The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill provides detailed methodology for escalating privileges on both Windows and Linux systems.
Evidence includes instructions for exploiting 'Potato' vulnerabilities (GodPotato, PrintSpoofer), service path hijacking, and SUID bit abuse in references/phases.md.
[PERSISTENCE]: Comprehensive instructions are provided for establishing long-term access to compromised systems.
Evidence in references/phases.md section 5 includes modifying shell profiles (.bashrc), creating cron jobs, installing systemd services, and adding unauthorized SSH keys to authorized_keys.
[DATA_EXFILTRATION]: The skill outlines techniques for harvesting sensitive data and credentials.
Evidence includes using tools like Mimikatz for memory dumping, harvesting NTLM hashes via NTLM Relay, and accessing sensitive local files like /etc/shadow or cloud metadata via SSRF in references/attack-playbooks.md and references/phases.md.
[COMMAND_EXECUTION]: Provides an extensive list of commands for network-capable and exploitation tools.
Evidence includes specific CLI strings for nmap, sqlmap, nuclei, and various 'Living Off the Land' binaries (LOLBins) like certutil, mshta, and bitsadmin used to download and execute remote payloads in references/evasion-cheatsheet.md.
[DYNAMIC_EXECUTION]: Describes advanced techniques for executing code in memory to bypass security software (EDR/AV).
Evidence includes instructions for direct system calls, process injection into low-monitoring targets (e.g., RuntimeBroker.exe), module stomping, and sleep encryption in references/evasion-cheatsheet.md.
[CREDENTIALS_UNSAFE]: Mentions the extraction of NTLM hashes, TGTs, and secrets from memory or configuration files using specialized tools like mimikatz and secretsdump.py.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon untrusted user input (target URLs, IPs, and scopes).
Ingestion points: Target parameters provided by the user during the 'confirmation gate' in SKILL.md.
Boundary markers: Includes a mandatory confirmation gate requiring explicit user permission before execution.
Capability inventory: Significant capabilities including remote command execution, file system manipulation, and network operations across all referenced scripts.
Sanitization: No explicit evidence of input sanitization or escaping is provided in the orchestration instructions.

attack-chain