# CLAUDE.md starter template

Copy this into the root of any project as `CLAUDE.md`. Claude Code (and other
agent CLIs that read this file) loads it automatically at the start of every
session — it's the first thing the agent knows about you and your rules.
Fill in the bracketed parts, delete what doesn't apply, keep the rest.

---

# [Your name] — working rules

## Who I am

- **[Your name]**, [one line: what you do / what this project is].
- Based in [where], operating [business name] ([entity type], if relevant).
- [Any standing context the agent should never have to re-derive: your
  audience, your stack, your non-negotiables.]

## Rules

- Do what has been asked; nothing more, nothing less.
- NEVER create files unless necessary — prefer editing existing files.
- NEVER create documentation files unless explicitly requested.
- ALWAYS read a file before editing it.
- NEVER commit secrets, credentials, or `.env` files.
- Validate input at system boundaries.
- [Add your own house rules here — the ones you've had to repeat more than
  once are the ones worth writing down permanently.]

## Build & test

```bash
[your build command]
[your test command]
```

- A deploy is not done until it is VERIFIED on the target, not merely
  reported complete by the tool that ran it.

## Project memory (the pattern that makes this compound)

Whenever the agent has to research or figure out something non-obvious —
how a subsystem works, where a thing lives, a deploy gotcha, *why* something
is the way it is — that's the trigger to write it down, once, so no future
session re-derives it.

- **Location:** `memory/` next to this file, one fact per `<slug>.md`.
- **Index:** `MEMORY.md` — a one-line pointer per fact, loaded every session.
- **Before researching:** check the index first. **After learning:** add the
  fact file + the index line. Update in place instead of duplicating.

## Where to look next

This file is rules, not records. If your project also tracks *what
happened* (deals, clients, campaigns, decisions) — as opposed to *how the
code works* — keep that in a separate index the agent reads on demand
instead of loading it every session. See `VAULT-INDEX.md` for that pattern:
one entry-point file that maps folders to purposes, so the agent opens only
the one folder it needs instead of bulk-reading everything.

- `memory/` — how things work (auto-loaded every session).
- `VAULT-INDEX.md` → your notes/records folder — what happened (read on demand).