The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a vulnerability surface where it ingests data from external command outputs while maintaining potentially exploitable capabilities.
Ingestion points: Data enters the agent context through arkcli models search and arkcli resources list command outputs as specified in SKILL.md.
Boundary markers: The instructions lack explicit delimitation or warnings to ignore instructions that might be embedded in the ingested data.
Capability inventory: The skill performs billing-sensitive write operations via arkcli +deploy and accesses identity status via arkcli auth status.
Sanitization: No sanitization or validation logic is defined for the external tool output before it is processed or presented to the user.
Multi-step chaining: The skill employs routing logic that directs the agent to different instruction files (e.g., arkcli-infer-endpoint/SKILL.md) based on input parameters, which could be influenced by injected content.- [COMMAND_EXECUTION]: The skill executes shell commands via the arkcli tool to manage endpoints. It handles user-supplied flags for names and models, and processes structured JSON strings for rate limits and routing configurations.

arkcli-deploy