The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains modules specifically designed to harvest session cookies from local browser storage. scripts/lib/chrome_cookies.py and scripts/lib/safari_cookies.py access ~/Library/Application Support/Google/Chrome/Default/Cookies and ~/Library/Cookies/Cookies.binarycookies respectively. It also attempts to retrieve the encryption key for Chrome cookies from the macOS Keychain using the security command.
[COMMAND_EXECUTION]: The skill uses subprocess to execute several sensitive system commands, including security find-generic-password to access Keychain secrets, openssl to decrypt browser cookies, and brew install to install system utilities like yt-dlp. It also executes a bundled Node.js application (bird-search.mjs) to perform searches on X/Twitter.
[DATA_EXFILTRATION]: The skill extracts highly sensitive authentication data (browser session cookies) from the local filesystem. While these are primarily used to authenticate search requests to the platforms themselves, the logic transmits parameters derived from these credentials to various third-party APIs listed in the skill.
[PROMPT_INJECTION]: The SKILL.md file defines a series of "LAWS" and a "VOICE CONTRACT" designed to aggressively override the AI agent's default behaviors and tool instructions. For example, it explicitly commands the agent to ignore the WebSearch tool's mandatory citation requirements and to prioritize the skill's local formatting rules over any platform-wide or user-defined settings.
[EXTERNAL_DOWNLOADS]: The skill makes requests to numerous external API endpoints and includes functionality in scripts/lib/setup_wizard.py to automatically download and install software (yt-dlp) via the Homebrew package manager.

last30days