Overview

license

production-grade CLI for auditing AI prompts. Detects invisible characters, homoglyph attacks, and adversarial Unicode poisoning in your development workflow.

✨ Features

CI/CD Ready: Exit codes for passing/failing builds based on threat severity.
Scanning: Detect invisible characters, BIDI overrides, and homoglyphs.
Sanitization: Automatically strip dangerous characters (where safe).
Configuration: Support for .promptshieldrc and .promptignore.
JSON Output: Machine-readable reports for dashboard integration.

📦 Installation

# Global installation
$ pnpm install -g @promptshield/cli

# Project dependency
$ pnpm add -D @promptshield/cli

🚀 Usage

View complete usage guide and configurations

Basic Scan

Scan all files in the current directory (respects .gitignore and .promptignore).

$ promptshield scan

Scan specific files or globs:

$ promptshield scan "prompts/**/*.txt" "src/**/*.ts"

Options

Option	Description
`--check`	Fail (exit code 1) on the first detected threat. Ideal for CI.
`--json`	Output results in JSON format.
`--strict`	Enable stricter sanitization rules.
`--write`	Write fixes to disk (used with `fix` or `sanitize`).
`--config <file>`	Path to a specific config file.
`--init`	Generate a default configuration file.

Commands

`scan`

Analyzes files for threats without modifying them.

$ promptshield scan --json > report.json

`fix`

Applies safe automatic fixes to files.

$ promptshield fix --write

`sanitize`

Aggressively cleans text content (may be destructive for some binary/encoded data).

$ promptshield sanitize --write

⚙️ Configuration

Create a promptshield.config.json (or use --init):

{
  "minSeverity": "HIGH",
  "disableHomoglyphs": false,
  "exclude": ["**/node_modules/**", "**/*.min.js"]
}

📚 Deep Dives: For comprehensive explanation of commands and config schemas, see the Documentation section.

License

This library is licensed under the MIT open-source license.

with 💖 by Mayank Kumar Chaudhari

On this page