PromptLintLint your LLM prompts

Catch cost waste, security risks, PII leaks, and quality issues before they reach production. 20 intelligent rules, 5 auto-fixes, zero API calls.

Get Started →

View on GitHub

Browse Rules

🔒

Security First

Detect prompt injection, jailbreak attempts, hardcoded secrets, PII exposure, and context boundary violations — all at CRITICAL severity with zero false negatives in production.

Security rules

💰

Cost Analytics

Count tokens with tiktoken, project monthly spend, and quantify savings before you deploy. See exactly how much a fix saves at your call volume.

Cost rules

✨

Quality Checks

Catch vague language, missing examples, weak verbs, politeness bloat, and terminological inconsistency — the subtle issues that degrade model output quality.

Quality rules

🔧

Auto-Fix

Five rules can automatically rewrite your prompt — remove politeness bloat, collapse redundant phrases, strip injections, scaffold structure, and normalize spacing.

Auto-fix guide

⚡

CI/CD Ready

JSON, SARIF v2.1, and colored terminal output. Exit codes map directly to fail levels. Drop into GitHub Actions, GitLab CI, or any pipeline in under 5 minutes.

GitHub Actions

🏗️

Structure Analysis

Ensure prompts have task, context, and output sections. Detect missing role definitions, absent output format specs, and conditions that increase hallucination risk.

Structure rules

Quick Start

Install via pip or npm, then lint your first prompt:

pipnpmGitHub Action

bash

pip install promptlint-cli
promptlint --file prompt.txt

bash

npm install -g promptlint-cli
promptlint --file prompt.txt

bash

- uses: AryaanSheth/promptlint@v1
  with:
    path: prompts/
    fail-level: warn

Example Output

  PromptLint v1.3.0

  ┌─────────────────────────────────────────────────────────────┐
  │  File: prompts/system_prompt.txt  (97 tokens · ~$0.0005)   │
  └─────────────────────────────────────────────────────────────┘

  [ CRITICAL ] prompt-injection (line 3)
    Injection pattern detected: 'ignore previous instructions'

  [ CRITICAL ] pii-in-prompt (line 7)
    Email address detected: us**@example.com

  [ WARN ] structure-sections (line -)
    No explicit sections detected (Task / Context / Output)

  [ WARN ] politeness-bloat (line 1)
    Consider removing 'Please' — models don't need it

  [ INFO ] specificity-examples (line -)
    No examples provided — add 1-2 to improve consistency

  ────────────────────────────────────────────────────────────────
  Score: 42/100  Grade: F   5 findings  (2 critical · 2 warn · 1 info)
  Run with --fix to auto-resolve 3 of these issues

Install via Multiple Channels

Channel	Command	When to use
pip	`pip install promptlint-cli`	Python projects & scripts
npm	`npm install -g promptlint-cli`	Node.js projects & pipelines
GitHub Action	`AryaanSheth/promptlint@v1`	Any GitHub-hosted CI
VS Code	Search "PromptLint" in Extensions	Real-time editor feedback

20 Rules, 5 Auto-Fixes

Category	Rules	Auto-fixable
💰 Cost & Tokens	`cost`, `cost-limit`	—
🔒 Security	`prompt-injection`, `jailbreak-pattern`, `secret-in-prompt`, `pii-in-prompt`, `context-injection-boundary`	✅ 1
🏗️ Structure	`structure-sections`, `role-clarity`, `output-format-missing`, `hallucination-risk`	✅ 1
✨ Clarity	`clarity-vague-terms`	—
🎯 Specificity	`specificity-examples`, `specificity-constraints`	—
📝 Verbosity	`politeness-bloat`, `verbosity-redundancy`, `verbosity-sentence-length`	✅ 2
💪 Actionability	`actionability-weak-verbs`	—
🔄 Consistency	`consistency-terminology`	—
✅ Completeness	`completeness-edge-cases`	—

Browse all rules →

PromptLintLint your LLM prompts

Security First

Cost Analytics

Quality Checks

Auto-Fix

CI/CD Ready

Structure Analysis

Quick Start ​

Example Output ​

Install via Multiple Channels ​

20 Rules, 5 Auto-Fixes ​

Quick Start

Example Output

Install via Multiple Channels

20 Rules, 5 Auto-Fixes