The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of over 269 Kali Linux CLI tools for reconnaissance, scanning, and system assessment. These commands are executed across local, remote (SSH), and containerized (Docker) environments.
[REMOTE_CODE_EXECUTION]: Contains instructions for the autonomous exploitation of remote targets using frameworks like Metasploit (msfconsole) and sqlmap. It specifically demonstrates executing OGNL injection and SQL injection attacks to gain access to remote systems.
[CREDENTIALS_UNSAFE]: The skill setup and operation involve accessing and managing sensitive SSH private keys (~/.ssh/kali_key) to authenticate with remote Kali servers, posing a risk of credential exposure if the environment is not secured.
[DATA_EXFILTRATION]: Includes post-exploitation workflows designed to dump credentials and sensitive information using tools like mimikatz, impacket-secretsdump, and crackmapexec (extracting SAM hashes).
[EXTERNAL_DOWNLOADS]: Orchestrates the download of external resources including the official Kali Linux Docker image (kalilinux/kali-rolling) and numerous security packages via apt install during environment initialization.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because the agent processes unvalidated output from external tools (nmap scans, web vulnerability results, etc.). Malicious content on a target system could potentially influence the agent's decision-making logic.
Ingestion points: Processing output from various scanners and enumeration tools as seen in the execution phases (SKILL.md).
Boundary markers: None implemented to distinguish between trusted instructions and untrusted data retrieved from targets.
Capability inventory: Significant capability to execute shell commands, write files, and perform network operations across multiple environments.
Sanitization: No evidence of sanitization or filtering of external data before it is interpreted by the agent's logic engine.

kali-pentest-ai-agent