tao-run-on-virtualenv
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill dynamically generates a Python wrapper script from a static template at runtime. This script is written to the job's results directory and executed to manage the process lifecycle, providing durable state and cleanup for the training job.
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the execution of external Python scripts and arguments provided by the agent. This represents an attack surface for indirect prompt injection if the inputs originate from untrusted sources. The implementation mitigates common risks by avoiding shell invocation and validating inputs.
- Ingestion points: Ingests script paths and arguments through the
submitcommand inreferences/virtualenv_runner.py. - Boundary markers: Lacks explicit boundary markers for script content, relying on existing local file permissions.
- Capability inventory: Utilizes
subprocesscalls to launch processes andosmodule for process group signals. - Sanitization: Includes validation of virtual environment structures and restricted placeholder rendering for arguments.
Audit Metadata