The Agent Skills Directory

[SAFE]: The skill serves as an orchestrator for model deployment, fine-tuning, and evaluation. Its primary operations are performed through documented scripts and subprocess calls using list-based arguments to prevent command injection.
[SAFE]: Data ingestion via ingest_dataset.py includes robust security checks, such as path-traversal (Zip Slip) prevention and zip-bomb guards, to handle untrusted external datasets safely.
[SAFE]: Uses ast.literal_eval() for parsing structured log data and metadata, ensuring that data is evaluated as Python literals rather than arbitrary executable code.
[SAFE]: The optional Web UI implements security best practices by binding to the local loopback interface by default and using an explicit environment variable allowlist for subprocesses to prevent accidental sensitive data exposure.

deepstream-eval-and-finetune