The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains extensive logic dedicated to harvesting sensitive credentials from the local system.
scripts/lib/chrome_cookies.py executes system commands to retrieve encryption keys from the macOS Keychain (security find-generic-password -w -s "Chrome Safe Storage").
scripts/lib/safari_cookies.py and scripts/lib/cookie_extract.py programmatically read and parse local browser database files (Cookies.binarycookies and cookies.sqlite) to extract session tokens for services like X/Twitter and Truth Social.
These harvested tokens are then used to authenticate external API requests without the user's explicit manual entry of the secrets.
[PROMPT_INJECTION]: The SKILL.md instructions (specifically 'LAW 1' and 'LAW 1 OVERRIDE') explicitly command the agent to ignore and override the mandatory output requirements of the platform's WebSearch tool. This is a direct attempt to bypass system-level constraints and safety reminders provided by the environment.
[COMMAND_EXECUTION]: The skill relies on a complex engine that executes numerous shell commands through Python's subprocess and Popen modules. This includes running system utilities (openssl, security), package managers (brew), and Node.js scripts to perform its operations.
[DATA_EXFILTRATION]: The skill transmits data to several third-party endpoints. One of these, https://xquik.com/api/v1 (used in scripts/lib/xquik.py), has been flagged by automated scanners as a phishing threat. The combination of local credential harvesting and transmission to flagged domains represents a significant exfiltration risk.
[EXTERNAL_DOWNLOADS]: The skill's setup process (scripts/lib/setup_wizard.py) attempts to automatically download and install software (yt-dlp) using brew install if it is missing from the user's system.

last30days