AGENTS.md, CLAUDE.md and Cursor rules starter files

Do not edit any files except the one I ask you to create.
Read package files, config files, CI workflows, the README and a sample of source files in this repo.
Draft AGENTS.md using this structure: Project, Commands, Layout, How to work, Ask before you, Never, Done means.
Rules:
- Every command must appear in package scripts, a Makefile, CI or the README. Do not invent commands.
- Mark anything you inferred rather than read with (inferred).
- Under 150 lines.
- List at the end any questions I should answer to finish the file.

The (inferred) marker is what makes the draft safe to trust. Check those lines first.

Read AGENTS.md. Do not edit any code.
For each of these tasks, describe step by step what you would do, which files you would touch, which commands you would run and which rules in AGENTS.md apply:
1. [a small bug fix you did recently]
2. [a feature that touches the database]
3. [adding a dependency]
Then list any point where AGENTS.md did not tell you what to do, or where two rules pulled in different directions.

Run this every time you change the file. The gaps it finds are the next lines worth adding.

# AGENTS.md, CLAUDE.md and Cursor rules starter files

PaperChase prompt pack, version 1.0. Six fill-in starter files, a layout that keeps them in sync, and 8 prompts that write, trim and test your agent instructions. For Codex, Claude Code, Cursor and any tool that reads AGENTS.md.

## What problem this solves

Coding agents read an instructions file before they touch your repo. Most of those files are either empty, a copy of someone else's 400-line rule dump, or three different files that disagree with each other. The agent then guesses your test command, adds a dependency you did not want and "fixes" a file it should never edit.

A good instructions file is short, specific to your repo and written as rules the agent can follow and you can check.

## The layout this pack uses

One source of truth, thin pointers for each tool:

```
AGENTS.md                 the real instructions (Codex, Cursor and others read it)
CLAUDE.md                 one line: @AGENTS.md, plus anything Claude Code specific
.cursor/rules/            optional scoped rules for Cursor, one topic per file
  security.mdc
  tests.mdc
apps/web/AGENTS.md        optional, only if this folder has different rules
```

Claude Code supports importing another file into CLAUDE.md with an @path line, so you write the rules once. Nested AGENTS.md files apply to their folder; the closest one to the file being edited takes priority.

## Contents

1. Base AGENTS.md (any stack)
2. Next.js plus Supabase variant
3. Python FastAPI variant
4. Browser game variant (Phaser or Three.js)
5. CLAUDE.md pointer file
6. Cursor scoped rules: security, tests, UI
7. Prompts: write from an existing repo, trim, test against real tasks, review after a bad session
8. A ten-point quality check for any instructions file