vast-gpu

Pass

Audited by Gen Agent Trust Hub on Apr 1, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to install the vastai Python package using pip to enable its core functionality.
  • [COMMAND_EXECUTION]: Extensively uses the Bash tool to execute local CLI commands for searching, creating, and managing cloud infrastructure via the vastai toolset.
  • [REMOTE_CODE_EXECUTION]: Establishes SSH connections to remote rented instances to install system dependencies, sync project code, and execute training scripts. This includes the use of rsync and scp for data transfer to external IP addresses provided by the service.
  • [PROMPT_INJECTION]: The skill implements an indirect prompt injection surface by ingesting and analyzing content from local experiment plans (refine-logs/EXPERIMENT_PLAN.md) and user-provided scripts to derive hardware requirements and execution parameters.
  • Ingestion points: Reads EXPERIMENT_PLAN.md, user scripts, and natural language task descriptions from $ARGUMENTS.
  • Boundary markers: None identified in the provided instructions.
  • Capability inventory: Uses the Bash tool for local CLI operations and remote command execution via ssh.
  • Sanitization: No explicit sanitization or validation of the ingested file content before it influences the generated CLI commands is described.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 1, 2026, 10:18 PM