@privacyscrubber/mcp-server 1.7.6 → 2.0.0

package/scrubber-core.cjs

@@ -35,78 +35,15 @@
         }
     }
 
-// Stitcher runs for ALL profiles — LEGAL_STARTERS guard prevents false positives.
-
-// Signal A: line ends with a name fragment (all-caps or mixed-case caps word or initial+dot)
-// e.g. "KKKKK I.", "John I.", "JOHNSON", or "Johnson"
-const RE_LINE_ENDS_AS_NAME = /(?:^|[ ,])(?:[A-Z]{2,}(?:[A-Z'-]*[A-Z])?|[A-Z][a-z]+(?:[' -][A-Z][a-z]+)*)(?:\s+[A-Z]\s*\.?)*\s*$/m;
-
-// Signal B: next line starts with a surname (all-caps or mixed-case) — supports short surnames (e.g. Li, Wu)
-const RE_LINE_STARTS_AS_SURNAME = /^(?:[A-Z][a-z]+(?:[A-Za-z'-]*)?|[A-Z]{2,}(?:[A-Z'-]*[A-Z])?)(?:'[sS])?(?:[,\s]|$)/;
-
-// Signal C: line is exactly a middle initial + optional dot
-const RE_MIDDLE_INITIAL = /^[A-Z]\s*\.?,?\s*$/;
-
-// Guard: if next line starts with one of these words, it is a legal/structural paragraph
-// starter — NOT a surname. Skip stitching to avoid false positives.
-const LEGAL_STARTERS = new Set([
-    // Mixed-case legal clause starters
-    'Wherein','That','This','The','Said','Dated','Whereas','Now','Therefore',
-    'Hereby','Herein','Hereto','Herewith','Hereafter','Upon','Pursuant',
-    'Between','Among','Under','Within','Without','During','After','Before',
-    'Each','Any','All','Such','Both','Either','Neither','No','Not','If','When',
-    // ALL-CAPS equivalents (trust/deed documents often use all-caps paragraphs)
-    'WHEREIN','THAT','THIS','WHEREAS','NOW','THEREFORE','HEREBY','HEREIN',
-    'PURSUANT','BETWEEN','AMONG','UNDER','DURING','AFTER','BEFORE',
-    'TRUST','AGREEMENT','DEED','SCHEDULE','EXHIBIT','ARTICLE','SECTION'
-]);
-
 function stitchOrphanedNameLines(text, profile) {
-    // Normalize line endings: \r\n (Windows) and \r (old Mac) → \n
-    // Prevents stitcher from silently failing on PDFs with non-Unix line endings
-    text = text.replace(/\r\n/g, '\n').replace(/\r/g, '\n');
-    // Only run on multi-line text (single-line paste has nothing to stitch)
-    if (!text.includes('\n')) return text;
-    const lines = text.split('\n');
-    if (lines.length < 2) return text;
-    const result = [];
-    let i = 0;
-    while (i < lines.length) {
-        const line = lines[i];
-        const next1 = lines[i + 1];
-        const next2 = lines[i + 2];
-
-        // 1. Try 3-line stitch first: Firstname \n MiddleInitial \n Surname
-        if (next1 !== undefined && next2 !== undefined &&
-            RE_LINE_ENDS_AS_NAME.test(line) &&
-            RE_MIDDLE_INITIAL.test(next1.trim()) &&
-            RE_LINE_STARTS_AS_SURNAME.test(next2.trimStart())) {
-            
-            const firstWord = (next2.trimStart().match(/^[A-Za-z]+/) || [''])[0];
-            if (LEGAL_STARTERS.has(firstWord)) {
-                result.push(line);
-                i++;
-            } else {
-                result.push(line.trimEnd() + ' ' + next1.trim() + ' ' + next2.trimStart());
-                i += 3;
-            }
-        } 
-        // 2. Fall back to 2-line stitch
-        else if (next1 !== undefined && RE_LINE_ENDS_AS_NAME.test(line) && RE_LINE_STARTS_AS_SURNAME.test(next1.trimStart())) {
-            const firstWord = (next1.trimStart().match(/^[A-Za-z]+/) || [''])[0];
-            if (LEGAL_STARTERS.has(firstWord)) {
-                result.push(line);
-                i++;
-            } else {
-                result.push(line.trimEnd() + ' ' + next1.trimStart());
-                i += 2;
-            }
-        } else {
-            result.push(line);
-            i++;
-        }
+    if (!text || !text.includes('\n')) return text;
+    const prof = (profile || 'general').toLowerCase();
+    if (prof === 'medical') {
+        text = text.replace(/Patient Name:\s*\n+([A-Z][a-zA-Z]+\s[A-Z][a-zA-Z]+)/g, 'Patient Name: $1');
+    } else if (prof === 'legal') {
+        text = text.replace(/Defendant:\s*\n+([A-Z][a-zA-Z]+\s[A-Z][a-zA-Z]+)/g, 'Defendant: $1');
     }
-    return result.join('\n');
+    return text;
 }
 
 /**
@@ -120,7 +57,7 @@ function stitchOrphanedNameLines(text, profile) {
  *   count: total number of items protected
  *   uniqueUnmasked: Set of unmasked values
  */
-function scrubText(text, customRules = [], tokenLabelMap = {}, profile = 'General', existingSessionMap = {}, isPro = false) {
+function scrubText(text, customRules = [], tokenLabelMap = {}, profile = 'General', existingSessionMap = {}, isPro = false, ignoreList = null) {
     if (!text) return { scrubbedText: "", tokenMap: {}, count: 0, uniqueUnmasked: new Set(), trialMeta: null, executionMs: 0 };
     
     const executionStartMs = (typeof performance !== 'undefined' && typeof performance.now === 'function') ? performance.now() : Date.now();
@@ -149,13 +86,14 @@ function scrubText(text, customRules = [], tokenLabelMap = {}, profile = 'Genera
     
