The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructs the agent to execute commands requiring elevated privileges. Specifically, sudo modprobe nvidia_peermem is used to load the kernel module required for GPUDirect RDMA functionality, and sudo dmesg is suggested for debugging driver-level issues. While these are standard administrative tasks for DOCA and CUDA environment setup, they involve the use of sudo for system-level modifications.
[COMMAND_EXECUTION]: The skill utilizes several system commands for environment discovery and validation. These include nvidia-smi for querying GPU status, nvcc for compiler version checks, and pkg-config for library configuration and versioning. These tools are used locally within the developer's environment to ensure hardware and software compatibility.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data provided by the user, such as CUDA source code (.cu files), build manifests (meson.build, CMakeLists.txt), and runtime logs (dmesg output).
Ingestion points: User-supplied source code, build configuration files, and system/application logs processed during the configure, modify, and debug tasks.
Boundary markers: Absent. The instructions do not specify the use of delimiters or safety warnings to ignore instructions embedded within the data being analyzed.
Capability inventory: The skill uses subprocess calls to nvidia-smi, nvcc, pkg-config, and modprobe, and performs file-system read operations to locate samples and version information.
Sanitization: Absent. There is no mention of sanitizing, escaping, or validating the external content before it is interpolated into the agent's context.

doca-gpunetio