Memory Manager

A suite of tools designed to bridge the gap between raw files and the PromptWar̊e ØS Vector Database.

Purpose

The Kernel provides the mechanism (Vector.Store), but not the policy for how to slice data. This skill provides the policy: intelligent chunking, metadata extraction, and formatting.

Tools

1. Chunker (chunker.ts)

Splits text or files into semantic chunks suitable for embedding.

Usage:

# Chunk a file
deno run -A chunker.ts --file README.md

# Chunk raw text
deno run -A chunker.ts "Some long text..."

# Custom chunk size (default: 1000 chars)
deno run -A chunker.ts --file README.md --size 500

# JSON Output (for programmatic use)
deno run -A chunker.ts --file README.md --json

Output Format (JSON):

[
  {
    "text": "Chunk 1 content...",
    "metadata": {
      "source": "README.md",
      "index": 0,
      "total": 5
    }
  },
  ...
]

Workflow

1. Hydrate: Agent loads this skill.
2. Chunk: Agent runs chunker.ts on a target file.
3. Store: Agent iterates over the output and calls Vector.Store for each chunk.

References

• RFC 0030: Vector Subsystem