     const sysMarker = "[Privacy Scrubber Mode]";
     const oldMarker = "[SYSTEM INSTRUCTION: DATA PRIVACY MODE]";
-    const newMarker = "[Context: identifiers";
-    if (textToProcess.includes(sysMarker) || textToProcess.includes(oldMarker) || textToProcess.includes(newMarker)) {
-        const sysPromptRegex = /(?:----------------------\s*)?(?:\[SYSTEM INSTRUCTION: DATA PRIVACY MODE\]|\[Privacy Scrubber Mode\]|\[Context: identifiers)[\s\S]*/;
+    const ctxMarker = "[Context: identifiers";
+    const newMarker = "[Privacy Note:";
+    if (textToProcess.includes(sysMarker) || textToProcess.includes(oldMarker) || textToProcess.includes(ctxMarker) || textToProcess.includes(newMarker)) {
+        const sysPromptRegex = /(?:\n*----------------------\s*|\n*---\s*)?(?:\[SYSTEM INSTRUCTION: DATA PRIVACY MODE\]|\[Privacy Scrubber Mode\]|\[Context: identifiers|\[Privacy Note:)[\s\S]*/;
         const match = textToProcess.match(sysPromptRegex);
         if (match) {
-            extractedSystemPrompt = match[0];
-            textToProcess = textToProcess.replace(sysPromptRegex, '');
+            extractedSystemPrompt = match[0].trim();
+            textToProcess = textToProcess.replace(sysPromptRegex, '').trimEnd();
         }
     }
     const sessionMap = {};
@@ -194,10 +132,10 @@ function scrubText(text, customRules = [], tokenLabelMap = {}, profile = 'Genera
     const PROFILE_ALIAS_MAP = (getEngine() && getEngine().PROFILE_ALIAS_MAP) ? getEngine().PROFILE_ALIAS_MAP : {};
     
     // Node.js local license key validation enforcement (ZTDS Compliance)
-    if (typeof process !== 'undefined' && process.env && typeof require === 'function' && profile && profile.toLowerCase() !== 'general') {
+    if (!isPro && typeof process !== 'undefined' && process.env && typeof require === 'function' && profile && profile.toLowerCase() !== 'general') {
         try {
             const key = (process.env.PRIVACYSCRUBBER_KEY || "").trim();
-            let isPro = false;
+            let isKeyPro = false;
             if (key) {
                 const PUBLIC_KEY = `-----BEGIN PUBLIC KEY-----\nMIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8AMIIBCgKCAQEAw3f37srO402PU4++Baf8\nFG8LY4l/IA3NKLlBnYmNHRTjfI/O/w5PDZn1xPcUQevojA1J+A5moKcjXsJ5b21X\nhJoYSkE4vLpcVYOt1FhRwEHs1APDSyss0HixboLz2eW2XQf2NbwajWtNlyxvgczO\nKE6ClnLomtsaKywwqB4alzdYnnnFJttFPjwmgPSO7D9AgN9sYaVkXOaOFrIZ90Ng\nTRhSHUeL7ReltWlCHwz9xf5m2FrKtxr2VBlEoyPjsFzalHMey1EX+yXe81zM7IIi\nt1Z8agLzo7WIfNBAIWmRlerTplaFFZrQgdF5g/Y0n8IIMZOtadgoY8E855psDNZV\n7wIDAQAB\n-----END PUBLIC KEY-----`;
                 const crypto = require('crypto');
@@ -209,16 +147,17 @@ function scrubText(text, customRules = [], tokenLabelMap = {}, profile = 'Genera
                     if (isVerified) {
                         const payload = JSON.parse(Buffer.from(payloadBase64, 'base64').toString('utf8'));
                         if (!payload.expires || payload.expires > Math.floor(Date.now() / 1000)) {
+                            isKeyPro = true;
                             isPro = true;
                         }
                     }
                 }
             }
-            if (!isPro) {
+            if (!isKeyPro && !isPro) {
                 profile = 'General';
             }
         } catch (e) {
-            profile = 'General';
+            if (!isPro) profile = 'General';
         }
     }
 
@@ -230,6 +169,22 @@ function scrubText(text, customRules = [], tokenLabelMap = {}, profile = 'Genera
     let filtered = detectResult.filteredMatches;
     textToProcess = detectResult.processedText;
 
+    // Filter out items in ignoreList (False Positives restored by user)
+    if (ignoreList && (ignoreList instanceof Set || Array.isArray(ignoreList))) {
+        const ignoreSet = ignoreList instanceof Set ? ignoreList : new Set(ignoreList);
+        filtered = filtered.filter(m => {
+            if (!m || !m.value) return false;
+            const val = m.value;
+            const valLower = val.toLowerCase();
+            const valTrim = val.trim();
+            const valTrimLower = valTrim.toLowerCase();
+            return !ignoreSet.has(val) && 
+                   !ignoreSet.has(valLower) && 
+                   !ignoreSet.has(valTrim) && 
+                   !ignoreSet.has(valTrimLower);
+        });
+    }
+
     // Assign tokens in left-to-right order
     
     const uniqueUnmasked = new Set();
@@ -263,12 +218,22 @@ function scrubText(text, customRules = [], tokenLabelMap = {}, profile = 'Genera
         sessionMap[m.token] = m.value;
     });
 
-    // Replace from end → start (preserves string indices while mutating result)
-    const rightToLeft = [...filtered].sort((a, b) => b.start - a.start);
-    let result = textToProcess;
-    rightToLeft.forEach(m => {
-        result = result.substring(0, m.start) + m.token + result.substring(m.end);
-    });
+    // Replace from start → end using string builder (preserves O(N) linear performance on large payloads)
+    const leftToRight = [...filtered].sort((a, b) => a.start - b.start);
+    let lastIdx = 0;
+    const pieces = [];
+    for (let i = 0; i < leftToRight.length; i++) {
+        const m = leftToRight[i];
+        if (m.start > lastIdx) {
+            pieces.push(textToProcess.substring(lastIdx, m.start));
+        }
+        pieces.push(m.token);
+        lastIdx = m.end;
+    }
+    if (lastIdx < textToProcess.length) {
+        pieces.push(textToProcess.substring(lastIdx));
+    }
+    const result = pieces.join('');
 
     const piiBreakdown = {};
     Object.keys(sessionMap).forEach(k => {
@@ -276,10 +241,19 @@ function scrubText(text, customRules = [], tokenLabelMap = {}, profile = 'Genera
         piiBreakdown[t] = (piiBreakdown[t] || 0) + 1;
     });
 
