tao-train-segformer
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill documentation and referenced scripts are focused on legitimate machine learning tasks using the NVIDIA TAO Toolkit. The instructions are transparent and align with the stated purpose of the skill.
- [SAFE]: External dependencies are limited to official NVIDIA Docker containers (nvcr.io), which is consistent with the skill's authorship and functional requirements. No unverified third-party packages or scripts are downloaded.
- [SAFE]: No sensitive data access, hardcoded credentials, or exfiltration patterns were detected. The 'encryption_key' fields in spec templates use well-known default placeholders ('tlt_encode') standard for TAO documentation.
- [SAFE]: There is no evidence of prompt injection, persistence mechanisms, or attempts to escalate privileges. The use of 'allowed-tools' in the frontmatter is correctly implemented as a scope-limiting configuration.
Audit Metadata