReadonlyMap<string, number>;
+  documentByFrame: ReadonlyMap<string, CapturedFrameDocument<SemanticAxNode>>;
+  domByFrame: ReadonlyMap<
+    string,
+    ReadonlyMap<number, CapturedFrameDocument<SemanticAxNode>["domNodes"][number]>
+  >;
+  rootSceneIdByFrame: ReadonlyMap<string, number>;
+}
+
+function nearestSceneParent(index: ProjectionIndex, group: RenderedSurfaceGroup): number | null {
+  const document = index.documentByFrame.get(group.frameId);
+  const domByBackend = index.domByFrame.get(group.frameId) ?? new Map();
+  let parentBackendNodeId = group.representative.parentBackendNodeId;
+  let guard = 0;
+  while (parentBackendNodeId !== null && guard <= domByBackend.size) {
+    const sceneId = index.sceneByBackend.get(frameBackendKey(group.frameId, parentBackendNodeId));
+    if (sceneId !== undefined) return sceneId;
+    parentBackendNodeId = domByBackend.get(parentBackendNodeId)?.parentBackendNodeId ?? null;
+    guard += 1;
+  }
+
+  if (document?.parentFrameId && document.ownerBackendNodeId !== undefined) {
+    const owner = index.sceneByBackend.get(
+      frameBackendKey(document.parentFrameId, document.ownerBackendNodeId),
+    );
+    if (owner !== undefined) return owner;
+  }
+
+  return index.rootSceneIdByFrame.get(group.frameId) ?? null;
+}
+
+function buildProjectionIndex(
+  scene: VomScene,
+  documents: CapturedFrameDocument<SemanticAxNode>[],
+): ProjectionIndex {
+  const rootSceneIdByFrame = new Map<string, number>();
+  for (const node of scene.nodes) {
+    if (
+      node.frameId &&
+      !rootSceneIdByFrame.has(node.frameId) &&
+      ["rootwebarea", "webarea"].includes(node.role?.toLowerCase() ?? "")
+    ) {
+      rootSceneIdByFrame.set(node.frameId, node.id);
+    }
+  }
+  return {
+    sceneByBackend: new Map(
+      scene.nodes.flatMap((node) =>
+        node.backendNodeId === undefined
+          ? []
+          : [[frameBackendKey(node.frameId ?? "", node.backendNodeId), node.id] as const],
+      ),
+    ),
+    documentByFrame: new Map(documents.map((document) => [document.frameId, document])),
+    domByFrame: new Map(
+      documents.map((document) => [
+        document.frameId,
+        new Map(document.domNodes.map((node) => [node.backendNodeId, node])),
+      ]),
+    ),
+    rootSceneIdByFrame,
+  };
+}
+
+export function projectRenderedSurfaces(
+  scene: VomScene,
+  documents: CapturedFrameDocument<SemanticAxNode>[],
+  clustered: ClusteredRenderedSurfaces,
+): VomScene {
+  const index = buildProjectionIndex(scene, documents);
+  const visualSurfaces: VomVisualSurface[] = [];
+
+  for (const group of clustered.groups) {
+    const parentId = nearestSceneParent(index, group);
+    visualSurfaces.push({
+      parentId,
+      backendNodeId: group.representative.backendNodeId,
+      frameId: group.frameId,
+      renderingKind: "canvas",
+      visibleRect: group.representative.visibleRect,
+      ...(group.label ? { label: group.label } : {}),
+      memberCount: group.memberCount,
+    });
+  }
+
+  return {
+    ...scene,
+    ...(visualSurfaces.length > 0 ? { visualSurfaces } : {}),
+    ...(clustered.truncated
+      ? {
+          visualSurfacesTruncated: true,
+          ...(clustered.omittedCount !== undefined
+            ? { omittedVisualSurfaceCount: clustered.omittedCount }
+            : {}),
+        }
+      : {}),
+  };
+}

apps/extension/src/tools/vom/rendered-surfaces/types.ts

@@ -0,0 +1,25 @@
+import type { Rect } from "@browser-skill/vom";
+
+export interface RenderedSurface {
+  renderingKind: "canvas";
+  frameId: string;
+  backendNodeId: number;
+  parentBackendNodeId: number | null;
+  visibleRect: Rect;
+  paintOrder: number;
+  label?: string;
+}
+
+export interface RenderedSurfaceGroup {
+  frameId: string;
+  parentBackendNodeId: number | null;
+  representative: RenderedSurface;
+  label?: string;
+  memberCount: number;
+}
+
+export interface ClusteredRenderedSurfaces {
+  groups: RenderedSurfaceGroup[];
+  truncated: boolean;
+  omittedCount?: number;
+}

apps/extension/src/transport/types.ts

@@ -23,6 +23,7 @@ export type ErrorCode =
 export type RpcErrorReason =
   | "agent_window_scope"
   | "element_not_visible"
+  | "ref_capability_denied"
   | "ref_not_found"
   | "selector_not_found"
   | "target_not_fillable"

