The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or security vulnerabilities were detected in the skill instructions or supporting files.\n- [EXTERNAL_DOWNLOADS]: The skill references the official nvcr.io/nvidia/tao/tao-toolkit Docker image. This is a well-known service and the authoritative source for the TAO Toolkit, and its use is considered safe and appropriate for this skill.\n- [COMMAND_EXECUTION]: The skill provides templates for executing TAO actions through docker run. These commands are standard for the toolkit and do not involve risky patterns such as piping remote content to a shell.\n- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests untrusted dataset paths and configuration files. This is a low-risk architectural property of the TAO workflow. The documentation actively mitigates risk by advising against using TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 for untrusted checkpoints.\n- [DATA_EXPOSURE]: The skill refers to an encryption_key parameter for model protection. No actual keys are hardcoded in the skill; it uses standard placeholders in the schemas and templates.

tao-train-reid