The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's core execution engine and Phase 1.3 (Hidden Systems) instructions in references/phase-details.md direct the agent to execute intensive system-level reconnaissance commands, including process listing (ps aux), network socket inspection (ss -tulpn), open file listing (lsof -i), and systemd unit auditing.
[DATA_EXFILTRATION]: The instructions in references/phase-details.md (Phase 1.3, 3, and 4) require the agent to read and analyze highly sensitive system files and credentials, including /etc/passwd, /etc/sudoers, SSH keys, bash history, .env files, and hardcoded secrets. While intended for auditing, this provides the agent with access to data that could be used for exfiltration.
[REMOTE_CODE_EXECUTION]: The 'Execution Engine' in SKILL.md contains a specific instruction (Step 7: EXPLOIT) that commands the agent to 'Attempt to exploit if vulnerability found'. This encourages the agent to generate and execute potentially malicious payloads (e.g., SQL injection, command injection, or path traversal) against the environment it is auditing.
[PROMPT_INJECTION]: The skill uses authoritative and restrictive language ('YOU MUST follow this loop... NO EXCEPTIONS', 'CRITICAL rules', 'NEVER skip') to enforce its complex logic and override default agent behaviors during the audit process.

ralph-promax