The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill provides instructions that involve executing commands with root privileges. The Command Appendix in TASKS.md suggests using sudo dmesg | tail -n 40 to inspect kernel logs for driver-level troubleshooting.
[COMMAND_EXECUTION]: The skill workflow requires the execution of multiple shell commands. These include pkg-config for retrieving compilation and linking flags, doca_caps for querying hardware device capabilities, and executing user-compiled binaries during the testing phase.
[INDIRECT_PROMPT_INJECTION]: The skill includes instructions for handling and analyzing external data in the form of cryptographic payloads.
Ingestion points: The skill facilitates the processing of ciphertext and Additional Authenticated Data (AAD) through the doca_aes_gcm_task_decrypt task.
Boundary markers: Documentation explicitly defines the authentication tag as a security boundary and instructs the agent to treat verification failures as a critical security signal (poisoned output).
Capability inventory: The agent has the capability to execute build tools (via pkg-config), perform hardware capability queries (via doca_caps), and read system diagnostics.
Sanitization: The security model relies on the underlying cryptographic library's tag verification mechanisms rather than input sanitization or escaping within the agent instructions.

doca-aes-gcm