The Agent Skills Directory

[PROMPT_INJECTION]: The skill is subject to indirect prompt injection because it is designed to process and search through untrusted data from various external sources.
Ingestion points: The grub() function accepts local filesystem paths, globs, and remote URLs (e.g., grub('./docs'), grub('https://example.com/guide')) to build search indices.
Boundary markers: The instructions lack specific delimiters or guardrails to ensure the agent ignores any instructions found within the data being indexed.
Capability inventory: The skill possesses the ability to read local files and perform network requests to retrieve website content.
Sanitization: There is no evidence of sanitization or content filtering performed on the data retrieved from the ingestion points.
[EXTERNAL_DOWNLOADS]: The skill relies on the grub Python library and mentions an optional semantic extension that downloads machine learning models from sentence-transformers. These are documented as vendor-provided resources associated with the author 'thorwhalen'.

grub-search