embedding-strategies

Pass

Audited by Gen Agent Trust Hub on Aug 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADS
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides templates that utilize the sentence-transformers library to download and load pre-trained embedding models from Hugging Face, such as BAAI/bge-large-en-v1.5 and intfloat/multilingual-e5-large. These are well-known, established models from a trusted AI repository.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (documents and queries) for the purpose of generating embeddings. This constitutes an attack surface for indirect prompt injection; however, the skill includes a _default_preprocess function that performs basic sanitization by removing special characters and excessive whitespace, which helps mitigate basic injection attempts.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The templates demonstrate the use of the OpenAI API for embedding generation. While this involves sending document content to a third-party service, OpenAI is a well-known provider and this is standard functionality for the skill's stated purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 17, 2026, 07:23 PM
Security Audit — agent-trust-hub — embedding-strategies