The Agent Skills Directory

[SAFE]: The skill is authored by NVIDIA, a trusted organization, and all external references target official NVIDIA domains (docs.nvidia.com) or repositories. The provided artifacts, including a cryptographic signature (skill.oms.sig), indicate a secure development and distribution process.
[COMMAND_EXECUTION]: The skill includes shell commands for building applications (meson, pkg-config) and diagnosing device capabilities (doca_caps). These commands are standard for the DOCA SDK and are used strictly for their intended technical purpose.
[PRIVILEGE_ESCALATION]: A command using sudo (dmesg | tail -n 40) is provided for troubleshooting hardware and driver issues. This is a common and appropriate diagnostic step in the context of high-performance networking and hardware-acceleration development.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface common to agent skills that ingest user-provided configuration and code.
Ingestion points: User-supplied parameters for task configuration and error messages from the DOCA runtime are processed via SKILL.md and TASKS.md.
Boundary markers: The skill mitigates risks by requiring explicit capability discovery and round-trip smoke tests before processing any bulk data.
Capability inventory: The skill enables the agent to help users build binaries, execute diagnostic tools, and perform local compression fixtures using Python's standard library.
Sanitization: The instructions emphasize byte-for-byte comparison and manual verification of accelerator outputs, ensuring that processing logic is validated by the user.

doca-compress