The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill includes examples using torch.load() to import embedding tensors from local storage. This is a standard operation in machine learning workflows, though it utilizes Python's pickle serialization format. As a best practice, users should ensure the files being loaded originate from trusted sources.
[EXTERNAL_DOWNLOADS]: The skill uses vec2text.load_pretrained_corrector("gtr-base"), which fetches pretrained models from Hugging Face. This is the intended behavior for the embedding-inversion task and uses a well-known model repository.
[INDIRECT_PROMPT_INJECTION]: The skill defines an interface for processing external data (embeddings and text strings).
Ingestion points: embeddings.pt file loading and list of strings in invert_strings.
Capability inventory: The skill converts data to text for user review; it does not pass output to dangerous commands, network calls, or write operations.
Boundary markers: None present.
Sanitization: None present.

vec2text