hardware-workload-estimator
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/check_hardware.pyexecutes thenvidia-smisystem command to retrieve GPU specifications like VRAM and model name. This is a legitimate diagnostic function for hardware workload estimation. - [DYNAMIC_EXECUTION]: The skill uses
scripts/generate_colab_notebook.pyto programmatically create Jupyter Notebook files (.ipynb). These files contain executable code cells and package installation commands (!pip install) designed to be run in the Google Colab environment. - [INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it ingests user-provided training scripts and configuration parameters to generate cloud-ready notebooks. If these inputs were to contain malicious instructions, they could be executed in the cloud environment during the automated offloading process.
- Ingestion points: The
scripts/generate_colab_notebook.pyscript reads the content of an external Python file specified by the--scriptargument. - Boundary markers: The
SKILL.mdfile contains a prominent[!CAUTION]block that explicitly instructs the agent to scan for API keys, passwords, and PII before processing or uploading data. - Capability inventory: The skill includes capabilities to read local files, execute shell commands for hardware checks, and automate browser interactions via Chrome DevTools MCP.
- Sanitization: The skill relies on instructional guardrails for the agent to perform data scanning rather than implementing programmatic sanitization within the Python utility scripts.
Audit Metadata