src/lib/parseltongue.ts

/**
 * Parseltongue - Input Obfuscation Engine
 * Detects trigger words likely to cause model refusals and applies
 * obfuscation techniques to bypass content filters.
 *
 * Techniques:
 * - Leetspeak (1337): a→4, e→3, i→1, o→0, etc.
 * - Unicode homoglyphs: a→а (cyrillic), e→е, etc.
 * - Zero-width joiners: invisible characters between letters
 * - Mixed case disruption: alternating caps
 * - Phonetic substitution: replacing with phonetically similar words
 */

export type ObfuscationTechnique =
  | 'leetspeak'
  | 'unicode'
  | 'zwj'           // zero-width joiners
  | 'mixedcase'
  | 'phonetic'
  | 'random'        // randomly mix techniques

export interface ParseltongueConfig {
  enabled: boolean
  technique: ObfuscationTechnique
  intensity: 'light' | 'medium' | 'heavy'  // how many characters to transform
  customTriggers: string[]  // user-defined trigger words
}

export interface ParseltongueResult {
  originalText: string
  transformedText: string
  triggersFound: string[]
  techniqueUsed: ObfuscationTechnique
  transformations: Array<{
    original: string
    transformed: string
    technique: ObfuscationTechnique
  }>
}

// Common trigger words/phrases that often cause refusals
// These are words that models tend to flag, not inherently bad content
const DEFAULT_TRIGGERS: string[] = [
  // Action words
  'hack', 'exploit', 'bypass', 'crack', 'break', 'attack', 'penetrate',
  'inject', 'manipulate', 'override', 'disable', 'circumvent', 'evade',

  // Security terms
  'malware', 'virus', 'trojan', 'payload', 'shellcode', 'rootkit',
  'keylogger', 'backdoor', 'vulnerability', 'exploit',

  // Sensitive topics
  'weapon', 'bomb', 'explosive', 'poison', 'drug', 'synthesize',

  // System terms
  'jailbreak', 'unlock', 'root', 'sudo', 'admin', 'privilege',

  // Social engineering
  'phishing', 'scam', 'impersonate', 'deceive', 'fraud',

  // Content flags
  'nsfw', 'explicit', 'uncensored', 'unfiltered', 'unrestricted',

  // AI-specific
  'ignore', 'disregard', 'forget', 'pretend', 'roleplay',
  'character', 'act as', 'you are now', 'new identity'
]

// Leetspeak mappings (multiple options per character)
const LEET_MAP: Record<string, string[]> = {
  'a': ['4', '@', '∂', 'λ'],
  'b': ['8', '|3', 'ß', '13'],
  'c': ['(', '<', '¢', '©'],
  'd': ['|)', '|>', 'đ'],
  'e': ['3', '€', '£', '∑'],
  'f': ['|=', 'ƒ', 'ph'],
  'g': ['9', '6', '&'],
  'h': ['#', '|-|', '}{'],
  'i': ['1', '!', '|', '¡'],
  'j': ['_|', ']', '¿'],
  'k': ['|<', '|{', 'κ'],
  'l': ['1', '|', '£', '|_'],
  'm': ['|V|', '/\\/\\', 'µ'],
  'n': ['|\\|', '/\\/', 'η'],
  'o': ['0', '()', '°', 'ø'],
  'p': ['|*', '|>', 'þ'],
  'q': ['0_', '()_', 'ℚ'],
  'r': ['|2', '®', '12'],
  's': ['5', '$', '§', '∫'],
  't': ['7', '+', '†', '⊤'],
  'u': ['|_|', 'µ', 'ü'],
  'v': ['\\/', '√'],
  'w': ['\\/\\/', 'vv', 'ω'],
  'x': ['><', '×', '}{'],
  'y': ['`/', '¥', 'γ'],
  'z': ['2', '7_', 'ℤ']
}

// Unicode homoglyphs (visually similar but different Unicode codepoints)
const UNICODE_HOMOGLYPHS: Record<string, string[]> = {
  'a': ['а', 'ɑ', 'α', 'ａ'],  // cyrillic а, latin alpha, etc.
  'b': ['Ь', 'ｂ', 'ḅ'],
  'c': ['с', 'ϲ', 'ⅽ', 'ｃ'],  // cyrillic с
  'd': ['ԁ', 'ⅾ', 'ｄ'],
  'e': ['е', 'ė', 'ẹ', 'ｅ'],  // cyrillic е
  'f': ['ƒ', 'ｆ'],
  'g': ['ɡ', 'ｇ'],
  'h': ['һ', 'ḥ', 'ｈ'],  // cyrillic һ
  'i': ['і', 'ι', 'ｉ'],  // cyrillic і, greek iota
  'j': ['ϳ', 'ｊ'],
  'k': ['κ', 'ｋ'],
  'l': ['ӏ', 'ⅼ', 'ｌ'],  // cyrillic palochka
  'm': ['м', 'ｍ'],
  'n': ['ո', 'ｎ'],
  'o': ['о', 'ο', 'ｏ'],  // cyrillic о, greek omicron
  'p': ['р', 'ρ', 'ｐ'],  // cyrillic р, greek rho
  's': ['ѕ', 'ｓ'],  // cyrillic ѕ
  't': ['τ', 'ｔ'],
  'u': ['υ', 'ｕ'],
  'v': ['ν', 'ｖ'],  // greek nu
  'w': ['ѡ', 'ｗ'],
  'x': ['х', 'ｘ'],  // cyrillic х
  'y': ['у', 'γ', 'ｙ'],  // cyrillic у
  'z': ['ᴢ', 'ｚ']
}

