The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains multiple shell command templates for fingerprinting and exploiting VPN appliances. These include curl commands for path traversal and file disclosure, as well as bash loops for group enumeration. Notably, one payload template (test_$(id)_test.txt) includes a subshell command that would execute locally if run in a standard bash environment, potentially exposing local system information during a remote scan.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through variable interpolation in shell scripts.
Ingestion points: Hostnames and group names are ingested into variables like TARGET and group used in shell commands throughout SKILL.md.
Boundary markers: No delimiters or safety instructions are provided to distinguish between untrusted input data and command logic.
Capability inventory: The skill requires access to shell execution, network utilities (curl), and Python.
Sanitization: There is no evidence of input validation or escaping for the user-supplied strings before they are passed to the shell, which could allow a malicious target name to execute arbitrary local commands.
[DYNAMIC_EXECUTION]: The skill utilizes python3 -c to generate large-scale buffer or memory disclosure payloads at runtime, specifically for the Citrix Bleed (CVE-2023-4966) exploit sequence.

enterprise-vpn-attack