apps/extension/test-fixtures/rendered-surfaces/README.md

@@ -0,0 +1,110 @@
+# Rendered Surface browser fixtures
+
+These pages exercise Canvas discovery through a real browser DOMSnapshot and AX tree. They are deliberately framework- and site-independent.
+
+## Serve
+
+From the repository root:
+
+```bash
+python3 -m http.server 4173 --bind 127.0.0.1 \
+  --directory apps/extension/test-fixtures/rendered-surfaces
+```
+
+Use an Agent Window large enough to keep the primary matrix visible:
+
+```bash
+bsk session start --width 1440 --height 1100
+bsk navigate http://127.0.0.1:4173/index.html --session <id>
+bsk observe --session <id>
+```
+
+Always stop the session after the run.
+
+## Pages
+
+### `index.html`
+
+Expected individual Surface labels:
+
+- `fixture-basic-large`
+- `fixture-tiny-8px`
+- `fixture-pointer-events-none`
+- `fixture-partial-ax`
+- `fixture-contained-outer`
+- `fixture-contained-inner`
+- `fixture-different-parent-a`
+- `fixture-different-parent-b`
+
+Expected one grouped Surface:
+
+- `fixture-layered-group`, with `layers=2`
+
+Expected absent:
+
+- `fixture-display-none`
+- `fixture-visibility-hidden`
+- `fixture-opacity-zero`
+- `fixture-parent-opacity-zero`
+- `fixture-hidden-attribute`
+- `fixture-inert-parent`
+- `fixture-aria-hidden-parent`
+- `fixture-zero-size`
+- `fixture-offscreen-right`
+- `fixture-fully-overflow-clipped`
+
+`fixture-partially-overflow-clipped` must remain present because part of it is visible. Its screenshot must contain only the browser's final composited view.
+
+### `unlabeled.html`
+
+This page contains eight Canvas elements but must render seven generic Surface markers because
+the two coincident layers form one group.
+
+- Every marker uses the stable fallback label `canvas visual surface`.
+- Exactly one marker has `layers=2`.
+- The two contained canvases remain separate markers.
+- `id`, `class`, `data-testid`, nearby prose, and fallback AX content never become a Surface label.
+- Blank `aria-label` and `title` values behave as no label.
+- Fallback AX content does not suppress the corresponding Surface.
+
+### `frames.html`
+
+Expected Surface labels:
+
+- `fixture-same-origin-frame`
+- `fixture-cross-origin-frame`
+- `fixture-nested-frame`
+
+The cross-origin frame uses `http://localhost:4173` while the parent uses `http://127.0.0.1:4173`. With Chromium site isolation this exercises the OOPIF path without an external service.
+
+### `viewport.html`
+
+Run with a `900x700` Agent Window.
+
+- `fixture-partial-viewport` is partly visible and must be represented.
+- `fixture-below-fold` is absent initially.
+- After scrolling it into view and observing again, `fixture-below-fold` must be represented with a fresh ref.
+
+### `dynamic.html`
+
+Use the normal DOM buttons and re-observe after each action:
+
+- Initial state: `fixture-dynamic` absent.
+- After **Insert canvas**: present.
+- After **Resize to zero**: absent.
+- After **Restore size**: present with a fresh ref.
+- After **Remove canvas**: absent.
+
+### `sparklines.html`
+
+Expected Surface labels `fixture-sparkline-01` through `fixture-sparkline-20` must all be present.
+The adjacent, non-overlapping canvases remain twenty independent screenshot-only Surface refs.
+
+## Invariants
+
+- `snapshot` must not emit any of these Surface markers.
+- `observe` must emit only currently visible Canvas surfaces.
+- Existing DOM/AX controls remain readable alongside Surface markers.
+- Missing labels never prevent discovery; naming and discovery are independent.
+- A Surface ref allows `screenshot --ref` and rejects ordinary click/fill/hover/select.
+- No fixture URL, id, label, or layout constant may appear in production discovery code.

apps/extension/test-fixtures/rendered-surfaces/dynamic.html

@@ -0,0 +1,46 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <title>Rendered Surface dynamic matrix</title>
+    <style>
+      html, body { margin: 0; font: 14px system-ui, sans-serif; }
+      body { padding: 16px; }
+      .controls { display: flex; gap: 8px; margin-bottom: 16px; }
+      button { padding: 8px 12px; }
+      canvas { display: block; width: 260px; height: 120px; background: #e8daef; border: 2px solid #7d3c98; }
+    </style>
+  </head>
+  <body>
+    <h1>Dynamic Canvas lifecycle</h1>
+    <div class="controls">
+      <button id="insert" type="button">Insert canvas</button>
+      <button id="zero" type="button">Resize to zero</button>
+      <button id="restore" type="button">Restore size</button>
+      <button id="remove" type="button">Remove canvas</button>
+    </div>
+    <div id="mount"></div>
+    <script>
+      const mount = document.querySelector("#mount");
+      const getCanvas = () => mount.querySelector("canvas");
+      document.querySelector("#insert").addEventListener("click", () => {
+        if (getCanvas()) return;
+        const canvas = document.createElement("canvas");
+        canvas.width = 260;
+        canvas.height = 120;
+        canvas.setAttribute("aria-label", "fixture-dynamic");
+        mount.append(canvas);
+      });
+      document.querySelector("#zero").addEventListener("click", () => {
+        const canvas = getCanvas();
+        if (canvas) canvas.style.cssText = "width:0;height:0;border:0";
+      });
+      document.querySelector("#restore").addEventListener("click", () => {
+        const canvas = getCanvas();
+        if (canvas) canvas.removeAttribute("style");
+      });
+      document.querySelector("#remove").addEventListener("click", () => getCanvas()?.remove());
+    </script>
+  </body>
+</html>

apps/extension/test-fixtures/rendered-surfaces/frame.html

@@ -0,0 +1,28 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <style>
+      html, body { margin: 0; font: 14px system-ui, sans-serif; }
+      body { padding: 12px; background: #fdfefe; }
+      canvas { display: block; width: 220px; height: 90px; background: #d5f5e3; border: 2px solid #1e8449; }
+      iframe { width: 280px; height: 150px; margin-top: 10px; border: 2px solid #85929e; }
+    </style>
+  </head>
+  <body>
+    <canvas id="frame-canvas" width="220" height="90"></canvas>
+    <script>
+      const params = new URLSearchParams(location.search);
+      const canvas = document.querySelector("#frame-canvas");
+      if (params.get("canvas") === "0") canvas.remove();
+      else canvas.setAttribute("aria-label", params.get("label") || "fixture-frame-default");
+      if (params.get("nested") === "1") {
+        const iframe = document.createElement("iframe");
+        iframe.title = "nested fixture frame";
+        iframe.src = "frame.html?label=fixture-nested-frame";
+        document.body.append(iframe);
+      }
+    </script>
+  </body>
+</html>

apps/extension/test-fixtures/rendered-surfaces/frames.html

@@ -0,0 +1,22 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <title>Rendered Surface frame matrix</title>
+    <style>
+      html, body { margin: 0; font: 14px system-ui, sans-serif; }
+      body { padding: 16px; }
+      .frames { display: grid; grid-template-columns: repeat(3, 320px); gap: 16px; }
+      iframe { width: 300px; height: 300px; border: 2px solid #5d6d7e; }
+    </style>
+  </head>
+  <body>
+    <h1>Frame matrix</h1>
+    <div class="frames">
+      <iframe title="same-origin fixture" src="frame.html?label=fixture-same-origin-frame"></iframe>
+      <iframe title="cross-origin fixture" src="http://localhost:4173/frame.html?label=fixture-cross-origin-frame"></iframe>
+      <iframe title="nested fixture" src="frame.html?canvas=0&nested=1"></iframe>
+    </div>
+  </body>
+</html>

