Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation.
Rust agentai-agentapplecomputer-usecomputer-use-agentcontainerizationcuadesktop-automationhacktoberfestlumemacosmanusoperatorswiftvirtualizationvirtualization-frameworkwindowswindows-sandbox
from __future__ import annotations
import os
import sys
from types import SimpleNamespace
import pytest
from cua_s1.pdf import PdfError, derive_entities, extract_entities, resolve_pdf_path
from cua_s1.schema import Entity
def test_pdf_path_is_confined_to_allowed_roots(tmp_path):
allowed = tmp_path / "allowed"
allowed.mkdir()
inside = allowed / "record.pdf"
inside.write_bytes(b"%PDF-1.4\n")
outside = tmp_path / "outside.pdf"
outside.write_bytes(b"%PDF-1.4\n")
assert resolve_pdf_path(inside, allowed_roots=[allowed]) == inside.resolve()
with pytest.raises(PdfError) as caught:
resolve_pdf_path(outside, allowed_roots=[allowed])
assert caught.value.code == "pdf_path_not_allowed"
def test_pdf_path_rejects_symlink_escape_and_oversized_file(tmp_path):
allowed = tmp_path / "allowed"
allowed.mkdir()
outside = tmp_path / "outside.pdf"
outside.write_bytes(b"%PDF-1.4\n")
link = allowed / "linked.pdf"
try:
link.symlink_to(outside)
except OSError:
pytest.skip("symlinks are not available")
with pytest.raises(PdfError) as escaped:
resolve_pdf_path(link, allowed_roots=[allowed])
assert escaped.value.code == "pdf_path_not_allowed"
local = allowed / "large.pdf"
local.write_bytes(b"12345")
with pytest.raises(PdfError) as oversized:
resolve_pdf_path(local, allowed_roots=[allowed], max_bytes=4)
assert oversized.value.code == "pdf_too_large"
assert oversized.value.details["size_bytes"] == 5
def test_extract_entities_enforces_page_bound_before_text_extraction(monkeypatch, tmp_path):
path = tmp_path / "record.pdf"
path.write_bytes(b"%PDF-1.4\n")
class Page:
def extract_text(self):
raise AssertionError("text extraction must not begin after the page limit is exceeded")
class Document:
pages = [Page(), Page()]
def __enter__(self):
return self
def __exit__(self, *_args):
return None
monkeypatch.setitem(sys.modules, "pdfplumber", SimpleNamespace(open=lambda _stream: Document()))
with pytest.raises(PdfError) as caught:
extract_entities(path, allowed_roots=[tmp_path], max_pages=1)
assert caught.value.code == "pdf_too_many_pages"
assert caught.value.details == {"path": str(path), "pages": 2, "max_pages": 1}
def test_extract_entities_refuses_a_symlink_swap_before_open(monkeypatch, tmp_path):
path = tmp_path / "record.pdf"
path.write_bytes(b"%PDF-1.4\n")
outside = tmp_path.parent / "outside.pdf"
outside.write_bytes(b"%PDF-1.4\nsecret")
real_open = os.open
swapped = False
def swap_then_open(candidate, flags):
nonlocal swapped
if not swapped and candidate == path.resolve():
swapped = True
path.unlink()
path.symlink_to(outside)
return real_open(candidate, flags)
def must_not_parse(_stream):
raise AssertionError("a swapped path must be refused before PDF parsing starts")
# Stub pdfplumber so this guard is exercised without the optional `pdf`
# extra (for example in the `four-b`-only inference environment).
monkeypatch.setitem(sys.modules, "pdfplumber", SimpleNamespace(open=must_not_parse))
monkeypatch.setattr(os, "open", swap_then_open)
with pytest.raises(PdfError) as caught:
extract_entities(path, allowed_roots=[tmp_path])
assert caught.value.code == "pdf_open_failed"
def test_derive_entities_adds_only_missing_name_variants():
derived = derive_entities([Entity("Full name", "Avery Chen")])
assert derived == [
Entity("Full name", "Avery Chen"),
Entity("First name", "Avery"),
Entity("Last name", "Chen"),
]
combined = derive_entities([Entity("First name", "Avery"), Entity("Last name", "Chen")])
assert combined[-1] == Entity("Full name", "Avery Chen")