RAG Tool Skill

rag-tool is an installable Agent Skill for preparing semantically coherent chunkvec ingest and query workflows.

Features

It provides two modes:

• store: chunk text, assign stable topic labels, and store it with cvstore
• search: search stored material with cvquery, using filters only when the user explicitly asks to narrow scope

This skill is designed for source material that needs to be stored and searched through chunkvec while preserving document ids, source/derived kind, optional page metadata, reusable topic labels, and provenance paths when available.

doc names the document identity, such as chapter1. Use kind for source versus derived. Subtypes such as notes, quiz, or flashcards are not first-class exact query filters in the current CLI.

Requirements

• cvstore and cvquery: Required to ingest and query chunkvec data.
• DeepInfra API Key: Required by cvstore and cvquery.
  • Set it via DEEPINFRA_API_KEY (recommended).
  • Or provide it via config.json next to the real cvstore and cvquery executables.

Installation

Using Codex

Install the published rag-tool skill with $skill-installer, using the repo and path where this renamed skill is hosted.

Manual Install

Copy or clone this skill directory into your agent's scanned skills path as ~/.agents/skills/rag-tool.

Modes

Store

Use store when the user wants to add material.

Use $rag-tool in store mode on notes.md.

Use $rag-tool to store chapter1.md.

Use $rag-tool to store derived notes for chapter1.md.

Search

Use search when the user wants to retrieve from stored material.

Use $rag-tool in search mode for: how do embeddings help search?

Use $rag-tool in search mode and search only in my chapter1 notes for regularization.

Use $rag-tool in search mode and search within the chapter1 source for vector search.