apps/extension/test-fixtures/rendered-surfaces/index.html

@@ -0,0 +1,118 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <title>Rendered Surface discovery matrix</title>
+    <style>
+      * { box-sizing: border-box; }
+      html, body { margin: 0; min-height: 100%; font: 14px/1.4 system-ui, sans-serif; }
+      body { padding: 16px; background: #f4f6f8; color: #17202a; }
+      h1 { margin: 0 0 12px; font-size: 20px; }
+      .matrix { display: grid; grid-template-columns: repeat(4, minmax(220px, 1fr)); gap: 12px; }
+      .case { min-height: 150px; padding: 10px; overflow: hidden; border: 1px solid #aeb6bf; border-radius: 6px; background: white; }
+      .case h2 { margin: 0 0 8px; font-size: 13px; }
+      canvas { display: block; background: #d6eaf8; border: 2px solid #2874a6; }
+      .stack { position: relative; width: 180px; height: 90px; }
+      .stack canvas { position: absolute; inset: 0; width: 180px; height: 90px; }
+      .stack canvas + canvas { background: rgb(244 208 63 / 45%); border-color: #b7950b; }
+      .contained { position: relative; width: 190px; height: 100px; }
+      .contained > canvas:first-child { width: 190px; height: 100px; }
+      .contained > canvas:last-child { position: absolute; left: 55px; top: 30px; width: 70px; height: 40px; background: #fadbd8; border-color: #c0392b; }
+      .same-place { position: relative; width: 180px; height: 90px; }
+      .same-place > div { position: absolute; inset: 0; }
+      .same-place canvas { width: 180px; height: 90px; }
+      .same-place > div:last-child canvas { border-color: #7d3c98; background: rgb(210 180 222 / 35%); }
+      .clip-box { width: 100px; height: 70px; overflow: hidden; border: 2px dashed #566573; }
+      .partially-clipped { width: 140px; height: 60px; transform: translateX(45px); }
+      .fully-clipped { width: 80px; height: 50px; transform: translateX(130px); }
+      .tiny { width: 8px; height: 8px; border-width: 1px; }
+    </style>
+  </head>
+  <body>
+    <h1>Rendered Surface discovery matrix</h1>
+    <main class="matrix">
+      <section class="case">
+        <h2>Visible named canvas</h2>
+        <canvas aria-label="fixture-basic-large" width="200" height="90"></canvas>
+      </section>
+
+      <section class="case">
+        <h2>Tiny canvas remains addressable</h2>
+        <canvas class="tiny" aria-label="fixture-tiny-8px" width="8" height="8"></canvas>
+      </section>
+
+      <section class="case">
+        <h2>Visible but pointer-events none</h2>
+        <canvas aria-label="fixture-pointer-events-none" width="180" height="80" style="pointer-events: none"></canvas>
+      </section>
+
+      <section class="case">
+        <h2>Partial AX/fallback content</h2>
+        <canvas aria-label="fixture-partial-ax" width="180" height="80">
+          <button type="button">Fallback button must not suppress the surface</button>
+        </canvas>
+      </section>
+
+      <section class="case">
+        <h2>Two real rendering layers</h2>
+        <div class="stack">
+          <canvas aria-label="fixture-layered-group" width="180" height="90"></canvas>
+          <canvas width="180" height="90"></canvas>
+        </div>
+      </section>
+
+      <section class="case">
+        <h2>Contained canvases are independent</h2>
+        <div class="contained">
+          <canvas aria-label="fixture-contained-outer" width="190" height="100"></canvas>
+          <canvas aria-label="fixture-contained-inner" width="70" height="40"></canvas>
+        </div>
+      </section>
+
+      <section class="case">
+        <h2>Same geometry, different parents</h2>
+        <div class="same-place">
+          <div><canvas aria-label="fixture-different-parent-a" width="180" height="90"></canvas></div>
+          <div><canvas aria-label="fixture-different-parent-b" width="180" height="90"></canvas></div>
+        </div>
+      </section>
+
+      <section class="case">
+        <h2>Partially overflow-clipped</h2>
+        <div class="clip-box">
+          <canvas class="partially-clipped" aria-label="fixture-partially-overflow-clipped" width="140" height="60"></canvas>
+        </div>
+      </section>
+
+      <section class="case">
+        <h2>Fully overflow-clipped</h2>
+        <div class="clip-box">
+          <canvas class="fully-clipped" aria-label="fixture-fully-overflow-clipped" width="80" height="50"></canvas>
+        </div>
+      </section>
+
+      <section class="case">
+        <h2>CSS-hidden variants</h2>
+        <canvas aria-label="fixture-display-none" width="80" height="40" style="display: none"></canvas>
+        <canvas aria-label="fixture-visibility-hidden" width="80" height="40" style="visibility: hidden"></canvas>
+        <canvas aria-label="fixture-opacity-zero" width="80" height="40" style="opacity: 0"></canvas>
+      </section>
+
+      <section class="case">
+        <h2>Hidden ancestor variants</h2>
+        <div style="opacity: 0"><canvas aria-label="fixture-parent-opacity-zero" width="80" height="40"></canvas></div>
+        <div hidden><canvas aria-label="fixture-hidden-attribute" width="80" height="40"></canvas></div>
+        <div inert><canvas aria-label="fixture-inert-parent" width="80" height="40"></canvas></div>
+        <div aria-hidden="true"><canvas aria-label="fixture-aria-hidden-parent" width="80" height="40"></canvas></div>
+      </section>
+
+      <section class="case">
+        <h2>Zero-size and offscreen</h2>
+        <canvas aria-label="fixture-zero-size" width="80" height="40" style="width: 0; height: 0; border: 0"></canvas>
+        <canvas aria-label="fixture-offscreen-right" width="80" height="40" style="position: fixed; left: 5000px; top: 0"></canvas>
+      </section>
+    </main>
+  </body>
+</html>
+

apps/extension/test-fixtures/rendered-surfaces/sparklines.html

@@ -0,0 +1,28 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <title>Rendered Surface sparkline matrix</title>
+    <style>
+      html, body { margin: 0; font: 14px system-ui, sans-serif; }
+      body { padding: 16px; }
+      main { display: grid; grid-template-columns: repeat(5, 120px); gap: 12px; }
+      canvas { display: block; width: 100px; height: 30px; background: #eaf2f8; }
+    </style>
+  </head>
+  <body>
+    <h1>Twenty independent sparklines</h1>
+    <main id="sparklines"></main>
+    <script>
+      const host = document.querySelector("#sparklines");
+      for (let index = 1; index <= 20; index += 1) {
+        const canvas = document.createElement("canvas");
+        canvas.width = 100;
+        canvas.height = 30;
+        canvas.setAttribute("aria-label", `fixture-sparkline-${String(index).padStart(2, "0")}`);
+        host.append(canvas);
+      }
+    </script>
+  </body>
+</html>

