The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/scraper.py can perform network requests to arbitrary URLs and read local files via the --schema or batch commands. It also writes scraped data to arbitrary local paths using the --output flag. This combination creates a potential channel for exfiltrating sensitive local data to a remote server.\n- [COMMAND_EXECUTION]: The script allows the agent to read and write to any file path on the system provided via command-line arguments. This capability can be abused to access private information or overwrite important system configuration files.\n- [EXTERNAL_DOWNLOADS]: The skill retrieves data from external websites and uses the Crawl4AI library, which may download browser binaries (Playwright) to support JavaScript rendering during scraping operations.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data from the internet.\n
Ingestion points: Website content is fetched in scripts/scraper.py using httpx or Crawl4AI.\n
Boundary markers: No explicit markers are used to isolate or identify untrusted content before it is processed by the agent.\n
Capability inventory: The script can perform file system reads/writes and network communication.\n
Sanitization: The scraped content is not sanitized for malicious instructions before being returned to the agent context.

web-scraper