You installed Graphify. The map exists. Now what? This is the part nobody explains — what to type, when to refresh, and how to tell it's actually working.

Haven't installed it yet? Start here: Graphify — Full Setup Guide (Claude Code)

The mental model (30 seconds)

Without a map, Claude Code opens your project like a stranger every session — reading files, guessing, re-reading. That's your usage limit burning on rediscovery.

With the map, Claude looks up graph.json first: what exists, what calls what, where things live. It reads the two files it actually needs instead of twenty it might.

So the whole workflow is one habit: keep the map fresh, and let Claude look at it before it starts digging.

The daily loop

1. Start the session by pointing at the map

First message of a new session, before you ask for anything:

Check graphify-out/graph.json before you touch anything.
Then tell me in 3 lines what this project is and where the entry points are.

If the answer is specific — real file names, real function names — the map is live. If it's vague, Claude isn't reading it (see Is it even working? below).

2. Ask questions that use the graph, not the filesystem

These are the prompts that pay for themselves:

• Using the graph, where is <feature> implemented? Don't read files until you've answered from the map.
• What breaks if I change <function>? List the callers from the graph.
• Show me everything that touches the database layer.
• What's dead code here — nodes nothing points at?

The pattern: ask the map first, open files second. Every prompt that starts with "using the graph" is a prompt that doesn't cost you a full-project crawl.