Can I search files in my current working directory with QMD?

No, QMD is optimized for external reference repositories. For files in your current working directory, standard tools like Grep and Glob are faster and more efficient.

When should I use qmd_query versus other search tools?

qmd_query is the recommended default as it combines keyword and semantic search with LLM reranking to provide the most accurate results.

What is the difference between qmd_search and qmd_vsearch?

qmd_search is a keyword-based tool (BM25) best for exact terms and identifiers, while qmd_vsearch uses vector embeddings for conceptual or semantic questions.

How does QMD maintain my conversation context window?

For broad research, QMD recommends delegating to a subagent that summarizes findings in its own context, keeping your main conversation clean.

What does the hybrid pipeline's 'reranking' do?

It uses an LLM cross-encoder to score the relevance of retrieved documents, then blends those scores with original search rankings to ensure the best results appear first.

QMD Search

Name: QMD Search
Author: ramonclaudio

byramonclaudio

Learning & Documentation

Performs high-precision hybrid keyword and semantic searches across indexed reference codebases for rapid pattern discovery.

About

QMD Search is a sophisticated retrieval skill for Claude Code designed to query external reference repositories stored in a dedicated directory. By utilizing a hybrid pipeline that combines keyword (BM25) and semantic (vector) search with LLM-based reranking, it enables developers to find specific implementation patterns, error strings, or architectural concepts across massive codebases. The skill features position-aware blending to ensure exact matches remain prominent while leveraging subagent delegation to maintain clean conversation context during deep research tasks.

Key Features

LLM-powered reranking with position-aware blending for high precision
0 GitHub stars
Automated context hygiene via subagent delegation for broad research
Status monitoring for collection indexing and embedding progress
Hybrid search pipeline combining BM25, vector search, and HyDE query expansion
Granular document retrieval with line number and range-offset support

Use Cases

Locating exact identifiers or error strings across multiple external codebases
Researching specific implementation patterns in large open-source reference repositories
Exploring architectural concepts and 'how-it-works' logic using semantic vector search

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add ramonclaudio/skills search

For use in Claude.ai and ChatGPT

Download Skill

GitHub