The Agent Skills Directory

[DYNAMIC_EXECUTION] (MEDIUM): The setup_webctl.py script performs a 'hot-patch' on the installed webctl library. It copies auth_proxy.py into the library's internal directory and programmatically modifies session_manager.py to import and execute the new module. Modifying library code at runtime is a sensitive operation that can introduce stability or security risks if the source is untrusted.
[COMMAND_EXECUTION] (LOW): The skill utilizes subprocess.run(shell=True) within setup_webctl.py to automate the installation and configuration of the environment. This includes running pip install and the webctl CLI. This is consistent with the skill's stated purpose of setting up a browser automation environment.
[EXTERNAL_DOWNLOADS] (LOW): The setup process involves downloading the webctl package from PyPI and subsequently using webctl setup to download the Chromium browser binary. These are standard external dependencies for this functional use case.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill provides the ability to browse and extract content from arbitrary websites via webctl snapshot and webctl query. This creates an attack surface where malicious web content could attempt to influence the agent's logic.
Ingestion points: webctl snapshot text output and webctl query results processed by the agent.
Boundary markers: None explicitly implemented within the skill's Python scripts; relies on the agent's internal handling.
Capability inventory: Includes file system write access (patching), network access (browsing), and command execution capabilities.
Sanitization: No explicit sanitization or filtering of web content is performed by the scripts before passing data to the agent.

using-webctl