The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the arkcli binary, which is a specialized tool provided by the vendor (Volcengine) for managing cloud resources. The commands are scoped to inference endpoint management.
[PRIVILEGE_ESCALATION]: The skill documentation explicitly describes a "three-branch confirmation gate" for deletion operations. It prevents AI agents from performing silent deletions by requiring either a real TTY interaction or an explicit environment variable ARKCLI_ALLOW_HEADLESS_DELETE=1 that must be set outside the normal automated flow, preventing unauthorized destructive actions by the AI.
[DATA_EXFILTRATION]: The skill uses --mine and --page-all flags to list resources associated with the current user. Access is restricted to authenticated SSO sub-users, and the results remain within the local CLI context unless explicitly redirected by the user.
[PROMPT_INJECTION]: The skill contains internal instructions (e.g., "Anti-trigger (Important)") to ensure the agent routes users to the correct deployment workflow (+deploy) rather than raw CRUD commands when appropriate. These are legitimate behavioral constraints for the agent and do not constitute malicious injection.

arkcli-infer-endpoint