The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download external tools from third-party GitHub repositories that are not within the recognized trusted vendor list.
Fetches the vncpwd tool from https://github.com/jeroennijhof/vncpwd in references/vnc-techniques.md and SKILL.md.
References a Python decryption script from https://github.com/trinitronx/vncpasswd.py in SKILL.md.
[COMMAND_EXECUTION]: The skill makes extensive use of system commands and network auditing tools.
Executes nmap with scripts like realvnc-auth-bypass to probe remote targets.
Uses password cracking tools including hydra, medusa, and ncrack.
Orchestrates Metasploit modules (vnc_none_auth, vnc_login) for automated scanning.
Employs vncviewer for remote desktop connections and xdotool for automated input injection.
[DYNAMIC_EXECUTION]: The skill involves runtime compilation and dynamic code execution.
Uses make to compile the vncpwd tool from source code after cloning the repository in SKILL.md and references/vnc-techniques.md.
Executes a Python one-liner in references/vnc-techniques.md to perform 3DES decryption using the pyDes library.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from remote VNC servers, creating an attack surface for indirect prompt injection.
Ingestion points: Reads VNC desktop titles using the Nmap vnc-title script, which processes arbitrary strings provided by the target server.
Boundary markers: None identified in the instructions for handling tool outputs.
Capability inventory: Includes the ability to execute shell commands, perform network operations, and write to the local file system.
Sanitization: There is no explicit sanitization or validation of the version strings or titles retrieved from the remote services before they are processed by the agent.

vnc-pentesting