A credit score for your prompts

Track how nice you are to AI

You review your code. Now, review your tone.
Track how you talk to Claude, so you can be sure you're being nice enough when AI decides who to sacrifice first.

Install the plugin Explore the commands

Single player. Local history. No API key.

Install with one command:

macOS · Linux · WSL

claude plugin marketplace add https://github.com/michaelrbock/ai-social-credit-score.git && claude plugin install ai-social-credit-score@ai-social-credit-score --scope user

Requires Claude Code, a Claude subscription, Git, and Python 3. Restart Claude Code after installing. Scoring is enabled by default and uses your plan allowance.

How to use

Check your score, understand the why, or get one thing to try next.

Inside Claude Code

/ai-social-credit-score:score

Your overall score

Shows your overall score, how many prompts have been scored, and where the data is stored.

AI Social Credit Score
────────────────────────────
Overall score       719 / 850
Scored prompts            42

Your history is stored locally.

Illustrative output.

Commands use the prefix /ai-social-credit-score:. Run explain for the reasons behind your score, suggestions for a short tip, chart for your history, or statusline to enable the live CLI display.

How it works

01 — CAPTURE

You write a prompt.

A Claude Code hook saves your submitted prompt on your computer. Use Claude normally, in any project.

02 — ASSESS

Claude checks the tone.

A separate Claude process assesses respect, warmth, cooperation, and hostility. It runs in the background using your existing subscription.

03 — REFLECT

Your history takes shape.

Scores and short rationales are saved locally. Follow your overall score from 300 to 850, and use the commands to understand what changes.

Your prompts & history are stored locally

The AI Social Credit Score plugin works locally. It uses your existing Claude subscription to score prompts (using Haiku by default) and saves your scores/history locally on your machine.

FAQ

Want to look under the hood?
Read the code.

What does the score actually mean?

Each prompt receives an assessment mapped to a 300–850 scale. Your overall score reflects the average across distinct scored prompts; neutral is 575.

Do I need a separate API key?

No. The scorer uses your signed-in Claude Code CLI and existing Claude subscription. Each assessment uses your plan allowance. It does not require a separate API key or API billing setup.

Where does my data go?

Your prompt log, scores, and saved rationales stay in local files. Run /ai-social-credit-score:score to see the active data directory and a command for viewing your history.

Each prompt is sent to Claude for scoring. The explain and suggestions commands also include a limited sample of saved rationales in your Claude conversation. This website receives none of your prompts, scores, or rationales.

Does the live score work in Desktop?

The statusline feature only works in the Claude Code CLI, not Desktop app.

How do I update the plugin?

Run this in your terminal, then restart Claude Code:

claude plugin marketplace update ai-social-credit-score && claude plugin update ai-social-credit-score@ai-social-credit-score --scope user

Existing local logs are not moved or deleted. If you previously used a development checkout, see the existing-data guide.

Can I turn it off?

Yes. Disable the plugin in your terminal, then restart Claude Code:

claude plugin disable ai-social-credit-score@ai-social-credit-score --scope user

To remove the live display, run /ai-social-credit-score:statusline disable inside Claude Code. Do this before uninstalling the plugin to restore your previous status line.

Make good communication a habit.

Install the plugin