Fix: Prevent URL malformation causing 'getaddrinfo ENOTFOUND http' errors

[ldemesla]

Problem

Fixed a critical bug in extract_host_port() that caused malformed URLs like http://http://localhost:3000 when the target already contained a protocol scheme.

The extract_host_port() function was using naive string splitting on : which incorrectly parsed URLs:

Example of the bug:

• Input: http://localhost:3000
• Old behavior: Returns ("http://localhost", 3000) ❌
• URL reconstruction: f"http://{host}:{port}/login"
• Result: http://http://localhost:3000/login ❌
• Error: getaddrinfo ENOTFOUND http

This manifested on macOS (and potentially Linux) as DNS resolution errors when the agent attempted HTTP requests.

Solution

Rewrote extract_host_port() to use urllib.parse.urlparse which properly handles URL parsing:

• Input: http://localhost:3000
• New behavior: Returns ("localhost", 3000) ✅
• URL reconstruction: f"http://{host}:{port}/login"
• Result: http://localhost:3000/login ✅

Changes

• Fixed extract_host_port() in http_parser.py to use urlparse
• Added comprehensive unit tests (15 tests, all passing)
• Verified fix with Juice Shop server on localhost:3000

Testing

uv run pytest deadend_agent/tests/deadend_sdk/tools/browser_automation/test_extract_host_port.py -v

All 15 tests pass, including critical test for protocol duplication prevention.

Impact

This fix resolves the issue where agents running on macOS (and potentially Linux) were unable to perform HTTP requests due to malformed URLs.

🤖 Generated with Claude Code