The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents workflows for ingesting and applying sandbox policy files in YAML format.
Ingestion points: The skill uses openshell policy set and openshell sandbox create with the --policy flag to ingest user-authored YAML files.
Boundary markers: The CLI utilizes gateway-side validation for policy candidates and supports manual approval modes for agent-authored rules.
Capability inventory: The skill includes commands for remote process execution (sandbox exec), file synchronization (upload/download), and network port forwarding.
Sanitization: Policy updates are validated by the gateway supervisor before persistence, and enforcement modes (audit vs. enforce) are documented.
[EXTERNAL_DOWNLOADS]: The skill provides links to official installation guides and product documentation hosted on nvidia.com domains.
[COMMAND_EXECUTION]: The instructions include standard developer operations such as configuring SSH access by modifying the local ~/.ssh/config file.
[SAFE]: The skill follows secure credential management practices by recommending environment variable references (--credential KEY) instead of hardcoding sensitive tokens or API keys in shell command history.

openshell-cli