Skip to content

Building AI agents, atomically

Reject backslashes before the path in result URLs; cover IPv6 literals; guard connect_ex in tests

Juampiab23d310cbc9
diff --git a/atomic-forge/tools/firecrawl_search/tests/test_firecrawl_search.py b/atomic-forge/tools/firecrawl_search/tests/test_firecrawl_search.py
index f92463b..cd9070a 100644
--- a/atomic-forge/tools/firecrawl_search/tests/test_firecrawl_search.py
+++ b/atomic-forge/tools/firecrawl_search/tests/test_firecrawl_search.py
@@ -1,5 +1,6 @@
import asyncio
import copy
+import errno
import logging
import os
import socket
@@ -75,6 +76,7 @@ def offline(monkeypatch) -> Iterator[List[str]]:
attempts: List[str] = []
real_getaddrinfo = socket.getaddrinfo
real_connect = socket.socket.connect
+ real_connect_ex = socket.socket.connect_ex
def resolve_loopback_only(host, *args, **kwargs):
if host not in LOOPBACK_HOSTS:
@@ -82,14 +84,23 @@ def offline(monkeypatch) -> Iterator[List[str]]:
raise socket.gaierror("network access is disabled in these tests")
return real_getaddrinfo(host, *args, **kwargs)
- def connect_loopback_only(sock, address):
+ def blocked(address) -> bool:
if isinstance(address, tuple) and address[0] not in LOOPBACK_HOSTS:
attempts.append(str(address[0]))
+ return True
+ return False
+
+ def connect_loopback_only(sock, address):
+ if blocked(address):
raise OSError("network access is disabled in these tests")
return real_connect(sock, address)
+ def connect_ex_loopback_only(sock, address):
+ return errno.ECONNREFUSED if blocked(address) else real_connect_ex(sock, address)
+
monkeypatch.setattr(socket, "getaddrinfo", resolve_loopback_only)
monkeypatch.setattr(socket.socket, "connect", connect_loopback_only)
+ monkeypatch.setattr(socket.socket, "connect_ex", connect_ex_loopback_only)
yield attempts
assert attempts == []
@@ -250,6 +261,7 @@ def test_to_item_drops_badly_typed_fields(field, value, attribute, expected):
"https://müller.de/straße",
"https://xn--mller-kva.de/",
"https://example.com/%C3%BC%20x.pdf",
+ "https://example.com/a\\b",
],
)
def test_to_item_keeps_valid_http_urls(url):
@@ -296,6 +308,10 @@ def test_to_item_keeps_valid_http_urls(url):
{"url": "https://example.com:0/"},
{"url": "https://example.com:\u0664\u0664\u0663/"},
{"url": "https://a [email protected]/"},
+ {"url": "https://[v1.x]/"},
+ {"url": "https://[::1%25eth0]/"},
+ {"url": "https://a.com\\@evil.com/"},
+ {"url": "https://evil.com\\.example.com/"},
],
)
def test_to_item_without_usable_url_is_none(hit):
diff --git a/atomic-forge/tools/firecrawl_search/tool/firecrawl_search.py b/atomic-forge/tools/firecrawl_search/tool/firecrawl_search.py
index 32da97e..0f9d22f 100644
--- a/atomic-forge/tools/firecrawl_search/tool/firecrawl_search.py
+++ b/atomic-forge/tools/firecrawl_search/tool/firecrawl_search.py
@@ -113,9 +113,10 @@ class FirecrawlSearchTool(BaseTool[FirecrawlSearchToolInputSchema, FirecrawlSear
KEY_PATTERN = re.compile(r"\bfc-[0-9a-f]{32}\b", re.IGNORECASE)
MIN_REDACTED_KEY_LENGTH = 8
# Shape of a usable result URL: http(s), optional userinfo, a host that starts with a letter or digit (or an
- # IPv6 literal), an ASCII port, and no whitespace anywhere. urlsplit then validates the port range and IPv6.
+ # IPv6 literal), an ASCII port, no whitespace anywhere, and no backslash before the path (parsers disagree on
+ # what host "a.com\@b.com" means). urlsplit then validates the port range and the IPv6 literal.
RESULT_URL_PATTERN = re.compile(
- r"https?://(?:[^\s/?#@]*@)?(?:[^\W_][^\s/?#@:\[\]]*|\[[0-9a-f:.]+\])(?::[0-9]*)?(?:[/?#]\S*)?", re.IGNORECASE
+ r"https?://(?:[^\s/?#@\\]*@)?(?:[^\W_][^\s/?#@:\[\]\\]*|\[[0-9a-f:.]+\])(?::[0-9]*)?(?:[/?#]\S*)?", re.IGNORECASE
)
def __init__(self, config: FirecrawlSearchToolConfig = FirecrawlSearchToolConfig()):