apps/extension/test-fixtures/rendered-surfaces/unlabeled.html

@@ -0,0 +1,68 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <title>Unlabeled Rendered Surface matrix</title>
+    <style>
+      * { box-sizing: border-box; }
+      html, body { margin: 0; min-height: 100%; font: 14px/1.4 system-ui, sans-serif; }
+      body { padding: 16px; background: #f4f6f8; color: #17202a; }
+      h1 { margin: 0 0 12px; font-size: 20px; }
+      .matrix { display: grid; grid-template-columns: repeat(3, minmax(240px, 1fr)); gap: 12px; }
+      .case { min-height: 150px; padding: 10px; overflow: hidden; border: 1px solid #aeb6bf; border-radius: 6px; background: white; }
+      .case h2 { margin: 0 0 8px; font-size: 13px; }
+      .case p { margin: 0 0 8px; }
+      canvas { display: block; background: #d6eaf8; border: 2px solid #2874a6; }
+      .stack { position: relative; width: 180px; height: 90px; }
+      .stack canvas { position: absolute; inset: 0; width: 180px; height: 90px; }
+      .stack canvas + canvas { background: rgb(244 208 63 / 45%); border-color: #b7950b; }
+      .contained { position: relative; width: 190px; height: 100px; }
+      .contained > canvas:first-child { width: 190px; height: 100px; }
+      .contained > canvas:last-child { position: absolute; left: 55px; top: 30px; width: 70px; height: 40px; background: #fadbd8; border-color: #c0392b; }
+    </style>
+  </head>
+  <body>
+    <h1>Unlabeled Rendered Surface matrix</h1>
+    <main class="matrix">
+      <section class="case">
+        <h2>Only engineering attributes</h2>
+        <canvas id="must-not-be-a-label" class="must-not-be-a-label" data-testid="must-not-be-a-label" width="200" height="90"></canvas>
+      </section>
+
+      <section class="case">
+        <h2>Blank explicit names</h2>
+        <canvas aria-label="   " title="   " width="180" height="80"></canvas>
+      </section>
+
+      <section class="case">
+        <h2>Nearby prose is not a Canvas name</h2>
+        <p>must-not-be-inferred-as-a-surface-label</p>
+        <canvas width="180" height="80"></canvas>
+      </section>
+
+      <section class="case">
+        <h2>Fallback AX content does not name or suppress the Canvas</h2>
+        <canvas width="180" height="80">
+          <button type="button">Fallback control remains separate</button>
+        </canvas>
+      </section>
+
+      <section class="case">
+        <h2>Unnamed rendering layers still cluster</h2>
+        <div class="stack">
+          <canvas width="180" height="90"></canvas>
+          <canvas width="180" height="90"></canvas>
+        </div>
+      </section>
+
+      <section class="case">
+        <h2>Unnamed contained canvases stay independent</h2>
+        <div class="contained">
+          <canvas width="190" height="100"></canvas>
+          <canvas width="70" height="40"></canvas>
+        </div>
+      </section>
+    </main>
+  </body>
+</html>

apps/extension/test-fixtures/rendered-surfaces/viewport.html

@@ -0,0 +1,22 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <title>Rendered Surface viewport matrix</title>
+    <style>
+      html, body { margin: 0; font: 14px system-ui, sans-serif; }
+      body { min-height: 1800px; padding: 16px; }
+      canvas { display: block; background: #fdebd0; border: 2px solid #ca6f1e; }
+      #partial { position: fixed; left: -70px; top: 120px; width: 180px; height: 100px; }
+      #below { position: absolute; left: 40px; top: 1350px; width: 240px; height: 120px; }
+    </style>
+  </head>
+  <body>
+    <h1>Viewport matrix</h1>
+    <p>The orange canvas on the left is intentionally clipped by the viewport.</p>
+    <canvas id="partial" aria-label="fixture-partial-viewport" width="180" height="100"></canvas>
+    <canvas id="below" aria-label="fixture-below-fold" width="240" height="120"></canvas>
+  </body>
+</html>
+

crates/bsk-cli/skill/SKILL.md

@@ -73,6 +73,16 @@ CSS-only hover menu looks like from here. It hovers a bounded set of likely trig
 few seconds and touches the live page; once you know which element hides the menu, `bsk hover <ref>`
 is cheaper and more precise.
 
+When `bsk observe` renders
+`@eN surface ... [visual-only; requires=image-understanding; ...]`, the ref identifies rendered
+canvas content that is not represented by the text observation, not an interactable DOM control.
+Keep using any reliable DOM/AX text and controls that appear alongside it. If you can actually
+inspect image output, use `bsk screenshot --ref @eN --session <id>` to obtain the visible crop. If
+you lack multimodal or image-reading capability, do not take a screenshot and pretend to know its
+contents, and never guess coordinates; tell the user that they need to switch to a model with
+image-understanding capability. Do not pass a visual surface ref to click, fill, hover, or select;
+those commands intentionally reject screenshot-only refs.
+
 Escalate page reading only as needed:
 
 1. `bsk observe` for normal semantic understanding, text, controls, and refs.

crates/bsk-cli/src/cli/render_error.rs

@@ -40,6 +40,7 @@ use bsk_protocol::ErrorCode;
 pub mod reason {
     pub const AGENT_WINDOW_SCOPE: &str = "agent_window_scope";
     pub const ELEMENT_NOT_VISIBLE: &str = "element_not_visible";
+    pub const REF_CAPABILITY_DENIED: &str = "ref_capability_denied";
     pub const REF_NOT_FOUND: &str = "ref_not_found";
     pub const SELECTOR_NOT_FOUND: &str = "selector_not_found";
     pub const TARGET_NOT_FILLABLE: &str = "target_not_fillable";
@@ -299,6 +300,13 @@ pub fn info_for_error(code: ErrorCode, data: Option<&serde_json::Value>) -> Rend
             ),
             exit_code: base.exit_code,
         },
+        (ErrorCode::PermissionDenied, reason::REF_CAPABILITY_DENIED) => RenderInfo {
+            summary: "snapshot ref does not support this operation",
+            hint: Some(
+                "visual surface refs are screenshot-only; inspect `bsk screenshot --ref <ref> --session <id>` with image-understanding capability",
+            ),
+            exit_code: base.exit_code,
+        },
         (ErrorCode::InvalidParams, reason::TARGET_NOT_SELECT) => RenderInfo {
             summary: "target element is not a <select>",
             hint: Some(
@@ -564,6 +572,18 @@ mod tests {
         assert_eq!(info.exit_code, 3);
     }
 
