The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes templates that download pre-trained machine learning models for local execution.
Evidence: Templates in SKILL.md reference BAAI/bge-large-en-v1.5 and intfloat/multilingual-e5-large for use with the sentence-transformers library, which fetches these from the Hugging Face Hub.
[INDIRECT_PROMPT_INJECTION]: The skill provides an ingestion pipeline for external data, which is a common surface for indirect prompt injection attacks.
Ingestion points: The process_documents method in the DomainEmbeddingPipeline class (SKILL.md) accepts raw document content for chunking and embedding.
Boundary markers: While the skill suggests preprocessing, it does not implement specific delimiters or instructions for the agent to ignore embedded commands within the ingested text.
Capability inventory: The skill interacts with embedding APIs (OpenAI, Voyage AI) and local libraries to generate vector representations.
Sanitization: Includes a _default_preprocess method that performs basic cleaning using regular expressions to remove excessive whitespace and non-alphanumeric characters.

embedding-strategies