vast-gpu

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill contains explicit commands and parameters to create and destroy paid vast.ai instances (e.g., vastai create instance and vastai destroy instance) and instructs setting a vast.ai API key. Those actions directly start and stop billed contracts on the vast.ai marketplace and can incur charges (the skill even reports estimated/actual cost and supports a max_budget field). This grants the agent the ability to execute actions that result in real financial charges, i.e., direct financial execution authority.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (medium risk: 0.60). The skill explicitly instructs using ssh with StrictHostKeyChecking=no (bypassing host key verification) and runs root-level package installs in created instances (via an onstart-cmd), which encourages bypassing SSH security checks and executing privileged changes on remote machines even though it doesn't ask to create local users or escalate local sudo.