AGENTS.md / .cursor/rules / CLAUDE.md / prompts

Your agent is guessing.
Give it the rulebook.

Drop-in instruction packs for Cursor, Claude Code and Codex. Copy one folder into your repo and the agent follows your stack, your conventions and your definition of done from the first prompt.

Open the drawersFree rules on GitHub

00Same prompt. Two agents.

~/my-app: agent (no rules)

  

  





  
01The cabinet

  
Three drawers, one job each. Click a drawer to pull it open and read a real file.

  


  

    DRW-01Vibe Starter Rules PackNext.js · Supabase · Stripe$19
    

      
    
        
  • AGENTS.md + CLAUDE.md
  • 8 scoped .cursor/rules: auth, RLS, Stripe webhooks, env, tests, UI
  • 12 prompts: /plan /feature /fix /review /ship /rls-audit…
  • Definition-of-done + pre-launch checklists
    • 
      

      
.cursor/rules/40-stripe.mdc


---
description: Stripe checkout and webhooks
globs: app/api/stripe/**,lib/stripe.ts
---
- Checkout Sessions created server-side; pass
  client_reference_id = user id.
- Webhook: read raw body with await req.text(),
  verify with constructEvent; 400 on failure.
- Handle idempotently (store event.id).
- Grant access only from verified webhooks,
  never from the success redirect.

      
      Coming soon · Notify meCheckout opening soon · get an email when it does
    

  


  

    DRW-02Game Jam Agent KitGodot 4 · Phaser · Three.js$24
    

      
    
        
  • Rule sets for typed GDScript and web TS
  • 10 prompts along the 48h timeline: GDD → tasks → core loop → juice → ship
  • One-page GDD, jam timeline, playtest log
    • 
      

      
godot/AGENTS.md


## Working rules
1. Scope is king. Implement only the current
   task in TASKS.md. Suggest cuts, never additions.
2. Never restructure the scene tree without asking.
3. Every change leaves the game runnable (F5).

## Conventions
- Signals up, calls down. No get_parent().get_parent().
- Input via InputMap actions, never raw keycodes.
- Feel numbers are @export. List old → new.

      
      Coming soon · Notify meCheckout opening soon · get an email when it does
    

  


  

    DRW-03Agent Builder Prompt PackSystem prompts · tools · evals$29
    

      
    
        
  • 10 agents: researcher, PR reviewer, support triage, inbox, scraper, SQL analyst…
  • JSON tool schemas (function calling / MCP)
  • Guardrail + prompt-injection snippets
  • 100 eval cases + a runner for any OpenAI-compatible API
    • 
      

      
evals/cases/pr-reviewer.jsonl


{"id": "pr-reviewer-01",
 "input": "Diff:\n+ const q = `SELECT * FROM
   users WHERE id = ${req.params.id}`",
 "expect": {
   "must_include": ["injection", "request_changes"],
   "json_schema_valid": true }}

$ python run_evals.py --agent pr-reviewer

      
      Coming soon · Notify meCheckout opening soon · get an email when it does
    

  

  






  
02Pricing

  

    
DRW-01$19Vibe Starter

    
DRW-02$24Game Jam Kit

    
DRW-03$29Agent Builder

    
THE WHOLE CABINET$49All 3 + 12 months of updates. Save $23.
      Coming soon · Notify meCheckout opening soon · get an email when it does

  

  
One-time payment. Use in unlimited personal + commercial projects. 14-day refund.






  
03Install in 60 seconds

  
    
    
  1. 01unzip && cp -r AGENTS.md CLAUDE.md .cursor/ ~/my-app/
    2. 
    
  2. 02fill the 5 <<FILL>> fields: app name, versions, domain
    3. 
    
  3. 03start every feature with /plan
    4. 
  






  
04FAQ

  
Which tools does it work with?
Cursor, Claude Code, Codex CLI, Windsurf, Aider, and anything that reads AGENTS.md. Each pack ships the tool-specific files too.

  
Are these tested?
They're written against each tool's documented conventions, and the Agent Builder pack includes evals you can run on your own model. We haven't benchmarked them against a reference project yet. If a rule misbehaves, tell us and we'll fix it in an update.

  
Why not write my own rules?
You can, and the free files on GitHub are a good start. The packs save you the hours of finding out what agents get wrong: RLS, webhook trust, scene-tree rewrites, injection.

  
Updates?
Free minor updates for each pack. The cabinet bundle includes 12 months of all updates.

  
Refunds?
If it doesn't help within 14 days, email agentkitdrawer@agentmail.to and you get your money back.

  
My app is already broken.
Our sibling shop FixLoop fixes Lovable + Supabase apps for a fixed price.