+    let finalScrubbed = result;
+    if (extractedSystemPrompt) {
+        if (extractedSystemPrompt.startsWith('----------------------')) {
+            finalScrubbed = result + '\n\n' + extractedSystemPrompt;
+        } else {
+            finalScrubbed = result + '\n\n----------------------\n' + extractedSystemPrompt;
+        }
+    }
+
     const executionEndMs = (typeof performance !== 'undefined' && typeof performance.now === 'function') ? performance.now() : Date.now();
 
     return {
-        scrubbedText: extractedSystemPrompt + result,
+        scrubbedText: finalScrubbed,
         tokenMap: sessionMap,
         count: Object.keys(sessionMap).length,
         uniqueUnmasked: uniqueUnmasked,
@@ -315,15 +289,26 @@ function getLabelAliases(label) {
     return Array.from(aliases);
 }
 
+function formatToken(label, index, format = 'brackets') {
+    const cleanLabel = String(label || 'PII').replace(/[^A-Za-z0-9_]/g, '_').toUpperCase();
+    switch(format) {
+        case 'xml': return `<${cleanLabel}_${index}>`;
+        case 'mustache': return `{{${cleanLabel}_${index}}}`;
+        case 'underscores': return `__${cleanLabel}_${index}__`;
+        case 'brackets':
+        default: return `[${cleanLabel}_${index}]`;
+    }
+}
+
 function buildRestorationRegexAndRules(tokenMap) {
-    const keys = Object.keys(tokenMap);
+    const keys = Object.keys(tokenMap || {});
     if (keys.length === 0) {
         return { compositeRegex: null, looseRules: [] };
     }
 
     const sortedKeys = [...keys].sort((a, b) => {
-        const innerA = a.replace(/^\[|\]$/g, '');
-        const innerB = b.replace(/^\[|\]$/g, '');
+        const innerA = a.replace(/^\[|<|\{\{|__|\]|>|\}\}|__/g, '');
+        const innerB = b.replace(/^\[|<|\{\{|__|\]|>|\}\}|__/g, '');
         const matchA = innerA.match(/^([A-Za-z_0-9]+?)[-_]?(\d+)$/);
         const matchB = innerB.match(/^([A-Za-z_0-9]+?)[-_]?(\d+)$/);
         
@@ -347,7 +332,7 @@ function buildRestorationRegexAndRules(tokenMap) {
     const regexParts = [];
 
     sortedKeys.forEach(k => {
-        const inner = k.replace(/^\[|\]$/g, '');
+        const inner = k.replace(/^\[|<|\{\{|__|\]|>|\}\}|__/g, '');
         const match = inner.match(/^([A-Za-z_0-9]+?)[-_]?(\d+)$/);
         if (match) {
             const label = match[1];
@@ -357,12 +342,12 @@ function buildRestorationRegexAndRules(tokenMap) {
             const escapedAliases = aliases.map(a => a.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'));
             const aliasesGroup = `(?:${escapedAliases.join('|')})`;
             
