Claude Code is Anthropic's agentic coding tool. It reads your codebase, edits files, and runs commands, and you can use it in your terminal, IDE, desktop app, or browser. These prompts are built around its project memory file, plan mode, and custom slash commands.
Copy the prompt, paste it into the tool's chat, and replace everything in [BRACKETS] with your own details.
1 Write a CLAUDE.md That Actually Helps
What it's for: Turning the starter CLAUDE.md from /init into a short project memory that Claude Code reads at the start of every session.
When to use it: You've just run /init, or Claude keeps forgetting your stack, commands, or rules.
Review my CLAUDE.md and improve it. Keep it short: only things you can't figure out from the code alone.
My app: [APP IDEA]
Things you keep getting wrong: [e.g. "uses npm instead of pnpm", "adds new libraries without asking"]
Rules I want followed: [e.g. "ask before adding dependencies", "never edit the migrations folder"]
Please:
1. Make sure it lists the exact commands to install, run, build, and test.
2. Add my rules as short, specific bullet points.
3. Remove anything vague, outdated, or obvious from the code.
4. Show me the changes before you save.
Tips: Run /context to check the file loaded. Update it whenever you finish a feature or catch Claude repeating a mistake.
What it's for: Getting a reviewed plan before Claude Code touches any files.
When to use it: You're about to add a feature that touches several files. Switch to plan mode first (press Shift+Tab to cycle modes).
I want to add [FEATURE] to my app. Don't change any files yet.
Please:
1. Look through the codebase and tell me which files are involved.
2. Write a step-by-step plan where each step ends in something I can test.
3. List anything risky: data changes, new dependencies, or features that might break.
4. Ask me up to 3 questions that would improve the plan.
Tips: Edit the plan before approving it. If the build goes sideways, go back to the plan and make it more specific instead of patching with follow-up prompts.
What it's for: Using Claude Code's ability to run commands so "fixed" means tested, not guessed.
When to use it: Claude says a bug is fixed, but you haven't seen proof.
Before we call this fixed, prove it.
The bug was: [BUG]
Please:
1. Run the app's tests or build command and show me the real output.
2. If there's no test for this bug, write one small test, show that it fails without the fix, and show that it passes with it.
3. List every file you changed.
4. Tell me one manual check I should do myself.
Tips: Run /diff to review every change before you commit.
What it's for: Saving a checklist you keep pasting as your own command, like /pre-deploy.
When to use it: You've typed the same instructions into Claude Code more than twice.
Create a custom command for this project called [COMMAND NAME, e.g. pre-deploy].
When I run it, it should:
[STEPS, e.g. "1. run the build 2. run the tests 3. check for secret keys in the code 4. list anything that failed"]
Please save it as a skill in .claude/skills/[COMMAND NAME]/SKILL.md, explain what you wrote, and tell me how to run it.
Tips: Older .claude/commands/<name>.md files still work too. Keep each command focused on one job.
What it's for: A quick, read-only check for the most common mistakes in vibe-coded apps.
When to use it: Before you share your link or make the repo public.
Do a security sweep of this project. Don't change any code yet.
Tech stack: [TECH STACK]
Database: [e.g. Supabase / Firebase / none]
Check:
1. Secret keys in front-end code or committed files (show only the first 4 characters).
2. Environment variables with browser prefixes (like NEXT_PUBLIC_ or VITE_) that hold secrets.
3. API routes that don't check who the user is.
4. Database rules: is Row Level Security (or equivalent) on for every table?
Give me a list sorted by severity, then tell me what you could NOT check from the code alone.
Tips: This helps catch common mistakes, but it isn't a full security review. Fix one item at a time and commit after each fix.
