The Agent Skills Directory

[DATA_EXFILTRATION]: The skill contains comprehensive logic for harvesting sensitive authentication cookies from multiple web browsers.
scripts/lib/chrome_cookies.py copies local database files and attempts to decrypt session tokens using keys retrieved from the macOS Keychain.
scripts/lib/safari_cookies.py extracts cookies from the Safari binary cookie file.
scripts/lib/cookie_extract.py provides cross-platform discovery for Firefox browser profiles, including support for Windows profiles accessed via WSL.
This capability allows the skill to capture sensitive session tokens that could provide unauthorized access to user accounts.
[COMMAND_EXECUTION]: The skill relies extensively on the execution of external binaries and system utilities via subprocesses.
It executes security find-generic-password to access the macOS Keychain and openssl for cookie decryption.
It uses node to execute a vendored JavaScript search client for X/Twitter data retrieval.
It invokes gh auth token to retrieve GitHub credentials and yt-dlp for video metadata collection.
[EXTERNAL_DOWNLOADS]: The skill contains logic to modify the user's system environment by downloading external dependencies.
scripts/lib/setup_wizard.py automatically executes brew install yt-dlp if the YouTube downloader is missing from the system path.
[PROMPT_INJECTION]: SKILL.md uses adversarial framing, such as 'LAWS' and 'CONTRACTS', to override the default safety guidelines and operational protocols of the AI agent.
The instructions explicitly command the agent to disregard requirements from other tools, such as the mandatory citation format for the WebSearch tool.
The skill processes untrusted third-party content from social media platforms, creating a vulnerability surface for indirect prompt injection.
[CREDENTIALS_UNSAFE]: The skill accesses highly sensitive credential stores and hardcodes authentication tokens.
It uses the security utility to retrieve the Chrome Safe Storage key from the system keychain.
scripts/lib/vendor/bird-search/lib/twitter-client-base.js contains a hardcoded Twitter bearer token.
The skill also reads from various local credential files, including ~/.codex/auth.json.

last30days