The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the download of several third-party security tools from GitHub repositories not belonging to the author or trusted organizations. These include LinPEAS, pspy, Linux Exploit Suggester, and deepce.\n
[REMOTE_CODE_EXECUTION]: Multiple instructions involve downloading remote scripts and executing them directly in the shell, such as curl -L [URL] | sh for LinPEAS in references/linux-privesc.md and similar patterns for deepce in references/container-breakout.md.\n
[COMMAND_EXECUTION]: The references/shells-and-pivoting.md file contains a large collection of reverse shell one-liners using bash, python, perl, netcat, php, ruby, and node.js, designed to establish remote control over a target system.\n
[DATA_EXFILTRATION]: The skill provides explicit instructions for harvesting sensitive information, including reading /etc/shadow, SSH private keys (id_rsa, id_ed25519), AWS credentials (~/.aws/credentials), and VPN configurations (/etc/wireguard/*.conf). It further demonstrates how to exfiltrate this data using curl --post-file or nc.\n
[PRIVILEGE_ESCALATION]: The core of the skill focuses on escalating privileges through SUID/SGID binary abuse, sudo misconfigurations (via GTFOBins references), kernel exploits, and Linux capabilities abuse.\n
[PERSISTENCE]: Instructions include techniques for maintaining access across sessions, such as adding attacker-controlled SSH keys to authorized_keys and creating malicious cron jobs that execute as root.\n
[DYNAMIC_EXECUTION]: The skill provides C source code for LD_PRELOAD exploitation and kernel exploits, along with instructions to compile them at runtime using gcc on the target system.\n
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (process listings, network scan results, file contents) which could be used to influence agent behavior, although no specific sanitization or boundary markers are documented for these ingestion points.

privilege-escalation