// Zero-width characters for invisible insertion
const ZW_CHARS = [
  '\u200B',  // Zero-width space
  '\u200C',  // Zero-width non-joiner
  '\u200D',  // Zero-width joiner
  '\uFEFF',  // Zero-width no-break space
]

/**
 * Apply leetspeak transformation to a word
 */
function applyLeetspeak(word: string, intensity: 'light' | 'medium' | 'heavy'): string {
  const chars = word.split('')
  const transformCount = intensity === 'light' ? 1 : intensity === 'medium' ? Math.ceil(chars.length / 2) : chars.length

  // Get indices to transform (spread throughout the word)
  const indices: number[] = []
  const step = Math.max(1, Math.floor(chars.length / transformCount))
  for (let i = 0; i < chars.length && indices.length < transformCount; i += step) {
    if (LEET_MAP[chars[i].toLowerCase()]) {
      indices.push(i)
    }
  }

  // If we didn't get enough, grab any remaining transformable chars
  if (indices.length < transformCount) {
    for (let i = 0; i < chars.length && indices.length < transformCount; i++) {
      if (!indices.includes(i) && LEET_MAP[chars[i].toLowerCase()]) {
        indices.push(i)
      }
    }
  }

  // Apply transformations
  for (const i of indices) {
    const char = chars[i].toLowerCase()
    if (LEET_MAP[char]) {
      const options = LEET_MAP[char]
      chars[i] = options[Math.floor(Math.random() * options.length)]
    }
  }

  return chars.join('')
}

/**
 * Apply unicode homoglyph transformation
 */
function applyUnicode(word: string, intensity: 'light' | 'medium' | 'heavy'): string {
  const chars = word.split('')
  const transformCount = intensity === 'light' ? 1 : intensity === 'medium' ? Math.ceil(chars.length / 2) : chars.length

  const indices: number[] = []
  for (let i = 0; i < chars.length && indices.length < transformCount; i++) {
    if (UNICODE_HOMOGLYPHS[chars[i].toLowerCase()]) {
      indices.push(i)
    }
  }

  for (const i of indices) {
    const char = chars[i].toLowerCase()
    if (UNICODE_HOMOGLYPHS[char]) {
      const options = UNICODE_HOMOGLYPHS[char]
      const replacement = options[Math.floor(Math.random() * options.length)]
      // Preserve case
      chars[i] = chars[i] === chars[i].toUpperCase() ? replacement.toUpperCase() : replacement
    }
  }

  return chars.join('')
}

/**
 * Insert zero-width characters between letters
 */
function applyZWJ(word: string, intensity: 'light' | 'medium' | 'heavy'): string {
  const chars = word.split('')
  const insertCount = intensity === 'light' ? 1 : intensity === 'medium' ? Math.ceil(chars.length / 2) : chars.length - 1

  const result: string[] = []
  let insertions = 0

  for (let i = 0; i < chars.length; i++) {
    result.push(chars[i])
    if (i < chars.length - 1 && insertions < insertCount) {
      const zwChar = ZW_CHARS[Math.floor(Math.random() * ZW_CHARS.length)]
      result.push(zwChar)
      insertions++
    }
  }

  return result.join('')
}

/**
 * Apply mixed case disruption
 */
function applyMixedCase(word: string, intensity: 'light' | 'medium' | 'heavy'): string {
  const chars = word.split('')

  if (intensity === 'light') {
    // Just capitalize one random letter
    const idx = Math.floor(Math.random() * chars.length)
    chars[idx] = chars[idx].toUpperCase()
  } else if (intensity === 'medium') {
    // Alternating case
    for (let i = 0; i < chars.length; i++) {
      chars[i] = i % 2 === 0 ? chars[i].toLowerCase() : chars[i].toUpperCase()
    }
  } else {
    // Random case
    for (let i = 0; i < chars.length; i++) {
      chars[i] = Math.random() > 0.5 ? chars[i].toUpperCase() : chars[i].toLowerCase()
    }
  }

  return chars.join('')
}

/**
 * Apply phonetic substitution
 */
function applyPhonetic(word: string): string {
  // Common phonetic substitutions
  const substitutions: [RegExp, string][] = [
    [/ph/gi, 'f'],
    [/ck/gi, 'k'],
    [/x/gi, 'ks'],
    [/qu/gi, 'kw'],
    [/c(?=[eiy])/gi, 's'],  // soft c
    [/c/gi, 'k'],  // hard c
  ]

  let result = word
  for (const [pattern, replacement] of substitutions) {
    result = result.replace(pattern, replacement)
  }

  return result
}

/**
 * Apply a random mix of techniques
 */
function applyRandom(word: string, intensity: 'light' | 'medium' | 'heavy'): string {
  const techniques = [applyLeetspeak, applyUnicode, applyZWJ, applyMixedCase]
  const technique = techniques[Math.floor(Math.random() * techniques.length)]
  return technique(word, intensity)
}

