One command reads every Claude Code, Codex, Cursor, Gemini, Copilot & Cline chat on your machine, scores how you prompt — and your English — then rewrites your worst prompts and coaches you to get better.
Not a lint check. Your own AI reads a sample of your real prompts and writes you a coaching report.
Prompts written, your AI crew, prime-time hour, catchphrases, rage moments — and your prompting personality.
Clarity, context, specificity, structure, efficiency and English — with what's dragging each one down.
Your worst real prompts, rewritten the way a senior engineer would write them. Copy, paste, learn.
Your actual grammar mistakes, corrected, with a CEFR estimate. Judged on your prose — never your code.
Finds the instructions you type every session and turns them into a CLAUDE.md, AGENTS.md or Cursor rules file.
Four timed mini-games built from your weak spots. No paste, no spell-check. Re-run monthly to watch your score climb.
Most wasted agent turns come from prompts that make the AI guess. See exactly which of yours do — and how to rewrite them.
"why because the token is expire" costs you a turn. Get your real mistakes, the fix and the rule — then practise.
Read-only scan of your AI tools' history folders. Only what you typed is kept — replies, tool output and pasted code/logs are dropped. Keys, tokens and passwords are redacted.
Dozens of stats are computed on your machine: length, timing, one-word "ok"s, frustration, context-giving, repeated instructions, punctuation, grammar patterns.
A sample of ~180 prompts goes to the agent you already use — Claude Code, Codex, Gemini CLI, your API key or local Ollama — and your report opens in the browser.
There is no account and no telemetry in the CLI. The deep analysis runs through your own AI login or key. Want zero network? Run npx howamiprompting --no-ai or use Ollama.
It's all open source — read every line.
Opt-in from the Share tab of your report. Only a nickname and scores are published.
| # | Dev | Score | Persona | Prompts | Tools |
|---|---|---|---|---|---|
| Loading… | |||||
No. It's free and MIT-licensed. The analysis uses a little of your own Claude Code / Codex / Gemini quota (a few calls), or nothing at all with --no-ai or Ollama.
Node.js 18.17+. For the deep analysis: Claude Code, Codex CLI, Gemini CLI, an Anthropic/OpenAI/OpenRouter API key, or Ollama. Cursor history needs Node 22.5+ or the sqlite3 command (built into macOS).
No — English is one of six skills. The others are clarity, context, specificity, structure and efficiency: the things that decide whether your agent gets it right first time.
Run npx howamiprompting --include my-export.md with any Markdown/JSON/text export, and open an issue so we add a collector.
Only on your machine, in ~/.howamiprompting/ (your past scores, so you can track progress). Delete the folder any time.
Founder from Kolkata, building SaaS products like GoDial and HyperScreenshot. I talk to Claude Code and Codex all day and wanted to know if I was any good at it — so I built the tool I wished existed.