Embedding Strategies Skill for Claude Code: Choose Embedding Models

A guide that helps your agent choose an embedding model and chunking approach for semantic search and RAG apps.

Installs: 11.6K; on skills.sh
Stars: 42.1K; on GitHub
Files: 1; besides SKILL.md
Updated: Oct 2026; last repo commit

This skill gives your agent a model comparison table and a set of Python templates for embeddings. The table lists 10 models, from voyage-3-large and text-embedding-3-large to bge-large-en-v1.5 and all-MiniLM-L6-v2, with dimensions, token limits and what each is best for. It recommends Voyage AI for Claude apps, saying Anthropic recommends it.

The six templates sit in references/details.md and cover Voyage, OpenAI, local Sentence Transformers, chunking, domain pipelines and quality evaluation. Their code reads a VOYAGE_API_KEY from the environment and uses an OpenAI client, so running them sends your text to those providers.

Good for

Developers building semantic search or RAG on top of Claude
Teams comparing Voyage, OpenAI and local embedding models
Anyone tuning chunk size, dimensions or multilingual search

Install embedding-strategies

Claude Code

Copy this prompt.

Install the embedding-strategies skill in Claude Code. Run this command:

/plugin marketplace add wshobson/agents
/plugin install llm-application-dev@claude-code-workflows

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Pick one route. The plugin routes install the whole llm-application-dev plugin: 3 agents, 3 commands and 8 skills (counted from the plugin folder in the repo tree, which has no hooks or MCP config files). The skills-only routes install just this skill, with no agents or commands. The skill is guidance with Python templates; running them needs Python plus the provider packages they import, and a Voyage or OpenAI account for those two templates.

From the project README. Source.

What's in the skill

When it loads

Select and optimize embedding models for semantic search and RAG applications. Use when choosing embedding models, implementing chunking strategies, or optimizing embedding quality for specific domains.

Sections in SKILL.md

When to Use This Skill
Core Concepts
1. Embedding Model Comparison (2026)
2. Embedding Pipeline
Templates and detailed worked examples
Best Practices
Do's
Don'ts

Key facts

Author: wshobson (owner of wshobson/agents)
Repo: wshobson/agents, folder plugins/llm-application-dev/skills/embedding-strategies
Installs on skills.sh: 11,612 as of 2026-11-02
GitHub stars: 42,081 for the whole wshobson/agents repo, 2026-11-02
License: MIT
Last repo commit: 2026-10-30
What is inside: 2 files: SKILL.md and references/details.md
Install routes in the README: The wshobson/agents README documents a plugin route and a skills-only route with `npx skills add` or `gh skill install`

What it includes

A table of 10 models with dimensions, token limits and best use.
Includes Voyage models for general, code, finance and law text, two OpenAI models, and open models such as bge-large-en-v1.5, all-MiniLM-L6-v2 and multilingual-e5-large.
Read · Embedding model comparison
Python code using VoyageAIEmbeddings with a VOYAGE_API_KEY.
Marked as recommended for Claude, with specialized models for domains.
Write · Template 1: Voyage AI embeddings
Python code that batches texts and can reduce dimensions.
Uses text-embedding-3-small by default and shows Matryoshka-style dimension reduction.
Write · Template 2: OpenAI embeddings
Runs an embedding model on your own machine.
Includes an E5 model example that uses instructions.
Write · Template 3: Local embeddings with Sentence Transformers
Ways to split documents before embedding.
The code uses tiktoken and nltk for token and sentence splitting.
Write · Template 4: Chunking strategies
An embedding pipeline with a code-specific variant.
The code-specific version uses tree_sitter_languages.
Write · Template 5: Domain-specific pipeline
Python code that scores retrieval quality at a cutoff k.
Compares retrieved documents with the relevant ones you list for each query.
Write · Template 6: Embedding quality evaluation
Do and do-not lists for model choice, chunking and caching.
Examples are matching the model to the use case, normalizing for cosine search, caching static content, and not mixing embedding models.
Read · Best practices

From the official docs

Permissions

Low risk

The skill bundles no scripts, but its templates call Voyage or OpenAI with your API key and send them your text. Use keys scoped to this work and only embed content you may share.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Template code sends your text to a provider
Running the Voyage or OpenAI templates sends document text to that provider. Check its data terms before embedding private content. github.com

Technical details

Turning on read-only	No switch. The skill is guidance, and templates only run when you run the code.
Talks to	Voyage AI, in Template 1. OpenAI, in Template 2
Sign-in and keys	Template 1 reads VOYAGE_API_KEY from the environment. Template 2 creates an OpenAI client without naming a key in the code.
Scopes	Writes Python code from its templates into your project when you ask Template code run later calls embedding providers with your text

Questions

What does the embedding-strategies skill do?

It helps your agent choose an embedding model and chunking approach for semantic search and RAG. It adds a 10-model comparison table and six Python templates.

How do I install it?

Use the install box at the top of this page. It offers the llm-application-dev plugin (8 skills, 3 agents, 3 commands, no hooks or MCP config; counts from the repo tree) or just this skill through the skills CLI or GitHub CLI. The skills-only routes add no agents or commands.

Which embedding model does it recommend for Claude?

Voyage AI. SKILL.md lists voyage-3-large as "Anthropic recommended" for Claude apps and voyage-3 as the cost-effective choice.

Does it need an API key?

The skill itself does not. Its Voyage template reads VOYAGE_API_KEY from your environment, and the OpenAI template uses the OpenAI client, so those templates need an account with the provider when you run them.

Is it safe to use?

It bundles no scripts. Code you run from its templates sends your text to Voyage or OpenAI, so only embed content you are allowed to send, and keep keys in environment variables.

Can I use it without paid APIs?

Yes for the local route. Template 3 uses Sentence Transformers, and the model table lists open models for local deployment such as bge-large-en-v1.5.

Information

updated Nov 2, 2026

Repository: wshobson/agents
Folder: plugins/llm-application-dev/skills/embedding-strategies

wshobson/agents skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 5

https://github.com/wshobson/agents/blob/HEAD/plugins/llm-application-dev/skills/embedding-strategies/SKILL.md · github.com
wshobson/agents README · github.com
llm-application-dev plugin.json · github.com
wshobson/agents marketplace.json · github.com
embedding-strategies details.md · github.com