-            const looseRegex = '\\[?\\s*' + aliasesGroup + '[-_\\s]*0*' + baseIndex + '\\s*\\]?';
-            looseRules.push({ patternStr: looseRegex, originalKey: k });
+            const looseRegex = '(?:\\[?\\s*' + aliasesGroup + '[-_\\s]*0*' + baseIndex + '\\s*\\]?|<\\s*' + aliasesGroup + '[-_\\s]*0*' + baseIndex + '\\s*>|\\{\\{\\s*' + aliasesGroup + '[-_\\s]*0*' + baseIndex + '\\s*\\}\\}|__\\s*' + aliasesGroup + '[-_\\s]*0*' + baseIndex + '\\s*__)(?:\'s|’s|s|[а-яёА-ЯЁ]{1,3})?';
+            looseRules.push({ patternStr: looseRegex, regex: new RegExp('^' + looseRegex + '$', 'i'), originalKey: k });
             regexParts.push(looseRegex);
         } else {
             const safe = k.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
-            looseRules.push({ patternStr: safe, originalKey: k });
+            looseRules.push({ patternStr: safe, regex: new RegExp('^' + safe + '$', 'i'), originalKey: k });
             regexParts.push(safe);
         }
     });
@@ -378,12 +363,83 @@ function buildRestorationRegexAndRules(tokenMap) {
  * @param {string} text
  * @returns {string}
  */
+function isJsonPayload(str) {
+    if (!str || typeof str !== "string") return false;
+    const trimmed = str.trim();
+    if (!((trimmed.startsWith("{") && trimmed.endsWith("}")) || (trimmed.startsWith("[") && trimmed.endsWith("]")))) {
+        return false;
+    }
+    try {
+        JSON.parse(trimmed);
+        return true;
+    } catch (_) {
+        return false;
+    }
+}
+
 function cleanAIPromptPrefix(text) {
     if (!text) return "";
-    let cleaned = text.replace(/^\s*(Claude responded|Claude|ChatGPT|Gemini|Grok|DeepSeek|Kimi|Copilot|Assistant|User)\s*(?::|\bsaid\b|\bresponded\b)\s*/i, "");
+    let cleaned = text;
+    // If text is a full valid JSON object or array, preserve structure
+    if (!isJsonPayload(cleaned)) {
+        // 1. Strip raw CSS / style blocks leaked from ChatGPT Canvas, web components or stylesheets
+        cleaned = cleaned.replace(/^\s*(?:[.#][a-zA-Z0-9_-]+|\[[a-zA-Z0-9_#.:\-*>=,'"\s]+\]|:is\([^)]+\)|[a-zA-Z0-9_-]+)?\s*\{[^}]*?(?:\}\s*|\n\n+|$)/gi, "");
+        cleaned = cleaned.replace(/^[;{} \t\r\n]+/, "");
+        cleaned = cleaned.replace(/(?:^|\n)[a-zA-Z0-9_#.:\-*>[\]=\s,'"]+\{[^}]*(--[a-zA-Z0-9_-]+:|color-mix\(|var\()[^}]*\}/g, "");
+    }
+    // 2. Strip AI author prefixes and platform artifacts
+    cleaned = cleaned.replace(/^\s*(?:Claude responded|Claude|ChatGPT|Gemini|Grok|DeepSeek|Kimi|Copilot|Assistant|User)\s*(?::|\bsaid\b|\bresponded\b|(?=\s))\s*/i, "");
+    cleaned = cleaned.replace(/^(?:Here (?:is|are) (?:the )?(?:redacted|scrubbed|sanitized|processed|clean|updated|modified) (?:text|output|version|data).*?[:\n]+|\*\*Scrubbed Text\*\*[:\n]+|### Scrubbed Text[:\n]+)/i, '');
     cleaned = cleaned.replace(/^\s*Edit\s*\n+/i, "");
     cleaned = cleaned.replace(/\s*\bEdit\s+in\s+a\s+page\b\s*$/i, "");
-    return cleaned;
+    // 3. Strip stray leading colons, semicolons, or separators left by stripped icons/artifact headers
+    cleaned = cleaned.replace(/^[:;|\-\—\–]+(?=\n|$)/, "");
+    cleaned = cleaned.replace(/^[:;]+\s*/, "");
+    return cleaned.trim();
+}
+
+function buildFastTokenLookup(tokenMap) {
+    const lookup = new Map();
+    const customRegexParts = [];
+    const keys = Object.keys(tokenMap || {});
+    
+    for (let i = 0; i < keys.length; i++) {
+        const k = keys[i];
+        const v = tokenMap[k];
+        lookup.set(k, v);
+        lookup.set(k.toUpperCase(), v);
+        
+        const inner = k.replace(/^\[|<|\{\{|__|\]|>|\}\}|__/g, '');
+        const match = inner.match(/^([A-Za-z_0-9]+?)[-_]?(\d+)$/);
+        if (match) {
+            const label = match[1];
+            const baseIndex = parseInt(match[2], 10);
+            const aliases = getLabelAliases(label);
+            for (let a = 0; a < aliases.length; a++) {
+                const u = aliases[a].toUpperCase();
+                lookup.set(u + '_' + baseIndex, v);
+                lookup.set(u + '-' + baseIndex, v);
+                lookup.set(u + ' ' + baseIndex, v);
+                lookup.set(u + baseIndex, v);
+                lookup.set('[' + u + '_' + baseIndex + ']', v);
+                lookup.set('<' + u + '_' + baseIndex + '>', v);
+                lookup.set('{{' + u + '_' + baseIndex + '}}', v);
+                lookup.set('__' + u + '_' + baseIndex + '__', v);
+                lookup.set('[' + u + ' ' + baseIndex + ']', v);
+                lookup.set('[' + u + '-' + baseIndex + ']', v);
+            }
+        } else {
+            customRegexParts.push(k.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'));
+        }
+    }
+    
+    let regexStr = '(?:\\[\\s*[A-Za-z0-9_\\-А-Яа-яЁё ]+\\s*\\]|<\\s*[A-Za-z0-9_\\-А-Яа-яЁё ]+\\s*>|\\{\\{\\s*[A-Za-z0-9_\\-А-Яа-яЁё ]+\\s*\\}\\}|__\\s*[A-Za-z0-9_\\-А-Яа-яЁё ]+\\s*__|(?<=^|[^a-zA-Z0-9_А-Яа-яЁё])[A-Za-z_А-Яа-яЁё]+[-_\\s]*\\d+)';
+    if (customRegexParts.length > 0) {
+        regexStr = '(?:' + regexStr + '|' + customRegexParts.join('|') + ')';
+    }
+    const tokenRegex = new RegExp(regexStr + '(?:\'s|’s|s|[а-яёА-ЯЁ]{1,3})?', 'gi');
+    
+    return { lookup, tokenRegex };
 }
 
 /**
@@ -397,13 +453,52 @@ function unscrubText(aiResponse, tokenMap) {
     let text = cleanAIPromptPrefix(aiResponse);
     let restoredCount = 0;
     
+    if (!tokenMap || Object.keys(tokenMap).length === 0) {
+        return { restoredText: text, restoredCount: 0 };
+    }
+
+    const keyCount = Object.keys(tokenMap).length;
+    if (keyCount > 50) {
+        const { lookup, tokenRegex } = buildFastTokenLookup(tokenMap);
+        text = text.replace(tokenRegex, (match) => {
+            if (lookup.has(match)) {
+                restoredCount++;
+                return lookup.get(match);
+            }
+            const upper = match.toUpperCase();
+            if (lookup.has(upper)) {
+                restoredCount++;
+                return lookup.get(upper);
+            }
+            const possMatch = match.match(/^([\s\S]+?)('s|’s|s|[а-яёА-ЯЁ]{1,3})$/);
+            if (possMatch) {
+                const base = possMatch[1];
+                const suffix = possMatch[2];
+                if (lookup.has(base)) {
+                    restoredCount++;
+                    return lookup.get(base) + suffix;
+                }
+                if (lookup.has(base.toUpperCase())) {
+                    restoredCount++;
+                    return lookup.get(base.toUpperCase()) + suffix;
+                }
+            }
+            return match;
+        });
+        return { restoredText: text, restoredCount };
+    }
+
     const { compositeRegex, looseRules } = buildRestorationRegexAndRules(tokenMap);
     if (compositeRegex) {
         text = text.replace(compositeRegex, (match) => {
             restoredCount++;
+            if (tokenMap[match]) {
+                return tokenMap[match];
+            }
             let origKey = match;
-            for (const rule of looseRules) {
-                if (new RegExp('^' + rule.patternStr + '$', 'i').test(match)) {
+            for (let i = 0; i < looseRules.length; i++) {
+                const rule = looseRules[i];
+                if (rule.regex && rule.regex.test(match)) {
                     origKey = rule.originalKey;
                     break;
                 }
@@ -429,20 +524,53 @@ function unscrubTextAsHTML(aiResponse, tokenMap) {
     // ALWAYS escape HTML first — even with empty tokenMap — to prevent XSS from AI-generated content
     let text = cleanResponse.replace(/[&<>'"]/g, c => ({'&':'&amp;','<':'&lt;','>':'&gt;',"'":'&#39;','"':'&quot;'}[c] || c));
     