+    #[test]
+    fn ref_capability_denied_explains_visual_surface_refs() {
+        let data = serde_json::json!({ "reason": reason::REF_CAPABILITY_DENIED });
+        let info = info_for_error(ErrorCode::PermissionDenied, Some(&data));
+        assert_eq!(info.summary, "snapshot ref does not support this operation");
+        assert!(
+            info.hint.unwrap().contains("screenshot-only"),
+            "expected visual-surface-specific hint"
+        );
+        assert_eq!(info.exit_code, 1);
+    }
+
     #[test]
     fn session_busy_overrides_timeout_copy() {
         let data = serde_json::json!({ "reason": reason::SESSION_BUSY });

packages/dsh-plugin-browserskill/src/tools.ts

@@ -489,7 +489,9 @@ function defineBrowserOperations(deps: ToolDeps, register: DefinitionRegistrar):
       ? "Capture an indented aria-tree snapshot of the session's active tab. Interactive elements " +
         "carry @eN refs for browser_interact actions. Prefer browser_inspect action=observe for a richer semantic view."
       : "Produce a semantic VOM observation of the session's active tab (roles, states, perception " +
-        "probes) with @eN refs for browser_interact actions. Read-only: never submits input.";
+        "probes) with @eN refs for browser_interact actions. Visual-only surfaces require image " +
+        "understanding and their refs are screenshot-only; keep using any DOM/AX semantics rendered " +
+        "alongside them. Read-only: never submits input.";
     register(
       defineTool({
         name,
@@ -754,12 +756,17 @@ function defineBrowserOperations(deps: ToolDeps, register: DefinitionRegistrar):
       name: "inspect.screenshot",
       description:
         "Capture a PNG screenshot of the session's active tab, or crop to a snapshot ref element. " +
+        "This is the supported operation for visual surface refs emitted by inspect.observe. Only use " +
+        "it to interpret a visual-only surface when you can inspect image output. If you lack multimodal " +
+        "or image-reading capability, tell the user to switch to a model with image understanding; do " +
+        "not guess content or coordinates. " +
         "Returns the image itself when the deployment supports image input, otherwise a file path.",
       parameters: {
         session: SESSION_PARAM,
         ref: {
           type: "string",
-          description: "Snapshot ref (@e3) from the last snapshot/observe; crops to that element.",
+          description:
+            "Ref (@e3) from the last snapshot/observe; crops to that DOM element or visual surface.",
         },
       },
       output: {

packages/vom/src/__tests__/render.test.ts

@@ -29,6 +29,159 @@ function coreRefs(refs: ReturnType<typeof renderVom>["refs"]) {
 }
 
 describe("renderVom single-layer page", () => {
+  it("renders visual surfaces as screenshot-only refs", () => {
+    const input = scene([
+      node({ id: 1, role: "RootWebArea", frameId: "main" }),
+      node({ id: 2, parentId: 1, role: "Iframe", frameId: "main", backendNodeId: 20 }),
+    ]);
+    input.visualSurfaces = [
+      {
+        parentId: 2,
+        backendNodeId: 99,
+        frameId: "child",
+        renderingKind: "canvas",
+        visibleRect: { x: 10, y: 20, w: 800, h: 500 },
+        label: "Data grid",
+        memberCount: 2,
+      },
+    ];
+    const out = renderVom(input);
+
+    expect(out.text).toContain("[visual surfaces]");
+    expect(out.text).toContain(
+      '@e1 surface "Data grid" [bounds=10,20,800,500; rendering=canvas; layers=2; visual-only; screenshot-only; requires=image-understanding]',
+    );
+    expect(out.refs).toContainEqual(
+      expect.objectContaining({
+        ref: "e1",
+        backendNodeId: 99,
+        kind: "surface",
+        visibleRect: { x: 10, y: 20, w: 800, h: 500 },
+        capabilities: ["screenshot"],
+      }),
+    );
+  });
+
+  it("renders an unnamed visual surface with a stable generic label", () => {
+    const input = scene([node({ id: 1, role: "RootWebArea", frameId: "main" })]);
+    input.visualSurfaces = [
+      {
+        parentId: 1,
+        backendNodeId: 99,
+        frameId: "main",
+        renderingKind: "canvas",
+        visibleRect: { x: 10, y: 20, w: 800, h: 500 },
+        memberCount: 1,
+      },
+    ];
+
+    const out = renderVom(input);
+
+    expect(out.text).toContain(
+      '@e1 surface "canvas visual surface" [bounds=10,20,800,500; rendering=canvas; visual-only; screenshot-only; requires=image-understanding]',
+    );
+    expect(out.refs).toContainEqual(
+      expect.objectContaining({
+        ref: "e1",
+        name: "canvas visual surface",
+        kind: "surface",
+        capabilities: ["screenshot"],
+      }),
+    );
+  });
+
+  it("appends visual refs after DOM refs so canvas density cannot renumber controls", () => {
+    const input = scene([
+      node({ id: 1, role: "RootWebArea", frameId: "main" }),
+      node({ id: 2, parentId: 1, role: "button", name: "Submit", tag: "button" }),
+    ]);
+    input.visualSurfaces = [
+      {
+        parentId: 1,
+        backendNodeId: 99,
+        frameId: "main",
+        renderingKind: "canvas",
+        visibleRect: { x: 10, y: 20, w: 80, h: 50 },
+        memberCount: 1,
+      },
+    ];
+
+    const withoutVisuals = renderVom({ ...input, visualSurfaces: [] });
+    const withVisuals = renderVom(input);
+
+    expect(withoutVisuals.refs[0]).toMatchObject({ ref: "e1", backendNodeId: 2, kind: "dom" });
+    expect(withVisuals.refs[0]).toMatchObject({ ref: "e1", backendNodeId: 2, kind: "dom" });
+    expect(withVisuals.refs[1]).toMatchObject({ ref: "e2", backendNodeId: 99, kind: "surface" });
+    expect(withVisuals.text.indexOf('@e1 button "Submit"')).toBeLessThan(
+      withVisuals.text.indexOf("[visual surfaces]"),
+    );
+  });
+
+  it("uses the appendix token budget and prefers explicitly labeled surfaces", () => {
+    const input = scene([node({ id: 1, role: "RootWebArea", frameId: "main" })]);
+    input.visualSurfaces = Array.from({ length: 12 }, (_, index) => ({
+      parentId: 1,
+      backendNodeId: 100 + index,
+      frameId: "main",
+      renderingKind: "canvas" as const,
+      visibleRect: { x: index * 10, y: 20, w: 8, h: 8 },
+      ...(index === 11 ? { label: "Important status" } : {}),
+      memberCount: 1,
+    }));
+
+    const out = renderVom(input);
+    const surfaceRefs = out.refs.filter((ref) => ref.kind === "surface");
+
+    expect(surfaceRefs).toHaveLength(12);
+    expect(surfaceRefs[0]).toMatchObject({ backendNodeId: 111, name: "Important status" });
+    expect(out.truncated).toBe(false);
+  });
+
+  it("renders twenty independent sparklines without changing DOM refs", () => {
+    const input = scene([
+      node({ id: 1, role: "RootWebArea", frameId: "main" }),
+      node({ id: 2, parentId: 1, role: "button", name: "Submit", tag: "button" }),
+    ]);
+    input.visualSurfaces = Array.from({ length: 20 }, (_, index) => ({
+      parentId: 1,
+      backendNodeId: 100 + index,
+      frameId: "main",
+      renderingKind: "canvas" as const,
+      visibleRect: { x: index * 12, y: 20, w: 10, h: 8 },
+      memberCount: 1,
+    }));
+
+    const out = renderVom(input);
+
+    expect(out.refs[0]).toMatchObject({ ref: "e1", backendNodeId: 2, kind: "dom" });
+    expect(out.refs.filter((ref) => ref.kind === "surface")).toHaveLength(20);
+    expect(out.truncated).toBe(false);
+  });
+
+  it("does not let a visual surface exhaust the budget before later DOM content", () => {
+    const input = scene([
+      node({ id: 1, role: "RootWebArea", name: "Doc", frameId: "main" }),
+      node({ id: 2, parentId: 1, role: "button", name: "First", tag: "button" }),
+      node({ id: 3, parentId: 1, role: "button", name: "Later", tag: "button" }),
+    ]);
+    input.visualSurfaces = [
+      {
+        parentId: 1,
+        backendNodeId: 99,
+        frameId: "main",
+        renderingKind: "canvas",
+        visibleRect: { x: 10, y: 20, w: 800, h: 500 },
+        memberCount: 1,
+      },
+    ];
+
+    const out = renderVom(input, { maxTokens: 30 });
+
+    expect(out.text).toContain('@e2 button "Later"');
+    expect(out.refs.filter((ref) => ref.kind === "surface")).toHaveLength(0);
+    expect(out.truncated).toBe(true);
+  });
+
   it("does not derive handle context across frame scopes", () => {
     const out = renderVom(
       scene([

