The Agent Skills Directory

[COMMAND_EXECUTION]: The script uses subprocess.run to call the surf CLI tool for browser automation tasks, including navigation, tab listing, and JavaScript execution.
[EXTERNAL_DOWNLOADS]: The requests library is used to download remote image assets to inline them as base64 data URIs.
[PROMPT_INJECTION]: The skill processes untrusted data from local HTML files and URLs, presenting a risk of indirect prompt injection.
Ingestion points: Untrusted content enters the skill in html_to_share.py through file reading and surf-mediated URL fetching.
Boundary markers: There are no explicit boundary markers or warnings provided to the agent to treat the processed content as untrusted data.
Capability inventory: The skill can execute the surf CLI and write files to the user's Desktop and cache directories.
Sanitization: Content is cleaned via regex to remove tracking and navigation elements and processed through a Markdown converter, but no robust security sanitization is applied.

share