llm-guard/dist/guards/profanity.js

A TypeScript library for validating and securing LLM prompts

"use strict";
Object.defineProperty(exports, "__esModule", { value: true });
exports.ProfanityGuard = void 0;
const base_1 = require("./base");
class ProfanityGuard extends base_1.BaseGuard {
    constructor(enabled = true) {
        super(enabled);
        // Initialize with a basic list - this should be expanded
        this.profanityList = new Set([
            'badword1',
            'badword2',
            // Add more profanity words here
        ]);
        this.commonSubstitutions = new Map([
            ['a', '@'],
            ['i', '1'],
            ['o', '0'],
            ['e', '3'],
            ['s', '$'],
            ['t', '7'],
        ]);
    }
    normalizeText(text) {
        let normalized = text.toLowerCase();
        // Remove common substitutions
        for (const [char, sub] of this.commonSubstitutions) {
            normalized = normalized.replace(new RegExp(sub, 'g'), char);
        }
        // Remove repeated characters
        normalized = normalized.replace(/(.)\1+/g, '$1');
        return normalized;
    }
    async validate(text) {
        if (!this.isEnabled()) {
            return this.createResult(true);
        }
        const normalizedText = this.normalizeText(text);
        const words = normalizedText.split(/\s+/);
        const matches = [];
        for (const word of words) {
            if (this.profanityList.has(word)) {
                matches.push(word);
            }
        }
        const score = matches.length === 0 ? 1.0 : Math.max(0, 1 - (matches.length * 0.2));
        return this.createResult(matches.length === 0, score, matches.map(word => ({
            rule: 'profanity_detection',
            message: `Found profanity: ${word}`,
            matched: word
        })));
    }
}
exports.ProfanityGuard = ProfanityGuard;