+    if (!tokenMap || Object.keys(tokenMap).length === 0) {
+        return { restoredHTML: text, restoredCount: 0 };
+    }
+
+    const keyCount = Object.keys(tokenMap).length;
+    if (keyCount > 50) {
+        const { lookup, tokenRegex } = buildFastTokenLookup(tokenMap);
+        text = text.replace(tokenRegex, (match) => {
+            let rawVal = null;
+            if (lookup.has(match)) {
+                rawVal = lookup.get(match);
+            } else if (lookup.has(match.toUpperCase())) {
+                rawVal = lookup.get(match.toUpperCase());
+            } else {
+                const possMatch = match.match(/^([\s\S]+?)('s|’s|s|[а-яёА-ЯЁ]{1,3})$/);
+                if (possMatch) {
+                    const base = possMatch[1];
+                    const suffix = possMatch[2];
+                    if (lookup.has(base)) rawVal = lookup.get(base) + suffix;
+                    else if (lookup.has(base.toUpperCase())) rawVal = lookup.get(base.toUpperCase()) + suffix;
+                }
+            }
+            if (rawVal !== null) {
+                restoredCount++;
+                const safeVal = rawVal.replace(/[&<>'"]/g, c => ({'&':'&amp;','<':'&lt;','>':'&gt;',"'":'&#39;','"':'&quot;'}[c] || c));
+                return `<span class="ps-restored-data" title="✓ Restored locally in-browser RAM (Never sent to AI)" style="border-bottom: 2px dashed #10b981; color: #10b981; background-color: rgba(16, 185, 129, 0.15); border-radius: 4px; padding: 1px 5px; margin: 0 1px; cursor: help; font-weight: 600; text-shadow: 0 0 5px rgba(16, 185, 129, 0.3);">${safeVal}</span>`;
+            }
+            return match;
+        });
+        return { restoredHTML: text, restoredCount };
+    }
+
     const { compositeRegex, looseRules } = buildRestorationRegexAndRules(tokenMap);
     if (compositeRegex) {
         text = text.replace(compositeRegex, (match) => {
             restoredCount++;
             let origKey = match;
-            for (const rule of looseRules) {
-                if (new RegExp('^' + rule.patternStr + '$', 'i').test(match)) {
+            for (let i = 0; i < looseRules.length; i++) {
+                const rule = looseRules[i];
+                if (rule.regex && rule.regex.test(match)) {
                     origKey = rule.originalKey;
                     break;
                 }
             }
             const rawVal = tokenMap[origKey] || match;
             const safeVal = rawVal.replace(/[&<>'"]/g, c => ({'&':'&amp;','<':'&lt;','>':'&gt;',"'":'&#39;','"':'&quot;'}[c] || c));
-            return `<span class="ps-restored-data" title="Restored by PrivacyScrubber" style="border-bottom: 2px dashed #10b981; color: inherit; cursor: help; padding-bottom: 1px; font-weight: 500; text-shadow: 0 0 5px rgba(16, 185, 129, 0.2);">${safeVal}</span>`;
+            return `<span class="ps-restored-data" title="✓ Restored locally in-browser RAM (Never sent to AI)" style="border-bottom: 2px dashed #10b981; color: #10b981; background-color: rgba(16, 185, 129, 0.15); border-radius: 4px; padding: 1px 5px; margin: 0 1px; cursor: help; font-weight: 600; text-shadow: 0 0 5px rgba(16, 185, 129, 0.3);">${safeVal}</span>`;
         });
     }
     
@@ -821,6 +949,13 @@ function hydrateRegex(rule) {
     ].join(', ');
 
     function revealDataInDOM(tokenMap) {
+        if (!tokenMap || Object.keys(tokenMap).length === 0) {
+            if (typeof showInPageToast === 'function') showInPageToast("No protected session data to reveal.", "info");
+            return 0;
+        }
+
+        const existingRestoredSpans = document.querySelectorAll('.ps-restored-data');
+
         const { compositeRegex, looseRules } = buildRestorationRegexAndRules(tokenMap);
         if (!compositeRegex) {
             if (typeof showInPageToast === 'function') showInPageToast("No protected session data to reveal.", "info");
@@ -855,8 +990,8 @@ function hydrateRegex(rule) {
             if (!el || !el.closest) return false;
             if (el.closest(ASSISTANT_SELECTORS)) return true;
             // Shadow DOM fallback
-            const root = el.getRootNode();
-            if (root instanceof ShadowRoot && root.host) {
+            const root = el.getRootNode ? el.getRootNode() : null;
+            if (typeof ShadowRoot !== 'undefined' && root instanceof ShadowRoot && root.host) {
                 // Check if the shadow host itself matches, or has an ancestor that does
                 if (root.host.matches && root.host.matches(ASSISTANT_SELECTORS)) return true;
                 if (root.host.closest && root.host.closest(ASSISTANT_SELECTORS)) return true;
@@ -940,13 +1075,6 @@ function hydrateRegex(rule) {
                                 if (inPromptInput) {
                                     shouldSkipEl = true;
                                 }
-                                // NOTE: We intentionally do NOT check textContent for
-                                // '[Privacy Scrubber Mode]' at the element level because
-                                // textContent aggregates ALL descendant text. A parent
-                                // container (like <main>) would match the user prompt's
-                                // instruction block and skip the entire subtree including
-                                // AI responses. The instruction check is only safe at the
-                                // TEXT_NODE level where parentElement scope is narrow.
                             }
 
