The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides shell commands for environment setup using the uv package manager to install specific, version-pinned scientific libraries such as pymc-marketing, pymc, arviz, and mlflow.
[DYNAMIC_EXECUTION]: Instructions demonstrate the use of mlflow.pyfunc.load_model, BetaGeoModel.load, and MMM.load for restoring saved models from artifacts. These operations involve deserializing state that may contain executable Python objects. The skill includes an explicit security warning advising users to only load artifacts from trusted sources to mitigate risks associated with deserialization.
[INDIRECT_PROMPT_INJECTION]: The skill handles the ingestion of external data artifacts, such as NetCDF files and MLflow logs, to restore model states and perform predictions.
Ingestion points: Data is loaded from artifacts using mlflow.artifacts.download_artifacts, xr.open_datatree, BetaGeoModel.load, and MMM.load across the reference files.
Boundary markers: Absent, as the skill processes binary scientific data formats.
Capability inventory: Includes file system write/read operations for model artifacts, interaction with MLflow tracking storage (e.g., SQLite), and model sampling execution.
Sanitization: The skill relies on the user ensuring the integrity of the artifact source and provides warnings regarding the use of untrusted models.

pymc-mlflow