The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/setup_check.py script invokes shell commands using subprocess.run to manage the installation environment, including calls to pip, python -m cloakbrowser, and the macOS xattr utility.
[EXTERNAL_DOWNLOADS]: The installation workflow (triggered by setup_check.py --install) fetches a third-party patched Chromium binary approximately 200 MB in size from a remote vendor server outside of standard package registries.
[REMOTE_CODE_EXECUTION]: The skill downloads and executes a third-party, closed-source binary fork of Chromium. This execution occurs during both the environment verification phase and the primary operation of the toolchain.
[PRIVILEGE_ESCALATION]: The setup script includes a function to recursively clear the macOS Gatekeeper quarantine flag (xattr -cr) on the downloaded application bundle. This action deliberately bypasses system-level security protections intended to prevent the execution of unverified third-party software.
[DYNAMIC_EXECUTION]: The toolchain is designed to run Chromium with the --no-sandbox flag enabled. Disabling the browser sandbox removes a critical security isolation layer, meaning that any vulnerability encountered during browsing could potentially result in a direct compromise of the host system.
[INDIRECT_PROMPT_INJECTION]: The skill processes structured diagnostic data from the cloakbrowser info --json command to drive its internal logic and reporting. This represents a potential attack surface where external tool output could influence the agent's decision-making process.
Ingestion points: The probe and cloakbrowser_info functions in scripts/setup_check.py parse the JSON output of an external command.
Boundary markers: None identified; the agent is instructed to report the diagnostic values (Chromium version and license tier) directly to the user.
Capability inventory: The skill has the ability to perform network downloads, execute binaries, and modify file system metadata.
Sanitization: The script uses json.loads to handle the command output but lacks validation for the integrity or safety of the content within the JSON fields.

playwright-stealth-setup