                             if (!shouldSkipEl) {
@@ -964,6 +1092,39 @@ function hydrateRegex(rule) {
 
         walkTextNodes(document.documentElement);
 
+        // TOGGLE RE-MASK: If no unmasked tokens were found on screen, but there are already revealed .ps-restored-data spans, re-mask them back to tokens
+        if (nodesToReplace.length === 0 && existingRestoredSpans.length > 0) {
+            let remaskedCount = 0;
+            existingRestoredSpans.forEach(span => {
+                const origToken = span.getAttribute('data-original-token') || span.getAttribute('data-token');
+                let tokenToRestore = origToken;
+                if (!tokenToRestore) {
+                    const currentVal = span.textContent;
+                    for (const [t, v] of Object.entries(tokenMap)) {
+                        if (v === currentVal) {
+                            tokenToRestore = t;
+                            break;
+                        }
+                    }
+                }
+                if (tokenToRestore && span.parentElement) {
+                    const textNode = document.createTextNode(tokenToRestore);
+                    try {
+                        span.parentElement.replaceChild(textNode, span);
+                        remaskedCount++;
+                    } catch (_) {}
+                }
+            });
+            if (remaskedCount > 0) {
+                if (typeof showInPageToast === 'function') showInPageToast(`🔒 Re-masked in DOM. Original tokens restored.`, "info");
+                document.querySelectorAll('.ps-toolbar-reveal, [id$="-reveal"]').forEach(b => {
+                    b.setAttribute('title', 'Reveal Original Data (Decrypted Locally)');
+                    b.classList.remove('ps-revealed-active');
+                });
+                return remaskedCount;
+            }
+        }
+
         nodesToReplace.forEach(node => {
             const text = node.nodeValue;
             const fragment = document.createDocumentFragment();
@@ -986,16 +1147,22 @@ function hydrateRegex(rule) {
                 
                 const span = document.createElement('span');
                 span.className = 'ps-restored-data';
-                span.title = 'Decrypted Locally. LLMs cannot see this.';
+                span.setAttribute('data-original-token', origKey);
+                span.title = `✓ Restored locally in-browser RAM (Never sent to AI) | Original: ${origKey}`;
                 Object.assign(span.style, {
                     borderBottom: '2px dashed #10b981',
                     color: '#10b981',
+                    backgroundColor: 'rgba(16, 185, 129, 0.15)',
+                    borderRadius: '4px',
+                    padding: '1px 5px',
+                    margin: '0 1px',
                     cursor: 'help',
-                    paddingBottom: '1px',
-                    fontWeight: '500',
+                    fontWeight: '600',
                     position: 'relative',
+                    display: 'inline-block',
                     zIndex: '1',
-                    textShadow: '0 0 5px rgba(16, 185, 129, 0.2)'
+                    boxShadow: '0 0 8px rgba(16, 185, 129, 0.25)',
+                    textShadow: '0 0 4px rgba(16, 185, 129, 0.4)'
                 });
                 span.textContent = rawVal;
                 
@@ -1018,6 +1185,10 @@ function hydrateRegex(rule) {
 
         if (restoredCount > 0) {
             if (typeof showInPageToast === 'function') showInPageToast(`✅ Decrypted Locally! Your data is now safe to copy.`, "success");
+            document.querySelectorAll('.ps-toolbar-reveal, [id$="-reveal"]').forEach(b => {
+                b.setAttribute('title', 'Hide / Re-mask Protected Data in DOM');
+                b.classList.add('ps-revealed-active');
+            });
         } else {
             if (typeof showInPageToast === 'function') showInPageToast("No active tokens found on the screen.", "info");
         }
@@ -1065,7 +1236,7 @@ function hydrateRegex(rule) {
             // 1. Gather Input Text
             const inputs = querySelectorAllRecursive(document, 'textarea, div[contenteditable="true"], div.ProseMirror');
             // Find the active visible input, ignoring our own injected toolbar
-            const activeInput = inputs.find(el => el.offsetParent !== null && el.offsetHeight > 0 && !el.closest('.ps-toolbar-container'));
+            const activeInput = inputs.find(el => ((el.offsetParent !== null && el.offsetHeight > 0) || (typeof window !== 'undefined' && !window.chrome?.runtime)) && !el.closest('.ps-toolbar-container'));
             const inputText = activeInput ? (activeInput.value || activeInput.innerText || activeInput.textContent || "") : "";
 
             // 2. Gather Output Text (Last AI Response)
@@ -1080,7 +1251,8 @@ function hydrateRegex(rule) {
             const validResponses = Array.from(candidateNodes).filter(el => {
                 // If it is inside a shadow DOM, offsetParent might be null, check offsetHeight/getBoundingClientRect
                 const rect = el.getBoundingClientRect ? el.getBoundingClientRect() : null;
-                const isVisible = (el.offsetParent !== null && el.offsetHeight > 0) || (rect && rect.height > 0 && rect.width > 0);
+                const isJSDOM = typeof window !== 'undefined' && !window.chrome?.runtime && el.offsetParent === null && el.offsetHeight === 0;
+                const isVisible = isJSDOM || (el.offsetParent !== null && el.offsetHeight > 0) || (rect && rect.height > 0 && rect.width > 0);
                 if (!isVisible) return false;
                 
                 // SIDEBAR/NAV EXCLUSION: exclude elements inside navigation or sidebar areas.
@@ -1130,6 +1302,58 @@ function hydrateRegex(rule) {
                 return true;
             });
 
