The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of the local arkcli binary to perform administrative and billing tasks. These include purchasing plans (plans buy), renewing subscriptions (plans renew), and managing team seats (plans team seat-assign).
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes data from remote Volcengine APIs via command outputs (e.g., plans get, model-list). This untrusted external data is used to drive the agent's logic for selecting plans and managing resources, creating a theoretical attack surface for indirect prompt injection.
Ingestion points: API responses processed by arkcli plans get, arkcli plans model-list, and arkcli plans team seat-list as described in SKILL.md and various reference files.
Boundary markers: The instructions mandate reading ../arkcli-shared/SKILL.md for shared security rules and implement a strict "agreement gate" that prevents execution of sensitive commands without explicit user consent.
Capability inventory: The skill can perform billing operations (plans buy, plans renew), credential rotation (rotate-apikey), and file system reads of local agent configurations (harness-status).
Sanitization: The skill enforces a procedural safety gate (the "Agreement Gate") requiring a multi-step confirmation process before allowing billing flags like --yes to be used.

arkcli-plans