hsm-attack
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill makes extensive use of powerful security utilities, including
pkcs11-tool,yubihsm-shell,hashcat, andghidra, to interact with hardware modules and network services. These tools are used for reconnaissance and active exploitation within the user's environment. - [REMOTE_CODE_EXECUTION]: Provides detailed Python templates for exploiting remote targets, such as
kali_luna_rce.py, which targets a pre-authentication buffer overflow in Thales Luna HSMs (CVE-2024-47787). The code handles binary protocol construction and payload delivery. - [DATA_EXFILTRATION]: Describes advanced techniques for extracting protected cryptographic keys from HSMs, specifically using the
C_WrapKeyPKCS#11 operation to wrap sensitive keys with an attacker-controlled public key for offline decryption. These methods are designed to move secret data from the hardware module to the host filesystem. - [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: The skill is designed to process output from network scanners (
nmap), cryptographic toolkits (pkcs11-tool), and local configuration files like/etc/Chrystoki.confand/var/luna/.ped. - Boundary markers: There are no protective delimiters or instructions provided to the agent to treat data from these external sources as untrusted.
- Capability inventory: The agent is granted capabilities for network communication (
socket,curl,nc), file system read/write access, and execution of arbitrary code via the Python interpreter. - Sanitization: The skill lacks logic to validate or sanitize tool outputs before the agent incorporates that information into its operational context, creating a vulnerability to poisoned tool responses.
Audit Metadata