packages/vom/src/__tests__/visual-surface-priority.test.ts

@@ -0,0 +1,35 @@
+import { describe, expect, it } from "vitest";
+import {
+  compareVisualSurfacePriority,
+  selectHighestPriorityVisualSurfaces,
+} from "../visual-surface-priority";
+
+function surface(backendNodeId: number, area: number, label?: string) {
+  return {
+    backendNodeId,
+    frameId: "main",
+    visibleRect: { x: backendNodeId, y: 0, w: area, h: 1 },
+    ...(label ? { label } : {}),
+  };
+}
+
+describe("visual surface priority selection", () => {
+  it("matches a full priority sort when selecting a bounded prefix", () => {
+    const surfaces = Array.from({ length: 600 }, (_, index) => surface(index, index + 1));
+    surfaces[599] = surface(599, 1, "Important status");
+
+    const selected = selectHighestPriorityVisualSurfaces(surfaces, 512);
+    const expected = [...surfaces].sort(compareVisualSurfacePriority).slice(0, 512);
+
+    expect(selected).toEqual(expected);
+    expect(selected[0]?.backendNodeId).toBe(599);
+  });
+
+  it("normalizes empty, fractional, and unbounded limits", () => {
+    const surfaces = [surface(1, 1), surface(2, 2), surface(3, 3)];
+
+    expect(selectHighestPriorityVisualSurfaces(surfaces, 0)).toEqual([]);
+    expect(selectHighestPriorityVisualSurfaces(surfaces, 1.9)).toEqual([surface(3, 3)]);
+    expect(selectHighestPriorityVisualSurfaces(surfaces, Number.POSITIVE_INFINITY)).toHaveLength(3);
+  });
+});

packages/vom/src/index.ts

@@ -15,6 +15,13 @@ export type {
   VomNode,
   VomOptions,
   VomRef,
+  VomRefCapability,
   VomResult,
   VomScene,
+  VomVisualSurface,
 } from "./types";
+export {
+  compareVisualSurfacePriority,
+  selectHighestPriorityVisualSurfaces,
+  type VisualSurfacePriorityInput,
+} from "./visual-surface-priority";

packages/vom/src/render.ts

@@ -9,7 +9,9 @@ import type {
   VomOptions,
   VomResult,
   VomScene,
+  VomVisualSurface,
 } from "./types";
+import { selectHighestPriorityVisualSurfaces } from "./visual-surface-priority";
 
 const SKIP_ROLES = new Set(["generic", "none", "presentation", "inlinetextbox"]);
 
