The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions specifically target high-value sensitive files and credential stores, including SSH private keys (id_rsa, *.pem, *.key), cloud credentials (~/.aws/credentials), Kubernetes configurations (~/.kube/config), database passwords in configuration files (.env, wp-config.php), and system shadow files (/etc/shadow).
[COMMAND_EXECUTION]: The methodology requires the agent to execute a broad range of system commands to gather situational awareness, including process monitoring (ps auxf, tasklist), network analysis (ss, netstat, netsh), and deep file searches using grep and find on system-wide paths.
[PRIVILEGE_ESCALATION]: The skill provides explicit instructions for identifying privilege escalation leads, such as checking sudo permissions (sudo -l), Windows user privileges (whoami /all, whoami /priv), and identifying misconfigured services or scheduled tasks that run with elevated permissions.
[DATA_EXFILTRATION]: The instructions direct the agent to collect discovered credentials, keyfiles, and sensitive data (referred to as "loot") into a specific directory (engagements/<target>/loot/) for further use in the attack chain.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface by ingesting and processing untrusted data from the target host (e.g., shell history files, application logs, and outputs from tools like linpeas).
Ingestion points: Host system files (/etc/passwd, ~/.bash_history, .env) and external tool outputs (linpeas.sh, winpeas, pspy).
Boundary markers: None present in the instructions to distinguish between benign data and potentially malicious embedded instructions.
Capability inventory: The agent has capabilities to execute arbitrary shell commands and write to the filesystem.
Sanitization: No filtering or sanitization of external content is specified before processing or following identified "leads."

privesc-enumeration