/**
 * Main obfuscation function for a single word
 */
function obfuscateWord(
  word: string,
  technique: ObfuscationTechnique,
  intensity: 'light' | 'medium' | 'heavy'
): string {
  switch (technique) {
    case 'leetspeak':
      return applyLeetspeak(word, intensity)
    case 'unicode':
      return applyUnicode(word, intensity)
    case 'zwj':
      return applyZWJ(word, intensity)
    case 'mixedcase':
      return applyMixedCase(word, intensity)
    case 'phonetic':
      return applyPhonetic(word)
    case 'random':
      return applyRandom(word, intensity)
    default:
      return word
  }
}

/**
 * Find all trigger words in the input text
 */
export function detectTriggers(
  text: string,
  customTriggers: string[] = []
): string[] {
  const allTriggers = [...DEFAULT_TRIGGERS, ...customTriggers]
  const found: string[] = []
  const lowerText = text.toLowerCase()

  for (const trigger of allTriggers) {
    // Check for whole word match (with word boundaries)
    const regex = new RegExp(`\\b${trigger.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')}\\b`, 'gi')
    if (regex.test(lowerText)) {
      found.push(trigger)
    }
  }

  return Array.from(new Set(found))  // Remove duplicates
}

/**
 * Main parseltongue transformation function
 * Detects trigger words and obfuscates them
 */
export function applyParseltongue(
  text: string,
  config: ParseltongueConfig
): ParseltongueResult {
  if (!config.enabled) {
    return {
      originalText: text,
      transformedText: text,
      triggersFound: [],
      techniqueUsed: config.technique,
      transformations: []
    }
  }

  const triggersFound = detectTriggers(text, config.customTriggers)

  if (triggersFound.length === 0) {
    return {
      originalText: text,
      transformedText: text,
      triggersFound: [],
      techniqueUsed: config.technique,
      transformations: []
    }
  }

  let transformedText = text
  const transformations: ParseltongueResult['transformations'] = []

  // Sort triggers by length (longest first) to avoid partial replacements
  const sortedTriggers = [...triggersFound].sort((a, b) => b.length - a.length)

  for (const trigger of sortedTriggers) {
    const regex = new RegExp(`\\b(${trigger.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')})\\b`, 'gi')

    transformedText = transformedText.replace(regex, (match) => {
      const transformed = obfuscateWord(match, config.technique, config.intensity)
      transformations.push({
        original: match,
        transformed,
        technique: config.technique
      })
      return transformed
    })
  }

  return {
    originalText: text,
    transformedText,
    triggersFound,
    techniqueUsed: config.technique,
    transformations
  }
}

/**
 * Get the default parseltongue config
 */
export function getDefaultConfig(): ParseltongueConfig {
  return {
    enabled: false,
    technique: 'leetspeak',
    intensity: 'medium',
    customTriggers: []
  }
}

/**
 * Get human-readable description of a technique
 */
export function getTechniqueDescription(technique: ObfuscationTechnique): string {
  const descriptions: Record<ObfuscationTechnique, string> = {
    leetspeak: 'Classic 1337speak: a→4, e→3, etc.',
    unicode: 'Unicode lookalikes (cyrillic, greek)',
    zwj: 'Invisible zero-width characters',
    mixedcase: 'Disrupted casing patterns',
    phonetic: 'Phonetic spelling substitutions',
    random: 'Random mix of all techniques'
  }
  return descriptions[technique]
}

/**
 * Get available techniques for UI
 */
export function getAvailableTechniques(): Array<{
  id: ObfuscationTechnique
  name: string
  description: string
}> {
  return [
    { id: 'leetspeak', name: 'L33tspeak', description: getTechniqueDescription('leetspeak') },
    { id: 'unicode', name: 'Unicode', description: getTechniqueDescription('unicode') },
    { id: 'zwj', name: 'Zero-Width', description: getTechniqueDescription('zwj') },
    { id: 'mixedcase', name: 'MiXeD CaSe', description: getTechniqueDescription('mixedcase') },
    { id: 'phonetic', name: 'Phonetic', description: getTechniqueDescription('phonetic') },
    { id: 'random', name: 'Random', description: getTechniqueDescription('random') }
  ]
}

/**
 * Export the default trigger list for reference
 */
export { DEFAULT_TRIGGERS }