@@ -87,6 +89,8 @@ const MAX_CUSTOM_RECOVERY_AREA = 180_000;
 const MAX_RECOVERABLE_SINGLE_INTENT_AREA = 360_000;
 const MAX_HANDLE_CONTEXT_ITEMS = 3;
 const MAX_SURFACE_ITEMS = 12;
+const DEFAULT_VISUAL_APPENDIX_TOKENS = 2_048;
+const MAX_VISUAL_SURFACE_REFS = 512;
 const MAX_SCOPE_LINES = 40;
 const RECOVERY_HANDLER_ATTRS = ["onclick", "onmousedown", "onkeydown", "onkeyup", "onkeypress"];
 const NATIVE_TAGS = new Set(["button", "input", "select", "textarea"]);
@@ -793,34 +797,117 @@ interface RenderState {
   duplicateRefNames: Set<string>;
   surfaceMap: Map<number, CondSurface>;
   scopeMap: Map<number, ActiveScopeBlock>;
+  visualSurfaces: VomVisualSurface[];
+  visualSurfacesTruncated: boolean;
+  omittedVisualSurfaceCount?: number;
+}
+
+function tryAppendLine(state: RenderState, line: string, tokenLimit = state.maxTokens): boolean {
+  const nextTokens = state.tokens + estimateTokens(line);
+  if (nextTokens > tokenLimit) return false;
+  state.lines.push(line);
+  state.tokens = nextTokens;
+  return true;
+}
+
+function visualSurfaceLine(surface: VomVisualSurface, ref: string): string {
+  const label = cleaned(surface.label) ?? `${surface.renderingKind} visual surface`;
+  const layers = surface.memberCount > 1 ? `; layers=${surface.memberCount}` : "";
+  const { x, y, w, h } = surface.visibleRect;
+  const bounds = [x, y, w, h].map(Math.round).join(",");
+  return `  @${ref} surface ${JSON.stringify(label)} [bounds=${bounds}; rendering=${surface.renderingKind}${layers}; visual-only; screenshot-only; requires=image-understanding]`;
+}
+
+function selectVisualSurfaces(
+  surfaces: readonly VomVisualSurface[],
+  nodesById: ReadonlyMap<number, VomNode>,
+): { selected: VomVisualSurface[]; eligibleCount: number } {
+  let eligibleCount = 0;
+  function* eligibleSurfaces(): Generator<VomVisualSurface> {
+    for (const surface of surfaces) {
+      if (surface.parentId !== null && !nodesById.has(surface.parentId)) continue;
+      eligibleCount += 1;
+      yield surface;
+    }
+  }
+  return {
+    selected: selectHighestPriorityVisualSurfaces(eligibleSurfaces(), MAX_VISUAL_SURFACE_REFS),
+    eligibleCount,
+  };
+}
+
+function emitVisualAppendix(state: RenderState): void {
+  if (state.stopped) return;
+  const { selected, eligibleCount } = selectVisualSurfaces(state.visualSurfaces, state.nodesById);
+  if (eligibleCount === 0 && !state.visualSurfacesTruncated) return;
+
+  const header = "[visual surfaces]";
+  const appendixStartTokens = state.tokens;
+  const appendixTokenLimit = Number.isFinite(state.maxTokens)
+    ? state.maxTokens
+    : appendixStartTokens + DEFAULT_VISUAL_APPENDIX_TOKENS;
+  let headerEmitted = false;
+  let emitted = 0;
+  for (const surface of selected) {
+    if (emitted >= MAX_VISUAL_SURFACE_REFS) break;
+    const ref = `e${state.nextRef}`;
+    const line = visualSurfaceLine(surface, ref);
+    const requiredTokens = estimateTokens(line) + (headerEmitted ? 0 : estimateTokens(header));
+    if (state.tokens + requiredTokens > appendixTokenLimit) break;
+    if (!headerEmitted) {
+      if (!tryAppendLine(state, header, appendixTokenLimit)) break;
+      headerEmitted = true;
+    }
+    if (!tryAppendLine(state, line, appendixTokenLimit)) break;
+    const label = cleaned(surface.label) ?? `${surface.renderingKind} visual surface`;
+    state.refs.push({
+      ref,
+      backendNodeId: surface.backendNodeId,
+      frameId: surface.frameId,
+      visibleRect: surface.visibleRect,
+      role: "surface",
+      name: label,
+      kind: "surface",
+      capabilities: ["screenshot"],
+      line: state.lines.length - 1,
+    });
+    state.nextRef += 1;
+    emitted += 1;
+  }
+
+  const locallyOmitted = eligibleCount - emitted;
+  const upstreamOmitted = state.omittedVisualSurfaceCount ?? 0;
+  const omitted = locallyOmitted + upstreamOmitted;
+  if (omitted > 0 || state.visualSurfacesTruncated) {
+    state.truncated = true;
+    if (headerEmitted) {
+      const count =
+        omitted > 0 ? `${state.visualSurfacesTruncated ? "at least " : ""}${omitted}` : "some";
+      tryAppendLine(state, `  … ${count} additional visual surfaces omitted`, appendixTokenLimit);
+    }
+  }
 }
 
 function emitActiveScopeBlock(scope: ActiveScopeBlock, depth: number, state: RenderState): void {
   if (state.stopped || scope.lines.length === 0) return;
   const indent = "  ".repeat(depth);
   const header = `${indent}[§ active: ${scope.label}]`;
-  const headerTokens = estimateTokens(header);
-  if (state.tokens + headerTokens > state.maxTokens) {
+  if (!tryAppendLine(state, header)) {
     state.truncated = true;
     state.stopped = true;
     return;
   }
-  state.lines.push(header);
-  state.tokens += headerTokens;
 
   for (const text of scope.lines.slice(0, MAX_SCOPE_LINES)) {
     if (state.stopped) return;
     const clean = cleaned(text);
     if (!clean) continue;
     const line = `${indent}  ${clean}`;
-    const nextTokens = state.tokens + estimateTokens(line);
-    if (nextTokens > state.maxTokens) {
+    if (!tryAppendLine(state, line)) {
       state.truncated = true;
       state.stopped = true;
       return;
     }
-    state.lines.push(line);
-    state.tokens = nextTokens;
   }
 }
 
@@ -855,15 +942,11 @@ function renderTree(children: Map<number | null, VomNode[]>, state: RenderState)
     const context = ref ? handleContext(node, state) : [];
     const surface = state.surfaceMap.get(node.id);
     const line = renderNodeLine(node, depth, ref, context, surface, state.redactValues);
-    const nextTokens = state.tokens + estimateTokens(line);
-    if (nextTokens > state.maxTokens) {
+    if (!tryAppendLine(state, line)) {
       state.truncated = true;
       state.stopped = true;
       break;
     }
-
-    state.lines.push(line);
-    state.tokens = nextTokens;
     if (ref) {
       const name = cleaned(node.name);
       state.refs.push({
@@ -873,6 +956,8 @@ function renderTree(children: Map<number | null, VomNode[]>, state: RenderState)
         ...(node.role ? { role: node.role } : {}),
         ...(name ? { name } : {}),
         ...(context.length > 0 ? { ctx: context.join(" > ") } : {}),
+        kind: "dom",
+        capabilities: ["interact", "screenshot"],
         line: state.lines.length - 1,
       });
       state.nextRef += 1;
@@ -894,6 +979,9 @@ function renderNodes(
   initialLines: string[],
   surfaces: CondSurface[] = [],
   activeScopeBlocks: ActiveScopeBlock[] = [],
+  visualSurfaces: VomVisualSurface[] = [],
+  visualSurfacesTruncated = false,
+  omittedVisualSurfaceCount?: number,
 ): RenderState {
   const children = buildChildren(nodes);
   const state: RenderState = {
@@ -913,6 +1001,9 @@ function renderNodes(
     duplicateRefNames: duplicateReferenceNames(nodes),
     surfaceMap: new Map(surfaces.map((surface) => [surface.triggerId, surface])),
     scopeMap: new Map(activeScopeBlocks.map((scope) => [scope.triggerId, scope])),
+    visualSurfaces,
+    visualSurfacesTruncated,
+    omittedVisualSurfaceCount,
   };
 
   renderTree(children, state);
@@ -1124,8 +1215,21 @@ function renderDoubleLayer(scene: VomScene, layer: BlockingLayer, options: VomOp
     "@layers 2 focus=L1",
     `L1 ${layer.kind} cover=${Math.round(layer.coverage * 100)}%`,
   ];
-  const state = renderNodes(visibleNodes, options, header, scene.surfaces, scene.activeScopeBlocks);
-  state.lines.push(renderPageOcclusionLine(hiddenCount));
+  const visibleSurface = (parentId: number | null): boolean =>
+    parentId === null || included.has(parentId);
+  const state = renderNodes(
+    visibleNodes,
+    options,
+    header,
+    scene.surfaces,
+    scene.activeScopeBlocks,
+    scene.visualSurfaces?.filter((surface) => visibleSurface(surface.parentId)),
+    scene.visualSurfacesTruncated,
+    scene.omittedVisualSurfaceCount,
+  );
+  const occlusionLine = renderPageOcclusionLine(hiddenCount);
+  if (!tryAppendLine(state, occlusionLine)) state.truncated = true;
+  emitVisualAppendix(state);
 
