The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides templates and instructions for fetching model weights and checkpoints from external registries including HuggingFace (hf://) and NVIDIA NGC (ngc://). These are standard services for hosting machine learning models and are used as intended for the skill's purpose.
[COMMAND_EXECUTION]: The instructions mandate the use of uv run for executing Python scripts and tests. This practice ensures that commands are executed within a managed virtual environment, which is a security best practice for Python development.
[REMOTE_CODE_EXECUTION]: Implementation skeletons include the use of torch.load to load model weights. While the skill notes that weights_only=False may be used when necessary, it provides clear guidance on its use within the context of loading authenticated model checkpoints.
[PROMPT_INJECTION]: The skill contains instructional constraints, such as prohibiting the reading of evaluation target files (evals/targets/). These are operational guidelines for the agent to ensure task integrity and do not represent malicious injection or safety bypass patterns.

earth2studio-create-diagnostic