3 prompts, zero hallucination

Hey! 👋 Here's what I promised you:

The 3 full prompts that make Claude Code plan before it writes, build without hallucinating, and test before it says "we're done" — copy/paste directly:

1. SPEC FIRST — it writes the plan for you before the code:

Before writing ANY code: write a full spec in SPEC.md for this feature — the goal, the user flow, every file you will create or touch, the edge cases, and a numbered implementation plan. Do NOT write code yet. Show me the spec and wait for my OK.

2. BUILD — it builds the features you talked to it about earlier, and follows the spec:

Now build the feature by following SPEC.md exactly, step by step in order. After each step, tell me which files you changed and why. If you need anything that is not in the spec, STOP and ask me first — no extra features, and never guess an API you are not sure exists.

3. TEST & FIX — after any code, it must test it and fix:

Now test everything you just built: run it, go through the whole flow, and list every bug, error, or broken state you find — system bugs, broken buttons, anything not working right. Then fix them one by one and re-test after each fix. Do not tell me it is done until it runs clean.

How to start

1. Open Claude Code in your project.
2. Prompt 1 → wait for the spec, read it, and give it the OK.
3. Prompt 2 → it builds according to the plan, feature by feature.
4. Prompt 3 → it hunts down the bugs and fixes them before it wraps up.
5. Repeat the same cycle for each new feature — no hallucination, no wasted tokens. ✅