The Agent Skills Directory

[COMMAND_EXECUTION]: The skill launches a local Google Chrome or Chromium binary via subprocess.Popen in scripts/web_fetch.py. This is the primary mechanism for page rendering. The implementation uses list-based command construction without shell=True, which is a secure practice that prevents shell injection attacks.
[DATA_EXFILTRATION]: The skill performs network operations to fetch content from external URLs. This is the intended purpose of the tool. It also communicates over a local loopback connection (127.0.0.1) to control the browser instance via a custom WebSocket implementation.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection because it retrieves and processes untrusted HTML content from the web.
Ingestion points: Data enters the system via the fetch_page function in scripts/web_fetch.py.
Boundary markers: The output is clearly delimited by format (JSON or text separators) as documented in SKILL.md.
Capability inventory: The script is focused on data retrieval and does not possess capabilities for file modification or remote code execution on the host.
Sanitization: Content is cleaned by removing <script>, <style>, and comment tags using regular expressions or the trafilatura library if available.

web-fetch