The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages the application lifecycle through the ./holohub wrapper script, executing commands for scaffolding, building, running, and testing within containerized environments.
[EXTERNAL_DOWNLOADS]: The workflow involves cloning the official HoloHub repository from GitHub (https://github.com/nvidia-holoscan/holohub). This targets a well-known service and a trusted organization, following secure practices by verifying the official remote and resolved commit hashes.
[PRIVILEGE_ESCALATION]: The instructions explicitly prohibit the use of sudo ./holohub and warn against broadening container privileges or changing file permissions without authorization.
[DATA_EXPOSURE]: The skill includes robust guidelines for protecting sensitive data, such as mounting input directories as read-only, keeping private data outside the Git checkout, and auditing environment diagnostic output before sharing it.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges the risk of processing untrusted repository content. It mitigates this by instructing the agent to treat all data, logs, and models as untrusted, using finite execution modes with bounded frame counts, and validating inputs against structured schemas.
[DYNAMIC_EXECUTION]: Host dependency installation is handled via ./holohub setup, but this is strictly limited to template scripts and requires explicit user review and authorization of the previewed actions before execution.

holohub-app-lifecycle