The Agent Skills Directory

[Unverifiable Dependencies & Remote Code Execution] (LOW): The script examples/local_embedder.py utilizes the sentence-transformers library, which automatically downloads pre-trained model weights from HuggingFace Hub during initialization. Although HuggingFace is a trusted source, the skill performs remote data retrieval at runtime.\n- [Indirect Prompt Injection] (LOW): The skill ingests and processes untrusted text data for chunking and embedding generation. It lacks boundary markers or sanitization, representing a standard vulnerability surface for indirect prompt injection.\n
Ingestion points: The embed_single, embed_batch, and chunk methods across the examples/ directory scripts accept arbitrary string inputs.\n
Boundary markers: None identified in the provided scripts.\n
Capability inventory: The skill performs mathematical computations (embeddings), caching (Redis), and text splitting; no high-risk capabilities such as arbitrary command execution or file system modification were detected in the processing pipeline.\n
Sanitization: No input validation, escaping, or filtering is performed on the ingested text data.

embedding-optimization