Free CLI on npm — score any repo, no account needed

Your prompts are code.
Score them. Improve them.
Ship them with confidence.

promptrix score — 0-100, offline, deterministic

Promptrix scans your repository for every prompt it contains — prompt files, agent context files, system prompts embedded in source — scores each one against 16 deterministic checks that run entirely on your machine, and proposes improvements as diffs you review before anything changes.

Or install:

AI-powered optimization ready

Deterministic checks

Optimization methods

Network calls to score

promptrix score

It works on your whole codebase

Prompts do not only live in a chat box. They are scattered across your repository — and they go stale exactly like code does. Point the CLI at a project and it finds every one of them.

Prompt files

prompts/**, *.prompt, .github/prompts/**

Agent context files

CLAUDE.md, AGENTS.md, .cursor/rules, copilot-instructions.md

Prompts inside your code

system prompts, role: 'system' messages, prompt templates

Scoring never leaves your machine

No API key, no network request, nothing uploaded. It is deterministic enough to fail a CI build on.

promptrix score

$ promptrix score

  agent-file (2)

   35  CLAUDE.md
   88  .claude/skills/review/SKILL.md

  prompt-file (1)

   25  prompts/summarize.md

  embedded (1)

   54  src/agent.ts:113

  4 prompts, average 51/100

  Failed checks
  • States purpose and situation (-12)
    → Say why you need this, who it is for
  • Specifies an output format (-4)

Read the docs

/promptrix:report

Or audit your prompts from inside Claude Code

Install the plugin, type one command, and get a full quality report written into the repository you are already working in.

Install it once

/plugin marketplace add pupik1989/promptrix-plugin
/plugin install promptrix@promptrix

Every prompt, weakest first

A score out of 100 per prompt, the category breakdown behind it, and each failing check with the specific edit that fixes it.

One file in your repo

promptrix-report.html, self-contained and openable in any browser. Re-running overwrites it instead of piling up.

A brief for your agent

Copy any prompt’s fix list straight back into Claude Code — or grab one brief covering everything below the threshold.

No account, nothing uploaded

The report is built on your machine from the same deterministic scorer. No API key, no network call, no model call.

claude code

> /promptrix:report

  Scanning 23 prompts…

  ✓ wrote promptrix-report.html

  average 68/100 · 6 below the threshold
  weakest 40 src/lib/queue/processors.ts:50
  top weakness Concrete references (19/23)

View on GitHub

How Promptrix Works

From first scan to CI gate in three steps

Scan

Install once with npm i -g promptrix-cli, then run promptrix score in any repo. It finds prompt files, CLAUDE.md, AGENTS.md, .cursor rules, and system prompts embedded in source. No account, no network, nothing uploaded.

Review the diffs

Every prompt gets a 0–100 score from 16 deterministic checks. Improvements come back as diffs — accept, edit, or undo. Nothing changes until you say so.

Gate and track

Publish scores to your projects dashboard and fail the build when quality drops. Scoring is offline and deterministic, so CI results are reproducible.

Your prompts are code.Score them. Improve them.Ship them with confidence.