re:prompt

Linter for your AI prompts — research-backed scoring, rule-based rewriting, and cross-tool analytics for your AI conversations. No LLM needed, <50ms per prompt.

Quick start

pip install reprompt-cli

reprompt check "fix the bug in auth.ts"   # score + lint + rewrite in one command
reprompt scan                              # auto-detect prompts from your AI tools
reprompt                                   # see your dashboard

See it in action

reprompt check -- full diagnostic in one command

reprompt rewrite -- rule-based prompt improvement

reprompt build -- assemble prompts from components

What a bad prompt looks like

What it does

Score, lint, and rewrite

Understand your AI interactions

Manage

Supported AI tools

Installation

pip install reprompt-cli            # core (all features, zero config)
pip install reprompt-cli[chinese]   # + Chinese prompt analysis (jieba)
pip install reprompt-cli[mcp]       # + MCP server for Claude Code / Continue.dev / Zed

Auto-scan after every session

reprompt install-hook               # adds post-session hook to Claude Code

Browser extension

Capture prompts from ChatGPT, Claude.ai, and Gemini directly in your browser. Live score badge shows prompt quality as you type -- click "Rewrite & Apply" to improve your prompt and replace the text directly in the input box.

1. Install the extension from Chrome Web Store or Firefox Add-ons
2. Connect to the CLI: reprompt install-extension
3. Verify: reprompt extension-status

Captured prompts sync locally via Native Messaging -- nothing leaves your machine.

CI integration

GitHub Action

# .github/workflows/prompt-lint.yml
name: Prompt Quality
on: pull_request

jobs:
  lint:
    runs-on: ubuntu-latest
    permissions:
      pull-requests: write    # needed for PR comments
    steps:
      - uses: actions/checkout@v4
      - uses: reprompt-dev/reprompt@main
        with:
          score-threshold: 50   # fail if avg prompt score < 50
          strict: true          # fail on warnings too
          comment-on-pr: true   # post quality report as PR comment

When comment-on-pr: true, every PR gets a quality report:

## reprompt lint Passed

| Metric          | Value          |
|-----------------|----------------|
| Prompts checked | 12             |
| Errors          | 0              |
| Warnings        | 2              |
| Avg Score       | 62/100 (threshold: 50) |

2 violation(s) [click to expand]

The comment updates on each push -- no duplicates. Uses GITHUB_TOKEN (no extra secrets needed).

pre-commit

# .pre-commit-config.yaml
repos:
  - repo: https://github.com/reprompt-dev/reprompt
    rev: v2.5.0
    hooks:
      - id: reprompt-lint

Direct CLI

reprompt lint --score-threshold 50  # exit 1 if avg score < 50
reprompt lint --strict              # exit 1 on warnings
reprompt lint --json                # machine-readable output

Project configuration

reprompt init   # generates .reprompt.toml with all rules documented

# .reprompt.toml (or [tool.reprompt.lint] in pyproject.toml)
[lint]
score-threshold = 50       # fail if avg score < 50

[lint.rules]
min-length = 20            # error if prompt < 20 chars (0 = off)
short-prompt = 40          # warning if < 40 chars (0 = off)
vague-prompt = true        # error on "fix it" etc (false = off)
debug-needs-reference = true

Prompt Science

Scoring is calibrated against 10 peer-reviewed papers covering 30+ features across 5 dimensions:

Cross-validated findings that inform our engine:

• Position bias is architectural -- present at initialization, not learned. Front-loading instructions is effective for prompts under 50% of context window (3 papers agree)
• Moderate compression improves output -- rule-based filler removal doesn't just save tokens, it enhances LLM performance (2505.00019)
• Prompt quality is independently measurable -- prompt-only scoring predicts output quality without seeing the response (ACL 2025, 2503.10084)

All analysis runs locally in <1ms per prompt. No LLM calls, no network requests.

How it works

                          ┌─────────────────────────┐
                          │     reprompt check       │  <- single entry point
                          └────────┬────────────────┘
                                   │
               ┌───────────────────┼───────────────────┐
               v                   v                   v
        ┌─────────────┐   ┌──────────────┐   ┌──────────────┐
        │  Score (0-100) │   │   Lint       │   │   Rewrite    │
        │  30+ features  │   │   rule-based │   │   4 layers   │
        │  5 dimensions  │   │   CI-ready   │   │   no LLM     │
        └──────┬──────┘   └──────┬───────┘   └──────┬───────┘
               │                  │                   │
               └───────────────────┼───────────────────┘
                                   v
                          ┌──────────────────┐
                          │  Unified Report   │
                          │  score + issues + │
                          │  rewritten prompt │
                          └──────────────────┘

 Data sources:
 ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
 │Claude Code│ │  Cursor  │ │  Aider   │ │ ChatGPT  │ │ 5 more.. │
 └─────┬────┘ └─────┬────┘ └─────┬────┘ └─────┬────┘ └─────┬────┘
       └─────────────┴───────────┴─────────────┴─────────────┘
                                 │
                    scan -> dedup -> store -> analyze
                                 │
              ┌──────────────────┼──────────────────┐
              v                  v                  v
        ┌──────────┐     ┌──────────────┐    ┌──────────┐
        │ insights │     │  patterns    │    │ sessions │
        │ style    │     │  repetition  │    │ projects │
        │ digest   │     │  privacy     │    │ agent    │
        └──────────┘     └──────────────┘    └──────────┘

Key design decisions:

• Pure rules, no LLM -- scoring and rewriting use regex + TF-IDF + research heuristics. Deterministic, private, <1ms per prompt.
• Adapter pattern -- each AI tool gets a parser that normalizes to a common Prompt model. Adding a new tool = one file.
• Two-layer dedup -- SHA-256 for exact matches, TF-IDF cosine similarity for near-dupes. Handles the "same prompt, slightly different wording" problem.
• Research-calibrated -- 10 peer-reviewed papers inform the scoring weights. Not vibes, not benchmarks-on-benchmarks.

Conversation Distillation

reprompt distill scores every turn in a conversation using 6 signals:

• Position -- first/last turns carry framing and conclusions
• Length -- substantial turns contain more information
• Tool trigger -- turns that cause tool calls are action-driving
• Error recovery -- turns that follow errors show problem-solving
• Semantic shift -- topic changes mark conversation boundaries
• Uniqueness -- novel phrasing vs repetitive follow-ups

Session type (debugging, feature-dev, exploration, refactoring) is auto-detected and signal weights adapt accordingly.

Privacy

• All analysis runs locally. No prompts leave your machine.
• reprompt privacy shows exactly what you've sent to which AI tool.
• Optional telemetry sends only anonymous 26-dimension feature vectors -- never prompt text.
• Open source: audit exactly what's collected.

Privacy policy

Links

• Website: getreprompt.dev
• Chrome Extension: Chrome Web Store
• Firefox Add-on: Firefox Add-ons
• PyPI: reprompt-cli
• Changelog: CHANGELOG.md
• Privacy: getreprompt.dev/privacy

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT