promptlint

Static analysis for LLM prompts — catch token waste, quality issues, and injection risks before they reach production.

Install

npm install -g promptlint
# or use without installing
npx promptlint --text "your prompt here"

Usage

# Lint a file
promptlint prompt.txt

# Lint inline text
promptlint --text "Please kindly write me some code"

# Pipe from stdin
cat prompt.txt | promptlint

# Lint multiple files / globs
promptlint prompts/**/*.txt

# Apply auto-fixes
promptlint --fix prompt.txt

# JSON output (for CI integration)
promptlint --format json prompt.txt

# Show token savings dashboard
promptlint --show-dashboard prompt.txt

# CI mode — exit 1 on warnings
promptlint --fail-level warn prompt.txt

Rules

| ID | Category | Severity | Auto-fix | |----|----------|----------|----------| | cost | Cost | INFO | — | | cost-limit | Cost | WARN | — | | prompt-injection | Security | CRITICAL | yes | | structure-sections | Structure | WARN | yes | | clarity-vague-terms | Quality | WARN | — | | specificity-examples | Quality | INFO | — | | specificity-constraints | Quality | INFO | — | | politeness-bloat | Quality | WARN | yes | | verbosity-sentence-length | Quality | INFO | — | | verbosity-redundancy | Quality | INFO | yes | | actionability-weak-verbs | Quality | INFO | — | | consistency-terminology | Quality | INFO | — | | completeness-edge-cases | Quality | INFO | — |

# List all rules
promptlint --list-rules

# Explain a rule
promptlint --explain prompt-injection

Configuration

Generate a starter config:

promptlint --init

This creates a .promptlintrc (YAML) in your current directory:

model: gpt-4o
token_limit: 800
cost_per_1k_tokens: 0.005
calls_per_day: 10000

rules:
  prompt_injection:
    enabled: true
    patterns:
      - ignore previous instructions
  politeness_bloat:
    enabled: true
    words: [please, kindly, thank you]

fix:
  enabled: true
  politeness_bloat: true
  verbosity_redundancy: true

Disable rules inline

Please write some code  # promptlint-disable politeness-bloat
Some vague stuff        # promptlint-disable

Programmatic API

import { analyze, applyFixes, loadConfig } from "promptlint";

const config = loadConfig(); // reads .promptlintrc
const findings = analyze("Please kindly write some code", config);
const fixed = applyFixes("Please kindly write some code", config);

console.log(findings);
// [{ level: 'WARN', rule: 'politeness-bloat', message: '...' }]

Exit codes

| Code | Meaning | |------|---------| | 0 | No issues (or below fail-level) | | 1 | Warnings found (with --fail-level warn) | | 2 | Critical issues found |

Links

• Website
• VS Code Extension
• Python package (PyPI)
• GitHub

License

Apache-2.0