The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation advises writing scraping logic to standalone Python scripts and executing them using system shell tools (Bash) for iterative development.\n- [INDIRECT_PROMPT_INJECTION]: As a web-scraping tool, the skill ingests untrusted content from the internet. It lacks explicit instructions for the agent to use boundary markers when processing this data, though it does include libraries for sanitizing HTML into markdown.\n
Ingestion points: Content is retrieved via scrape_page, AsyncFetcher, and Spider classes defined in SKILL.md and references/api-reference.md.\n
Boundary markers: None identified in the prompt examples.\n
Capability inventory: The environment supports network requests, file writing (to_json), and shell command execution (Bash).\n
Sanitization: Employs html_to_markdown and trafilatura for content cleaning.\n- [DATA_EXFILTRATION]: The skill permits network connections to any URL, which is the core functionality for scraping but inherently allows data to be sent to external endpoints.\n- [DYNAMIC_EXECUTION]: Supports runtime script generation and execution via ExecuteCode and Bash to handle complex scraping and crawling tasks.

web-scraping