The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads an external script (linpeas.sh) from a non-trusted GitHub repository (github.com/carlospolop) and executes it with elevated permissions within the Kali environment. It also instructs the agent to download and run arbitrary exploits from user-provided URLs.
[CREDENTIALS_UNSAFE]: Instructions are provided for the agent to locate and read sensitive credential files, including the system shadow file (/etc/shadow), private SSH keys (id_rsa), and configuration files that may contain hardcoded passwords.
[COMMAND_EXECUTION]: The skill makes extensive use of the kali-exec tool to perform arbitrary shell command execution, including package management, service manipulation, and execution of custom scripts like auto_enum.sh.
[PRIVILEGE_ESCALATION]: The skill methodology includes specific techniques for escalating privileges to root, such as targeting SUID binaries, kernel exploits, and using LD_PRELOAD to bypass sudo restrictions.
[DATA_EXFILTRATION]: The workflow includes steps to harvest and exfiltrate sensitive data, including password hashes, system configuration details, and 'flags' from the target environment.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes unvalidated data from network services (e.g., service banners, HTTP headers, and web page content) via tools like curl, nmap, and nikto. Malicious content from a target system could be used to influence the agent's behavior.
Ingestion points: Command outputs from nmap, curl, nikto, and gobuster are read into the agent context in SKILL.md and scripts/auto_enum.sh.
Boundary markers: No specific delimiters or warnings are used to isolate tool output from the agent's instructions.
Capability inventory: The agent has access to kali-exec (shell access), network tools (curl, nc), and file system write access.
Sanitization: There is no evidence of filtering or sanitization of external service data before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: Multiple external resources and tools are downloaded at runtime, including scripts from GitHub and various security tools via apt-get.
[OBFUSCATION]: The reference files contain various obfuscated payloads used for testing, including URL-encoded command injection strings and hexadecimal-encoded IP addresses to bypass security filters.
[DYNAMIC_EXECUTION]: The skill dynamically generates and executes scripts at runtime, such as the Python-based port scanner created in the /tmp directory.

penetration-testing