The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes commands on a remote devkit via SSH. This is the primary intended function of the skill for hardware management. All sensitive commands, such as docker run with --privileged and sudo, are documented and require explicit user confirmation through phase gates.
[EXTERNAL_DOWNLOADS]: The skill interacts with Docker containers and remote repositories. These operations are restricted to the user's own devkit environment and the official NVIDIA 'hololink-demo' container, which is consistent with the skill's purpose for NVIDIA hardware development.
[PROMPT_INJECTION]: The skill uses clear phase gates (Gate 1, Gate 2, Gate 3) and mandatory interaction patterns that ensure the AI agent remains within the defined hardware management workflow. It includes specific rules for handling user input to prevent unauthorized behavior.
[DATA_EXFILTRATION]: While the skill reads remote files (VERSION, READMEs, logs) and FPGA metadata, this data remains within the user's controlled session for the purpose of application selection and debugging. There are no patterns of sending data to unauthorized external domains.
[SAFE]: The skill includes robust error handling and failure analysis guidelines. It explicitly prevents unsafe automated remediations, such as automatically adding --headless or suggesting kernel/driver modifications, ensuring the user maintains control over the hardware state.

hsb-app