The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches pre-trained embedding models from the HuggingFace registry using the standard sentence-transformers library. This is a common and necessary operation for the skill's primary purpose.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local Qdrant collections to perform semantic benchmarks. 1. Ingestion points: Data is read through the client.scroll method in the create_from_qdrant function within SKILL.md. 2. Boundary markers: Absent; the content is passed directly to the model for encoding. 3. Capability inventory: Includes model downloads from HuggingFace (network) and saving results to benchmark_results.json (file write). 4. Sanitization: Absent; the skill is designed to evaluate raw text content for embedding quality.

embedding-comparison