The Agent Skills Directory

[SAFE]: The skill consists of configuration schemas, metadata, and instructions for using the NVIDIA TAO Toolkit for person re-identification tasks. No malicious patterns or security risks were identified.
[COMMAND_EXECUTION]: The skill defines structured shell commands for the re_identification CLI tool (e.g., re_identification train -e {config_path}). These commands are strictly scoped to the primary functional actions of the skill: training, evaluation, inference, and model export.
[CREDENTIALS_UNSAFE]: No hardcoded credentials, API keys, or secrets were detected. The skill includes fields for encryption_key, but these are left as empty defaults or placeholders for user-supplied values.
[PROMPT_INJECTION]: The instructions provided in SKILL.md are task-oriented and do not contain patterns designed to bypass AI safety filters, extract system prompts, or hijack agent behavior.
[DATA_EXFILTRATION]: No suspicious network operations or unauthorized file access patterns were found. Network settings (such as wandb integration) and file paths (for datasets and results) are standard for machine learning workflows and under user control.
[SAFE]: The documentation includes a security-conscious warning regarding the use of the TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 environment variable, explicitly advising users not to enable it for untrusted checkpoints to prevent potential code execution during model loading.

tao-train-reid