The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages cloud resources by executing various subcommands of the arkcli utility, such as create, list, get, start, stop, delete, and update. All commands are within the expected scope of a CLI-based management skill.
[INDIRECT_PROMPT_INJECTION]: The skill processes data returned from external services, such as endpoint listings and model search results, which represents a potential attack surface for indirect prompt injection if the service output were to contain malicious instructions.
Ingestion points: JSON output from arkcli infer endpoint list and arkcli models search (in SKILL.md and references/arkcli-infer-endpoint-list.md).
Boundary markers: None explicitly defined, but instructions rely on structured JSON parsing rather than natural language interpretation of the raw output.
Capability inventory: Includes file system writes (via shell redirects), multiple CLI command executions, and local script execution.
Sanitization: Employs jq to select specific fields and filter data, which inherently reduces the risk of accidental instruction execution.
[DYNAMIC_EXECUTION]: The skill provides an example of generating and executing a temporary Node.js script to process and format JSON reports. This is used for complex data manipulation (date conversions) that is difficult to achieve with standard shell tools, and the script logic is static and local to the reporting task.

arkcli-infer-endpoint