The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill provides extensive documentation on escalating privileges, including abusing sudo wildcard parameter injection using fnmatch() and placing SUID binaries in user-writable directories to be copied by root cronjobs in references/linux-privesc.md. It also covers escaping restricted shells by exploiting eval context and parameter expansion in references/bashjails.md.
[COMMAND_EXECUTION]: Numerous files demonstrate direct execution of system commands. references/bashjails.md uses os.system() and bash built-ins like read to bypass environment restrictions. references/pyjails.md documents escaping Python sandboxes by traversing the MRO chain to access os._wrap_close and other executable classes.
[REMOTE_CODE_EXECUTION]: Several high-risk remote execution patterns are described. references/games-and-vms-3.md documents the use of marshal.loads() to deserialize and execute arbitrary Python code objects. references/linux-privesc.md details how to use PostgreSQL's COPY TO PROGRAM function to execute OS commands as the database user.
[DATA_EXFILTRATION]: The skill identifies methods for accessing and exfiltrating sensitive data. This includes using Apache error logs via the Include directive to read files like /root/root.txt in references/linux-privesc.md, and exfiltrating data via network sockets (sock.send) or DNS tunneling techniques in references/games-and-vms-3.md and references/dns.md.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it is designed to ingest and process untrusted external data (e.g., 'mystery' files via binwalk, IQ signals, and network packets) while maintaining high-privilege capabilities such as file system writing and arbitrary command execution across multiple scripting environments.

ctf-misc