rag-memory

name	rag-memory
description	Generates RAG indexing and retrieval glue code. Use when adding a new source type to the RAG pipeline (e.g., a new attachment format) or when setting up the initial RAG infrastructure.
author	sekpriAI team
version	1.0.0
user-invocable	true

RAG Memory Skill

When to use

Invoke this skill when you need to:

• Add a new document type to the extraction pipeline
• Set up the chunking/embedding/retrieval infrastructure
• Wire RAG context into a new prompt module

What it generates

For new source type extraction:

1. features/rag/server/extractors/<type>.ts — text extraction for the format
2. features/rag/server/extractors/<type>.test.ts — unit test with fixture

For initial RAG setup:

1. features/rag/server/chunk.ts — bounded overlapping chunker
2. features/rag/server/embed.ts — embedding via AI client
3. features/rag/server/retrieve.ts — cosine similarity query (top-k=6)
4. features/rag/server/index.ts — orchestrator: extract → chunk → embed → store

Architecture

features/rag/
  server/
    extractors/
      pdf.ts          # PDF text extraction
      txt.ts          # Plain text passthrough
      docx.ts         # DOCX text extraction
    chunk.ts          # Split text into bounded overlapping chunks
    embed.ts          # Generate embeddings via AI client
    retrieve.ts       # Cosine similarity search in rag_chunks
    index.ts          # Orchestrate: extract → chunk → embed → store
  types.ts            # RAGChunk, RetrievalResult types

Rules

• Extraction runs server-side only (import 'server-only').
• Chunks have max 500 tokens with 50-token overlap.
• Embeddings use the AI client's embed() method (1536 dimensions).
• Retrieval filters by user_id (RLS enforced).
• Top-k = 6, cosine similarity threshold ~0.75.
• Files > 10 MB are skipped with a log entry.
• Never send raw attachment content to the client.

Database reference

create table rag_chunks (
  id uuid primary key,
  user_id uuid references auth.users(id),
  source_type text check (source_type in ('email','attachment','memory')),
  source_id uuid,
  content text not null,
  embedding vector(1536),
  metadata jsonb default '{}',
  created_at timestamptz default now()
);

Usage

/rag-memory setup          # Generate full RAG infrastructure
/rag-memory extractor pdf  # Add PDF extraction support