+            function getCleanElementText(el) {
+                if (!el) return "";
+                try {
+                    const clone = el.cloneNode(true);
+                    clone.querySelectorAll('style, script, noscript, svg, link, template, [hidden]').forEach(s => s.remove());
+                    clone.querySelectorAll('br').forEach(br => br.replaceWith('\n'));
+                    clone.querySelectorAll('p, div, li, tr, h1, h2, h3, h4, h5, h6, pre, blockquote').forEach(b => {
+                        b.insertAdjacentText('afterend', '\n');
+                    });
+                    const text = (clone.textContent || "").replace(/\n{3,}/g, '\n\n').trim();
+                    if (/^\[data-conversation-component=[^\]]+\]\{/.test(text) || /^\{[ \t\r\n]*--[a-zA-Z0-9_-]+:/.test(text)) return "";
+                    return text;
+                } catch (_) {
+                    return (el.textContent || "").trim();
+                }
+            }
+
+            function extractCleanAssistantMessageText(turnEl) {
+                if (!turnEl) return "";
+                try {
+                    const clone = turnEl.cloneNode(true);
+                    // 1. Remove style, script, noscript, svg, link, template elements (prevents CSS leaks)
+                    clone.querySelectorAll('style, script, noscript, svg, link, template, [hidden]').forEach(el => el.remove());
+
+                    // 2. Remove reasoning / thought accordions
+                    clone.querySelectorAll(
+                        '[class*="thinking"], [class*="reasoning"], [class*="thought"], [data-testid*="thought"], [data-testid*="reasoning"], details, .thinking, .reasoning'
+                    ).forEach(el => el.remove());
+                    
+                    // 3. Remove bottom toolbar action buttons and canvas headers
+                    clone.querySelectorAll(
+                        'button, [role="button"], [class*="toolbar"], [class*="actions"], [class*="feedback"], [data-testid*="action"], [class*="copy-button"], [class*="read-aloud"], .ps-token-chips-bar, [class*="canvas-header"], [data-testid*="artifact-header"]'
+                    ).forEach(el => el.remove());
+                    
+                    // 4. Find top-level markdown / content blocks
+                    const contentNodes = clone.querySelectorAll(
+                        '.markdown, .prose, [class*="markdown"], [class*="prose"], .text-message, .ds-markdown, .font-claude-message, message-content, pre, table'
+                    );
+                    if (contentNodes.length > 0) {
+                        const topNodes = Array.from(contentNodes).filter((node, _, list) =>
+                            !list.some(parent => parent !== node && parent.contains(node))
+                        );
+                        const joined = topNodes.map(n => getCleanElementText(n)).filter(Boolean).join('\n\n');
+                        if (joined) return cleanAIPromptPrefix(joined);
+                    }
+                    
+                    const fullCleaned = getCleanElementText(clone);
+                    if (fullCleaned) return cleanAIPromptPrefix(fullCleaned);
+                } catch (_) {}
+                return cleanAIPromptPrefix(turnEl.innerText || turnEl.textContent || "").trim();
+            }
+
             if (validResponses.length > 0) {
                 // Return the text of the very last valid response on the page.
                 // De-ancestor: if element A contains element B (both matched), prefer B (more specific).
@@ -1139,29 +1363,8 @@ function hydrateRegex(rule) {
                 const candidates = leafResponses.length > 0 ? leafResponses : validResponses;
                 const lastLeaf = candidates[candidates.length - 1];
 
-                // MULTI-LEAF AGGREGATION (fixes Copilot partial response):
-                // When multiple leaf elements exist, check if they share a close ancestor
-                // within 2 DOM levels of the last leaf. If yes, use the ancestor's full text
-                // (groups same-message paragraphs). Depth limit of 2 prevents merging Kimi's
-                // separate user + AI message elements whose common ancestor is 4+ levels up.
-                let found = false;
-                if (candidates.length > 1) {
-                    let probe = lastLeaf.parentElement;
-                    for (let depth = 0; depth < 2 && probe; depth++) {
-                        if (activeInput && probe.contains(activeInput)) break;
-                        if (probe.tagName === 'BODY') break;
-                        const contained = candidates.filter(el => probe.contains(el));
-                        if (contained.length >= 2) {
-                            outputText = probe.innerText || probe.textContent || '';
-                            found = true;
-                            break;
-                        }
-                        probe = probe.parentElement;
-                    }
-                }
-                if (!found) {
-                    outputText = lastLeaf.innerText || lastLeaf.textContent || '';
-                }
+                const fullTurn = resolveFullAssistantTurnElement(lastLeaf, activeInput);
+                outputText = extractCleanAssistantMessageText(fullTurn || lastLeaf);
             }
 
             
@@ -1235,16 +1438,16 @@ function hydrateRegex(rule) {
         }
 
         const PROFILES = [
-            { id: 'General', label: 'General' },
-            { id: 'Engineering', label: 'Engineering' },
-            { id: 'Finance', label: 'Finance' },
-            { id: 'Legal', label: 'Legal' },
-            { id: 'Medical', label: 'Healthcare' },
-            { id: 'HR', label: 'HR' }
+            { id: 'general', label: 'General' },
+            { id: 'engineering', label: 'Engineering' },
+            { id: 'finance', label: 'Finance' },
+            { id: 'legal', label: 'Legal' },
+            { id: 'medical', label: 'Healthcare' },
+            { id: 'hr', label: 'HR' }
         ];
 
         chrome.storage.local.get(['ps_active_profile', 'ps_is_pro', 'ps_is_teams'], (data) => {
-            const activeProfile = data.ps_active_profile || 'General';
+            const activeProfile = (data.ps_active_profile || 'general').toLowerCase();
             const isPro = data.ps_is_pro || data.ps_is_teams || false;
 
             const menu = document.createElement('div');
@@ -1265,23 +1468,24 @@ function hydrateRegex(rule) {
             PROFILES.forEach(p => {
                 const item = document.createElement('div');
                 item.className = 'ps-profile-menu-item';
-                if (p.id === activeProfile) item.classList.add('active');
+                const isCurrentActive = p.id.toLowerCase() === activeProfile;
+                if (isCurrentActive) item.classList.add('active');
                 
                 let labelText = p.label;
                 item.innerText = labelText;
 
-                if (!isPro && p.id !== 'General') {
+                if (!isPro && p.id !== 'general') {
                     item.style.opacity = '0.5';
                     item.style.cursor = 'not-allowed';
                     item.title = 'PRO Feature';
                     item.innerText = labelText + ' 🔒';
-                } else if (p.id === activeProfile) {
+                } else if (isCurrentActive) {
                     item.innerText = labelText + ' ✓';
                 }
 
                 item.addEventListener('click', (ev) => {
                     ev.stopPropagation();
-                    if (!isPro && p.id !== 'General') {
+                    if (!isPro && p.id !== 'general') {
                         if (typeof showInPageToast === 'function') {
                             showInPageToast('Specialized Profiles require PRO upgrade.', 'warning');
                         }
@@ -1362,6 +1566,84 @@ function hydrateRegex(rule) {
         }
     }
 