   return {
     text: state.lines.join("\n"),
@@ -1152,7 +1256,11 @@ export function renderVom(scene: VomScene, options: VomOptions = {}): VomResult
     ],
     scene.surfaces,
     scene.activeScopeBlocks,
+    scene.visualSurfaces,
+    scene.visualSurfacesTruncated,
+    scene.omittedVisualSurfaceCount,
   );
+  emitVisualAppendix(state);
 
   return {
     text: state.lines.join("\n"),

packages/vom/src/types.ts

@@ -61,6 +61,12 @@ export interface VomScene {
   rootFrameId?: string;
   surfaces?: CondSurface[];
   activeScopeBlocks?: ActiveScopeBlock[];
+  /** Non-semantic rendered regions discovered alongside the AX/DOM tree. */
+  visualSurfaces?: VomVisualSurface[];
+  /** Upstream surface discovery/aggregation stopped at a resource guardrail. */
+  visualSurfacesTruncated?: boolean;
+  /** Exact omitted count when upstream could determine it. */
+  omittedVisualSurfaceCount?: number;
 }
 
 export interface VomOptions {
@@ -94,6 +100,22 @@ export interface VomRef {
   ref: string;
   backendNodeId: number;
   frameId?: string;
+  /** Observation-time top-viewport crop for screenshot-only surfaces. */
+  visibleRect?: Rect;
+  kind?: "dom" | "surface";
+  capabilities?: VomRefCapability[];
+}
+
+export type VomRefCapability = "interact" | "screenshot";
+
+export interface VomVisualSurface {
+  parentId: number | null;
+  backendNodeId: number;
+  frameId: string;
+  renderingKind: "canvas";
+  visibleRect: Rect;
+  label?: string;
+  memberCount: number;
 }
 
 export interface RenderedRef extends VomRef {

packages/vom/src/visual-surface-priority.ts

@@ -0,0 +1,80 @@
+export interface VisualSurfacePriorityInput {
+  label?: string;
+  visibleRect: { x: number; y: number; w: number; h: number };
+  frameId: string;
+  backendNodeId: number;
+}
+
+function hasLabel(label: string | undefined): boolean {
+  return (label?.trim().length ?? 0) > 0;
+}
+
+function area(surface: VisualSurfacePriorityInput): number {
+  return Math.max(0, surface.visibleRect.w) * Math.max(0, surface.visibleRect.h);
+}
+
+/** Negative means `a` should be emitted before `b`. */
+export function compareVisualSurfacePriority(
+  a: VisualSurfacePriorityInput,
+  b: VisualSurfacePriorityInput,
+): number {
+  return (
+    Number(hasLabel(b.label)) - Number(hasLabel(a.label)) ||
+    area(b) - area(a) ||
+    a.frameId.localeCompare(b.frameId) ||
+    a.visibleRect.y - b.visibleRect.y ||
+    a.visibleRect.x - b.visibleRect.x ||
+    a.backendNodeId - b.backendNodeId
+  );
+}
+
+function siftWorstUp<T extends VisualSurfacePriorityInput>(heap: T[], start: number): void {
+  let index = start;
+  while (index > 0) {
+    const parent = Math.floor((index - 1) / 2);
+    if (compareVisualSurfacePriority(heap[index], heap[parent]) <= 0) break;
+    [heap[index], heap[parent]] = [heap[parent], heap[index]];
+    index = parent;
+  }
+}
+
+function siftWorstDown<T extends VisualSurfacePriorityInput>(heap: T[], start: number): void {
+  let index = start;
+  while (true) {
+    const left = index * 2 + 1;
+    if (left >= heap.length) return;
+    const right = left + 1;
+    let worst = left;
+    if (right < heap.length && compareVisualSurfacePriority(heap[right], heap[left]) > 0) {
+      worst = right;
+    }
+    if (compareVisualSurfacePriority(heap[worst], heap[index]) <= 0) return;
+    [heap[index], heap[worst]] = [heap[worst], heap[index]];
+    index = worst;
+  }
+}
+
+/** Selects the best `limit` surfaces and returns them in priority order. */
+export function selectHighestPriorityVisualSurfaces<T extends VisualSurfacePriorityInput>(
+  surfaces: Iterable<T>,
+  limit: number,
+): T[] {
+  if (!Number.isFinite(limit)) return [...surfaces].sort(compareVisualSurfacePriority);
+  const normalizedLimit = Math.max(0, Math.floor(limit));
+  if (normalizedLimit === 0) return [];
+  if (Array.isArray(surfaces) && surfaces.length <= normalizedLimit) {
+    return [...surfaces].sort(compareVisualSurfacePriority);
+  }
+
+  const heap: T[] = [];
+  for (const surface of surfaces) {
+    if (heap.length < normalizedLimit) {
+      heap.push(surface);
+      siftWorstUp(heap, heap.length - 1);
+    } else if (compareVisualSurfacePriority(surface, heap[0]) < 0) {
+      heap[0] = surface;
+      siftWorstDown(heap, 0);
+    }
+  }
+  return heap.sort(compareVisualSurfacePriority);
+}

skill/SKILL.md

@@ -73,6 +73,16 @@ CSS-only hover menu looks like from here. It hovers a bounded set of likely trig
 few seconds and touches the live page; once you know which element hides the menu, `bsk hover <ref>`
 is cheaper and more precise.
 
+When `bsk observe` renders
+`@eN surface ... [visual-only; requires=image-understanding; ...]`, the ref identifies rendered
+canvas content that is not represented by the text observation, not an interactable DOM control.
+Keep using any reliable DOM/AX text and controls that appear alongside it. If you can actually
+inspect image output, use `bsk screenshot --ref @eN --session <id>` to obtain the visible crop. If
+you lack multimodal or image-reading capability, do not take a screenshot and pretend to know its
+contents, and never guess coordinates; tell the user that they need to switch to a model with
+image-understanding capability. Do not pass a visual surface ref to click, fill, hover, or select;
+those commands intentionally reject screenshot-only refs.
+
 Escalate page reading only as needed:
 
 1. `bsk observe` for normal semantic understanding, text, controls, and refs.