The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from external URLs, creating an attack surface where malicious instructions in web pages could influence agent behavior. Content is ingested via the AsyncWebCrawler interface in several scripts including basic_crawler.py and batch_crawl.py. The skill lacks explicit security boundary markers for the returned scraped data. While it includes relevance-based filters like BM25ContentFilter, these do not provide security sanitization. The execution environment maintains capabilities for network access and file system writing.
[PRIVILEGE_ESCALATION]: The references/anti-detection.md file provides guidance for installing patchright using sudo to manage OS-level dependencies. Although this is a standard requirement for some scraping engines, it involves using elevated privileges.
[COMMAND_EXECUTION]: The tests/run_all_tests.py script uses subprocess.run to execute integration test files located within the project directory. This is a standard development practice but involves the execution of local scripts.

crawl4ai