tao-run-deft-aoi
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
subprocess.run(withinscripts/deft_exec.py) anddocker runcommands to manage the machine learning pipeline stages, including training, inference, and root cause analysis. These operations are essential for the skill's functionality. - [EXTERNAL_DOWNLOADS]: Fetches model weights and datasets from HuggingFace and container images from NVIDIA's NGC registry. These are handled through official CLI tools and libraries.
- [DATA_EXPOSURE]: The script
scripts/align_token_usage.pyreads local Claude Code project metadata to calculate token usage for benchmarking. This access is limited to the local filesystem for metric generation and does not perform network operations. - [SAFE]: The skill includes a robust security feature in
scripts/deft_exec.pythat enforces an 'air-gap' policy. This policy actively blocks potentially dangerous operations like package installations, remote git fetches, and network probes when restricted execution is enabled.
Audit Metadata