The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/identity.py dynamically calculates a file path to a sibling skill directory (playwright-stealth/scripts) and imports a Python module (launch.py) from that location using sys.path.insert. This runtime modification of the module search path is used to share logic between skills but introduces a dependency on external skill placement.
[DATA_EXFILTRATION]: The skill manages directories containing sensitive browser session data, including cookies, saved logins, and localStorage within the .stealth/<identity>/user-data/ path. While the skill mandates security practices such as updating .gitignore and setting restrictive file permissions (0700/0600), the access and manipulation of these credentials by the agent constitutes a sensitive data exposure surface.
[COMMAND_EXECUTION]: The workflow instructions in SKILL.md direct the agent to execute shell commands, specifically python3 -m cloakbrowser and scripts/identity.py, using arguments derived from user input. This includes the execution of a preflight check for proxies.
[INDIRECT_PROMPT_INJECTION]: The skill exposes a potential attack surface by ingesting untrusted data—such as proxy URLs, geography preferences, and humanization presets—from user prompts and interpolating them into shell commands.
Ingestion points: User-provided proxy strings (e.g., http://user:pass@host:port), country names, and humanization presets defined in the SKILL.md workflow.
Boundary markers: Command templates use double quotes for arguments, but there are no instructions to validate or sanitize these strings for shell metacharacters before execution.
Capability inventory: The agent utilizes python3 for script execution, writes to the local file system (creation of .stealth/ hierarchy), and performs network operations via the cloakbrowser utility.
Sanitization: While the identity.py script uses argparse, the initial interpolation by the agent into the shell command represents a point of potential misuse.
[EXTERNAL_DOWNLOADS]: The skill documentation refers to well-known IP echo services (ipify.org, checkip.amazonaws.com) and established proxy vendors (Bright Data, Oxylabs) for egress verification and network configuration. These are standard services for the skill's intended purpose.

playwright-stealth-identity