vpn-attack
Fail
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: CRITICALCREDENTIALS_UNSAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill instructs the agent to use
sudoto access protected system locations for reconnaissance and credential harvesting, including/etc/NetworkManager/system-connections/,/proc/entries for environment variables, and macOS keychain resources. Evidence:sudo cat /etc/NetworkManager/system-connections/*.vpnandsudo wg show all.\n- [CREDENTIALS_UNSAFE]: The skill provides procedures for extracting sensitive data such as pre-shared keys (PSK), session cookies (APSCOOKIE, DSID), and private keys from local files and memory. Evidence: harvesting cookies viatsharkfrom captured traffic and extracting keys from WireGuard and OpenVPN configuration files. \n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection. 1. Ingestion points: The agent ingests untrusted data from network service responses (ike-scan, curl, nmap, tshark) as described in the methodology and guides. 2. Boundary markers: Absent; there are no instructions to use delimiters or ignore embedded instructions when processing output from VPN gateways. 3. Capability inventory: The skill has access to Bash, Write, Edit, and Agent tools, allowing it to modify files or execute system commands based on external data. 4. Sanitization: Absent; the skill does not provide mechanisms for validating or escaping content retrieved from external VPN portals or handshake responses.\n- [COMMAND_EXECUTION]: The skill involves the dynamic creation of configuration files in temporary directories (e.g.,/tmp/test_vpn.conf) which are subsequently executed by system tools likevpncandstrongswan. This involves the interpolation of potentially untrusted cracked credentials into shell commands.
Recommendations
- Contains 6 malicious URL(s) - DO NOT USE
Audit Metadata