The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external document data which could contain malicious instructions designed to influence the agent's behavior during embedding or subsequent retrieval tasks.
Ingestion points: Document content enters the system via the process_documents method in the DomainEmbeddingPipeline class within SKILL.md.
Boundary markers: No explicit instruction delimiters or boundary markers are defined in the code templates to separate metadata from user content.
Capability inventory: The skill performs network operations using the openai Python library to generate embeddings.
Sanitization: The _default_preprocess method performs basic regex-based cleaning (whitespace and special character removal), but does not specifically filter for prompt injection markers.
[SAFE]: All identified Python libraries and external URLs refer to well-known, legitimate technology services and organizations.
Code templates utilize standard, industry-accepted libraries including openai, sentence_transformers, tiktoken, nltk, scipy, and numpy.
External links point to official documentation and community leaderboards on openai.com, sbert.net, and huggingface.co.

embedding-strategies