The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download data and software from external, non-whitelisted sources during setup and tutorials.
In references/docs/quickstart/vector-search.md, the agent is directed to fetch the SIFT1M dataset from a university domain using wget ftp://ftp.irisa.fr/local/texmex/corpus/sift.tar.gz.
In references/docs/quickstart/index.md, the documentation provides instructions to install beta versions of the library from a Gemfury registry using pip install --extra-index-url https://pypi.fury.io/lance-format pylance, which is a non-standard package source.
[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for building agents that ingest and process external datasets, creating a significant surface for indirect prompt injection.
Ingestion points: Data enters the agent's context through lance.dataset(uri) and subsequent to_table() or scanner() calls as documented in references/docs/guide/read_and_write.md.
Boundary markers: The documentation does not describe the use of delimiters or specific instructions for the agent to ignore any natural language instructions that might be embedded within the columnar data.
Capability inventory: The described execution environment includes full Python script execution via pylance, the ability to write to the local filesystem (e.g., UDF checkpoints mentioned in references/docs/guide/data_evolution.md), and extensive network access for cloud storage providers (AWS S3, GCS, Azure) detailed in references/docs/guide/object_store.md.
Sanitization: No mechanisms for sanitizing or validating the content of the ingested data files are discussed or implemented.

lance-format