tao-train-rtdetr
Pass
Audited by Gen Agent Trust Hub on Jun 30, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a configuration-only resource designed for the NVIDIA TAO Toolkit.
- [DATA_EXPOSURE]: No sensitive information is exposed. Variables such as 'encryption_key' use standard placeholder defaults ('tlt_encode') and dataset paths use example placeholders ('s3://bucket/data/train').
- [EXTERNAL_DOWNLOADS]: The skill references official NVIDIA container images ('tao_toolkit.pyt' and 'tao_toolkit.deploy') which are verified vendor resources.
- [INDIRECT_PROMPT_INJECTION]: While the skill defines workflows that ingest external datasets (images and annotations), this is a standard function of model training. The skill does not perform any unsafe interpolation of untrusted data into its own instruction logic.
- [NO_CODE]: The skill package does not include executable scripts or binaries, relying instead on the established 'rtdetr' CLI provided by the TAO environment.
Audit Metadata