Groq Prompt Scorer — AI skill for Claude Code

groq-prompt-scorer

A Claude Code hook that uses Groq's free LLM API to catch vague prompts before they waste tokens. Clear prompts pass through instantly. Underspecified prompts get flagged so Claude asks you clarifying questions first.

How it works

You type a prompt
       |
  Hook fires (UserPromptSubmit)
       |
  Bypass? (slash commands, short prompts, * prefix)
       |-- yes --> pass through silently
       |-- no  --> send to Groq for classification
                       |
                  verdict: "pass"  --> silent pass-through
                  verdict: "guide" --> inject guidance context
                       |
                  Claude asks clarifying questions
                  You refine, Claude executes

The hook uses Groq's `llama-3.3-70b-versatile` as the primary classifier, falling back to `llama-3.1-8b-instant` on rate limits. Both are free tier.

**Fail-open:** If Groq is unreachable, the API key is missing, or anything errors, the hook exits silently and your prompt passes through unchanged. No disruption.

Examples

Your prompt	Verdict	What happens
fix the TypeError in src/auth.ts line 42	pass	Passes through, Claude acts immediately
refactor getUserData to handle null returns	pass	Specific target + action, passes through
add dark mode	guide	Claude asks: which components? toggle mechanism? color system?
build me a full stack app	guide	Claude asks: which stack? what features? starting state?
it's broken fix everything	guide	Claude asks: what error? which file? reproduction steps?

Requirements

• Claude Code v2.0+
• Python 3.10+
• Free Groq API key

Install

1. Install the groq package

pip install groq

2. Copy the hook

Copy `groq-prompt-scorer.py` to your Claude Code hooks directory:

# macOS / Linux
cp groq-prompt-scorer.py ~/.claude/hooks/session/

#