/memory-optimize

name: memory-optimize
description: Choose the cheapest context path across direct read, markdown conversion, graph build, graph update, or graph query.
argument-hint: "[path-or-url] [question]"
allowed-tools:
  - Read
  - Bash
  - Glob
  - Grep

Objective

Recommend the cheapest memory workflow that preserves answer quality and future reuse.

Context

Inputs:

• optional path or URL
• optional question or task goal

Optimization targets:

• lower token cost now
• lower reread cost later
• better persistence when repeated queries are likely

Process

1. Inspect input shape and reuse horizon:

• plain text or small local file
• noisy web page
• PDF/Office doc or other hard-to-read binary-ish source
• mixed corpus
• existing graph with new question
• existing graph with changed sources

2. Check requirements:

• `apx check graphify`
• `apx check markitdown-file-intake`
• `apx check defuddle`

3. Recommend exactly one primary path:

• direct read for small, simple, one-shot text
• `defuddle` -> Markdown -> read or build when web chrome is the main problem
• `markitdown-file-intake` -> Markdown -> read or build when document format is the main problem
• `graphify` full build when repeated questions or relationship navigation justify persistent memory
• `graphify --update` when a graph already exists and sources changed
• graph query against existing graph when the graph exists and the task is retrieval, tracing, or explanation

4. State why that path is cheapest:

• lower immediate token cost
• lower repeated reread cost
• better persistence across future sessions

5. Include exact next command(s) for the recommended path. 6. If a required tool for the best path is missing, either:

• recommend the next-best lower-capability path that still preserves answer quality, or
• stop and say the missing requirement makes the cheap path unavailable

Stop Conditions

• input is too ambiguous to choose between direct read, helper conversion, build, update, or query
• the recommended path depends on a missing tool and no credible fallback preserves answer quality
• user asks for persistent graph advice but the target path or URL is unavailable

Rules

• optimization target = lower token cost, lower reread cost, better persistence
• `graphify` is long-lived memory path
• helpers are normalization tools only
• keep `graphify` primary for persistent memory, helpers secondary, Obsidian optional
• do not recommend a graph build for trivial one-shot text when direct read is cheaper
• do not claim a helper or graph path is available before requirement checks pass
• do not claim bundled upstream code