The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the Bash(*) tool to execute local commands via the vastai CLI and remote commands via ssh. This allows for broad system interaction necessary for cloud infrastructure management.\n- [COMMAND_EXECUTION]: The skill uses ssh -o StrictHostKeyChecking=no to connect to remote instances. This configuration disables host key verification, which facilitates connections to dynamic IP addresses but increases susceptibility to man-in-the-middle (MITM) attacks.\n- [PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface by ingesting untrusted local data to parameterize shell commands.\n
Ingestion points: Reads task requirements and hardware hints from refine-logs/EXPERIMENT_PLAN.md and local scripts (.py, .yaml, .json) during the Provision action.\n
Boundary markers: Absent. The skill interprets content from these files as authoritative instructions for configuring GPU searches and setup routines.\n
Capability inventory: Uses Bash(*) for local CLI execution, ssh for remote command execution, and rsync for file synchronization.\n
Sanitization: Absent. Extracted values are interpolated directly into shell command templates (e.g., in vastai search offers).

vast-gpu