Does this skill require specific hardware?

While it can run elsewhere, it is optimized for Apple Silicon using the MLX framework for high-performance embedding generation.

What is the primary benefit of Skill Embedding VSS?

It allows you to treat agent skills as vectors, making it possible to mathematically compare functionality, find similar tools, and identify unique contributions in a large codebase.

Can I use this for standard text search?

Yes, it includes an embed_query function that allows you to find the most relevant skills based on an arbitrary natural language string.

What is P-adic ultrametric distance?

It is a non-Archimedean metric used for hierarchical clustering that better represents nested or branching relationships between code structures than standard Euclidean distance.

How does novelty detection work in this context?

The skill calculates the distance of a new skill from all existing skills; those with the greatest distance to their nearest neighbor are flagged as the most 'novel' or unique.

Skill Embedding VSS

Name: Skill Embedding VSS
Author: plurigrid

byplurigrid

•

データサイエンスとML

Performs advanced vector similarity search and relational analysis for agent skills using MLX embeddings and DuckDB.

Skill Embedding VSS is a high-performance utility designed to index, compare, and analyze the relational structure of agent skills. By leveraging Snowflake Arctic 1024-bit embeddings optimized for Apple Silicon via MLX and indexing them with DuckDB's HNSW extension, it enables lightning-fast approximate nearest neighbor searches. Beyond simple similarity, it implements P-adic ultrametric distances for hierarchical clustering and GF(3) balanced triad walks, allowing developers to identify redundant capabilities, discover novel code patterns, and navigate complex skill repositories through natural language queries.

主な機能

01GF(3) triad conservation logic for structured skill relationship mapping

02Novelty detection to identify unique skills furthest from existing implementations

038 GitHub stars

04P-adic ultrametric distance calculation for non-Archimedean hierarchical clustering

05MLX-optimized Snowflake Arctic embeddings for high-speed inference on Apple Silicon

06DuckDB HNSW indexing for rapid vector similarity search (VSS)

ユースケース

01Searching for specific code capabilities using natural language descriptions instead of keywords

02Analyzing large agent repositories to find overlapping or redundant skill sets

03Mapping the 'topological' relationship between different AI tool definitions

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add plurigrid/asi skill-embedding-vss

For use in Claude.ai and ChatGPT

Download Skill