The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied data such as dates, model names, and seat IDs to construct CLI commands. This represents a potential attack surface for prompt injection.
Ingestion points: User queries are processed in SKILL.md to derive parameters for CLI execution.
Boundary markers: The instructions mandate reading shared security rules and reference files before command execution to ensure proper context.
Capability inventory: The skill uses the arkcli binary to perform network-based queries for usage and billing data.
Sanitization: The skill explicitly directs the agent to use safe API assembly methods rather than manual string concatenation and enforces strict validation for input formats like dates (YYYY-MM-DD).
[DATA_EXPOSURE]: The skill is designed to retrieve and display usage statistics, account identifiers, and team seat information.
Evidence: Commands like usage stats and usage plan access AccountID, AuthToken (masked), and user_id.
Mitigation: The instructions prioritize the --mine flag to restrict queries to the current user's resources and require authentication verification via arkcli auth whoami to prevent unauthorized cross-tenant or cross-user data access.

arkcli-usage