Skip to content

Automate browser based workflows with AI

perf(SKY-11332): apply scrape_exclude to get_frame_text (#6770)

LawyZheng0074d7cf0eec
diff --git a/skyvern/webeye/scraper/scraper.py b/skyvern/webeye/scraper/scraper.py
index 906f461e7..6b7080889 100644
--- a/skyvern/webeye/scraper/scraper.py
+++ b/skyvern/webeye/scraper/scraper.py
@@ -274,10 +274,13 @@ async def scrape_website(
)
-async def get_frame_text(iframe: Frame) -> str:
+async def get_frame_text(iframe: Frame, scrape_exclude: ScrapeExcludeFunc | None = None) -> str:
"""
Get all the visible text in the iframe.
:param iframe: Frame instance to get the text from.
+ :param scrape_exclude: Optional ``filter_frames``-style predicate; True
+ means skip. The top-level caller must pass a starting frame the
+ predicate would not itself exclude.
:return: All the visible text from the iframe.
"""
js_script = "() => document.body.innerText"
@@ -297,6 +300,10 @@ async def get_frame_text(iframe: Frame) -> str:
if child_frame.is_detached():
continue
+ # Skip excluded frames before any CDP probe.
+ if scrape_exclude is not None and await scrape_exclude(child_frame.page, child_frame):
+ continue
+
try:
child_frame_element = await child_frame.frame_element()
except Exception:
@@ -310,7 +317,7 @@ async def get_frame_text(iframe: Frame) -> str:
if not await child_frame_element.is_visible():
continue
- text += await get_frame_text(child_frame)
+ text += await get_frame_text(child_frame, scrape_exclude)
return text
@@ -505,7 +512,7 @@ async def scrape_web_unsafe(
if not elements and not support_empty_page:
raise NoElementFound()
- text_content = await get_frame_text(page.main_frame)
+ text_content = await get_frame_text(page.main_frame, scrape_exclude)
html = ""
window_dimension = None
diff --git a/tests/unit/test_get_frame_text_scrape_exclude.py b/tests/unit/test_get_frame_text_scrape_exclude.py
new file mode 100644
index 000000000..a9f64268a
--- /dev/null
+++ b/tests/unit/test_get_frame_text_scrape_exclude.py
@@ -0,0 +1,136 @@
+from unittest.mock import AsyncMock, MagicMock, patch
+from urllib.parse import urlparse
+
+import pytest
+
+from skyvern.webeye.scraper import scraper
+
+
+def _make_frame(url: str, child_frames: list | None = None) -> MagicMock:
+ """Build a Playwright-Frame stand-in good enough to drive get_frame_text."""
+ frame = MagicMock()
+ frame.url = url
+ frame.child_frames = child_frames or []
+ frame.is_detached = MagicMock(return_value=False)
+
+ element = MagicMock()
+ element.is_visible = AsyncMock(return_value=True)
+ frame.frame_element = AsyncMock(return_value=element)
+
+ page = MagicMock()
+ page.url = "https://example.com/"
+ frame.page = page
+ return frame
+
+
+async def test_get_frame_text_evaluates_main_frame_without_exclude() -> None:
+ main = _make_frame("https://example.com/")
+
+ with patch.object(scraper.SkyvernFrame, "evaluate", new=AsyncMock(return_value="hello")):
+ text = await scraper.get_frame_text(main)
+
+ assert text == "hello"
+
+
+async def test_get_frame_text_default_walks_every_child_for_backwards_compat() -> None:
+ child_ad = _make_frame("https://ads.example.invalid/iframe")
+ child_real = _make_frame("https://example.com/embed")
+ main = _make_frame("https://example.com/", child_frames=[child_ad, child_real])
+
+ visited: list[str] = []
+
+ async def _evaluate(frame, expression):
+ visited.append(frame.url)
+ return f"text({frame.url})"
+
+ with patch.object(scraper.SkyvernFrame, "evaluate", new=AsyncMock(side_effect=_evaluate)):
+ text = await scraper.get_frame_text(main)
+
+ # No predicate => historical behavior: every visible non-detached child is walked.
+ assert visited == [
+ "https://example.com/",
+ "https://ads.example.invalid/iframe",
+ "https://example.com/embed",
+ ]
+ assert "text(https://ads.example.invalid/iframe)" in text
+
+
+async def test_get_frame_text_skips_excluded_child_and_its_subtree() -> None:
+ grandchild_under_ad = _make_frame("https://ads.example.invalid/grandchild")
+ child_ad = _make_frame("https://ads.example.invalid/iframe", child_frames=[grandchild_under_ad])
+ child_real = _make_frame("https://example.com/embed")
+ main = _make_frame("https://example.com/", child_frames=[child_ad, child_real])
+
+ async def _exclude(page, frame) -> bool:
+ hostname = urlparse(frame.url or "").hostname or ""
+ return hostname == "ads.example.invalid"
+
+ visited: list[str] = []
+
+ async def _evaluate(frame, expression):
+ visited.append(frame.url)
+ return f"text({frame.url})"
+
+ with patch.object(scraper.SkyvernFrame, "evaluate", new=AsyncMock(side_effect=_evaluate)):
+ text = await scraper.get_frame_text(main, scrape_exclude=_exclude)
+
+ # The excluded child must be dropped *before* recursion, so its grandchild
+ # is unreachable too.
+ assert visited == [
+ "https://example.com/",
+ "https://example.com/embed",
+ ]
+ assert "ads.example.invalid" not in text
+ # frame_element() must not be probed on excluded frames either, so we don't
+ # pay a CDP round trip for frames we already decided to drop.
+ child_ad.frame_element.assert_not_awaited()
+
+
+async def test_get_frame_text_keeps_child_when_exclude_returns_false() -> None:
+ child_keep = _make_frame("https://other.example.com/iframe")
+ main = _make_frame("https://example.com/", child_frames=[child_keep])
+
+ exclude_calls: list[tuple[object, object]] = []
+
+ async def _exclude(page, frame) -> bool:
+ exclude_calls.append((page, frame))
+ return False
+
+ visited: list[str] = []
+
+ async def _evaluate(frame, expression):
+ visited.append(frame.url)
+ return f"text({frame.url})"
+
+ with patch.object(scraper.SkyvernFrame, "evaluate", new=AsyncMock(side_effect=_evaluate)):
+ await scraper.get_frame_text(main, scrape_exclude=_exclude)
+
+ # The predicate is invoked exactly once with (child_frame.page, child_frame)
+ # — matching filter_frames' contract.
+ assert len(exclude_calls) == 1
+ seen_page, seen_frame = exclude_calls[0]
+ assert seen_page is child_keep.page
+ assert seen_frame is child_keep
+ assert visited == [
+ "https://example.com/",
+ "https://other.example.com/iframe",
+ ]
+
+
+async def test_get_frame_text_still_skips_detached_children_before_predicate() -> None:
+ detached_child = _make_frame("https://ads.example.invalid/dead")
+ detached_child.is_detached = MagicMock(return_value=True)
+ main = _make_frame("https://example.com/", child_frames=[detached_child])
+
+ async def _exclude(page, frame) -> bool:
+ raise AssertionError("predicate must not run for detached frames")
+
+ with patch.object(scraper.SkyvernFrame, "evaluate", new=AsyncMock(return_value="root")):
+ text = await scraper.get_frame_text(main, scrape_exclude=_exclude)
+
+ assert text == "root"