vastai-core-workflow-a
Pass
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
subprocess.runmodule in Python to execute local system commands, including thevastaiCLI,ssh, andscp. These commands are used to search for GPU offers, provision instances, transfer files, and manage the instance lifecycle. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided inputs such as GPU specifications, Docker image names, and startup commands. These inputs are interpolated into arguments for the
vastaiandsshcommands. - Ingestion points: Function arguments in
SKILL.md(e.g.,gpu_name,image,onstart_cmd) and environment variables used in Bash snippets. - Boundary markers: None identified.
- Capability inventory: Local execution of
vastai,ssh, andscpviasubprocess.runand direct Bash commands. - Sanitization: The Python scripts use
subprocess.runwith a list of arguments, which provides a layer of protection against shell injection by avoiding a shell context, but no specific input validation logic is present for the parameter values. - [EXTERNAL_DOWNLOADS]: The workflow involves provisioning remote instances that download Docker images from external registries and install software dependencies (e.g., via
pip) during the startup process.
Audit Metadata