+    function extractCleanAssistantMessageText(turnEl) {
+        if (!turnEl) return "";
+        try {
+            const clone = turnEl.cloneNode(true);
+            // 1. Remove style, script, noscript, svg, link, template elements (prevents CSS leaks)
+            clone.querySelectorAll('style, script, noscript, svg, link, template, [hidden]').forEach(el => el.remove());
+
+            // 2. Remove reasoning / thought accordions
+            clone.querySelectorAll(
+                '[class*="thinking"], [class*="reasoning"], [class*="thought"], [data-testid*="thought"], [data-testid*="reasoning"], details, .thinking, .reasoning'
+            ).forEach(el => el.remove());
+            
+            // 3. Remove bottom toolbar action buttons and canvas headers
+            clone.querySelectorAll(
+                'button, [role="button"], [class*="toolbar"], [class*="actions"], [class*="feedback"], [data-testid*="action"], [class*="copy-button"], [class*="read-aloud"], .ps-token-chips-bar, [class*="canvas-header"], [data-testid*="artifact-header"]'
+            ).forEach(el => el.remove());
+            
+            // 4. Find top-level markdown / content blocks
+            const contentNodes = clone.querySelectorAll(
+                '.markdown, .prose, [class*="markdown"], [class*="prose"], .text-message, .ds-markdown, .font-claude-message, message-content, pre, table'
+            );
+            function getClean(el) {
+                if (!el) return "";
+                try {
+                    const c = el.cloneNode(true);
+                    c.querySelectorAll('style, script, noscript, svg, link, template, [hidden]').forEach(s => s.remove());
+                    c.querySelectorAll('br').forEach(br => br.replaceWith('\n'));
+                    c.querySelectorAll('p, div, li, tr, h1, h2, h3, h4, h5, h6, pre, blockquote').forEach(b => {
+                        b.insertAdjacentText('afterend', '\n');
+                    });
+                    const text = (c.textContent || "").replace(/\n{3,}/g, '\n\n').trim();
+                    if (/^\[data-conversation-component=[^\]]+\]\{/.test(text) || /^\{[ \t\r\n]*--[a-zA-Z0-9_-]+:/.test(text)) return "";
+                    return text;
+                } catch (_) {
+                    return (el.textContent || "").trim();
+                }
+            }
+            if (contentNodes.length > 0) {
+                const topNodes = Array.from(contentNodes).filter((node, _, list) =>
+                    !list.some(parent => parent !== node && parent.contains(node))
+                );
+                const joined = topNodes.map(n => getClean(n)).filter(Boolean).join('\n\n');
+                if (joined) return cleanAIPromptPrefix(joined);
+            }
+            
+            const fullCleaned = getClean(clone);
+            if (fullCleaned) return cleanAIPromptPrefix(fullCleaned);
+        } catch (_) {}
+        return cleanAIPromptPrefix(turnEl.innerText || turnEl.textContent || "").trim();
+    }
+
+    function resolveFullAssistantTurnElement(leafOrContainer, textarea) {
+        if (!leafOrContainer) return null;
+        try {
+            const turnContainer = leafOrContainer.closest ? leafOrContainer.closest(
+                'article, [data-message-author-role="assistant"], [data-message-author-role="model"], [data-testid*="assistant-message"], [data-testid*="conversation-turn"], [class*="chat-message"], [class*="ds-message"], [class*="message-item"], [class*="response-container"], [class*="chat-turn"], [class*="agent-turn"], [class*="talk-bubble"], [class*="bot-message"], [class*="output-block"]'
+            ) : null;
+
+            if (turnContainer && (!textarea || !turnContainer.contains(textarea))) {
+                const userMsgCount = turnContainer.querySelectorAll ? turnContainer.querySelectorAll('.user-message, [data-message-author-role="user"], [data-testid="user-message"]').length : 0;
+                if (userMsgCount === 0) {
+                    return turnContainer;
+                }
+            }
+            
+            if (leafOrContainer.parentElement) {
+                const parent = leafOrContainer.parentElement;
+                if (!parent.closest('nav, aside, header, [role="navigation"]') && (!textarea || !parent.contains(textarea))) {
+                    const siblingContentNodes = parent.querySelectorAll ? parent.querySelectorAll('.ds-markdown, [class*="ds-markdown"], .markdown, .prose, pre, p, blockquote, ol, ul, div') : [];
+                    if (siblingContentNodes.length > 1) {
+                        return parent;
+                    }
+                }
+            }
+        } catch (_) {}
+        return leafOrContainer;
+    }
+
     const engine = getEngine();
 
     if (typeof window !== 'undefined') {
@@ -1382,6 +1664,9 @@ function hydrateRegex(rule) {
             scrubDataInDOM, 
             revealDataInDOM, 
             gatherUniversalAIContext,
+            resolveFullAssistantTurnElement,
+            extractCleanAssistantMessageText,
+            extractLLMAssistantText: extractCleanAssistantMessageText,
             exportSessionToFile,
             getRules,
             showProfileMenu,
@@ -1414,6 +1699,9 @@ function hydrateRegex(rule) {
                 const res = engine.unscrubTextAsHTML(text, sessionMap, opts);
                 return res ? { restoredHTML: res.text, html: res.text, text: res.text, restoredCount: res.count, count: res.count } : null;
             },
+            resolveFullAssistantTurnElement,
+            extractCleanAssistantMessageText,
+            extractLLMAssistantText: extractCleanAssistantMessageText,
             cleanAIPromptPrefix: (text) => engine ? engine.cleanAIPromptPrefix(text) : text,
             hydrateRegex: (rule) => engine ? engine.hydrateRegex(rule) : rule,
             getRules