The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the umap-learn package and its optional dependencies (such as TensorFlow for Parametric UMAP) using standard Python package managers. These are well-known, legitimate libraries in the data science ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection as it is designed to ingest and process external datasets for dimensionality reduction and clustering.
Ingestion points: Data arrays passed to the fit, fit_transform, and transform methods across all examples in SKILL.md and references/api_reference.md.
Boundary markers: Absent; the skill expects pre-loaded numerical data without specific markers to distinguish between data and potential instructions.
Capability inventory: Includes local file creation for visualizations (plt.savefig in SKILL.md) and package installation instructions (pip install in SKILL.md).
Sanitization: Absent; while standard scaling (StandardScaler) is used, it is intended for feature normalization rather than security-based sanitization of input data.
[DYNAMIC_EXECUTION]: The skill demonstrates the use of Numba-compiled custom distance metrics. Numba performs Just-In-Time (JIT) compilation of Python code into machine code to optimize performance. In the provided context, this is a standard application of the library's performance features for user-defined distance functions.

umap-learn