Fix auto-detect false positives for configured values inside compound matches

Home path, shell prompt, git remote, and env var assignment patterns match
a compound string (e.g. "/home/johnsmith") that only *contains* a configured
identity/mapping value ("johnsmith"), not equals it. The exact-match
configured.has() check never recognized these as already handled, so the
"potential PII" warning fired on every message containing one of these
shapes regardless of configuration, and the fallback auto-redact step
overwrote already-substituted text with a generic placeholder.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S3nGKyoLaRghi1okiNyRKD

manifest.firefox.json

@@ -1,7 +1,7 @@
 {
   "manifest_version": 3,
   "name": "Silent Send",
-  "version": "0.9.46",
+  "version": "0.9.47",
   "description": "Intercepts personal info and substitutes it with user-defined replacements before sending to AI services.",
   "browser_specific_settings": {
     "gecko": {

manifest.json

@@ -1,7 +1,7 @@
 {
   "manifest_version": 3,
   "name": "Silent Send",
-  "version": "0.9.46",
+  "version": "0.9.47",
   "description": "Intercepts personal info and substitutes it with user-defined replacements before sending to AI services.",
   "permissions": [
     "storage",

package.json

@@ -1,6 +1,6 @@
 {
   "name": "silent-send",
-  "version": "0.9.46",
+  "version": "0.9.47",
   "private": true,
   "license": "MIT",
   "description": "Browser extension that substitutes personal data before sending to AI services",

sign-firefox.sh

@@ -86,7 +86,9 @@ for attempt in $(seq 1 $MAX_ATTEMPTS); do
   echo "=== Attempt $attempt: signing v$NEW_VERSION ==="
 
   # Capture output to check for specific errors
-  OUTPUT=$(npx web-ext sign \
+  # NODE_OPTIONS forces IPv4 DNS resolution first — Node.js 18+ defaults to IPv6,
+  # which silently fails when the system can't reach addons.mozilla.org over IPv6.
+  OUTPUT=$(NODE_OPTIONS='--dns-result-order=ipv4first' npx web-ext sign \
     --no-config-discovery \
     --source-dir "$SCRIPT_DIR/dist/firefox" \
     --artifacts-dir "$SCRIPT_DIR/dist/firefox-signed" \

src/content/content.js

@@ -51,8 +51,8 @@
 
     for (const m of sorted) {
       if (!m.enabled || !m.real?.trim() || !m.substitute?.trim()) continue;
-      const pattern = wordBoundary(m.real);
-      const regex = new RegExp(pattern, m.caseSensitive ? 'g' : 'gi');
+      const escaped = esc(m.real);
+      const regex = new RegExp(escaped, m.caseSensitive ? 'g' : 'gi');
       let match;
       while ((match = regex.exec(result)) !== null) {
         replacements.push({
@@ -71,8 +71,8 @@
     const sorted = [...maps].sort((a, b) => b.substitute.length - a.substitute.length);
     for (const m of sorted) {
       if (!m.enabled || !m.real?.trim() || !m.substitute?.trim()) continue;
-      const pattern = wordBoundary(m.substitute);
-      const regex = new RegExp(pattern, m.caseSensitive ? 'g' : 'gi');
+      const escaped = esc(m.substitute);
+      const regex = new RegExp(escaped, m.caseSensitive ? 'g' : 'gi');
       result = result.replace(regex, m.real);
     }
     return result;
@@ -82,15 +82,6 @@
     return str.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
   }
 
-  // Add \b only on word-character edges so "not" doesn't match inside "nothing",
-  // while leaving non-word edges (e.g. "@foo") alone since they self-delimit.
-  function wordBoundary(str) {
-    const escaped = esc(str);
-    const left = /^\w/.test(str) ? '\\b' : '';
-    const right = /\w$/.test(str) ? '\\b' : '';
-    return left + escaped + right;
-  }
-
   // ============================================================
   // Smart Pattern Engine (inline for page world)
   // ============================================================
@@ -358,6 +349,22 @@
 
   const CONTEXT_WORDS_RE = /\b(?:born|birthday|dob|birth|passport|license|driver|ssn|social\s*security|address|zip|postal|date\s+of\s+birth)\b/i;
 
+  // Some patterns (home paths, shell prompts, git remotes, env var assignments)
+  // match a compound string that only *contains* a configured value rather
+  // than equaling it exactly (e.g. "/home/johnsmith" contains the configured
+  // username "johnsmith"). A plain Set.has() exact-match check never fires
+  // for these, so the same configured value gets flagged as "unconfigured"
+  // every time. Length-gated to avoid short tokens suppressing unrelated
+  // findings via accidental substring collision.
+  function matchesConfiguredValue(value, configured) {
+    const lower = value.toLowerCase();
+    if (configured.has(lower)) return true;
+    for (const c of configured) {
+      if (c.length >= 3 && lower.includes(c)) return true;
+    }
+    return false;
+  }
+
   // Common English words that are capitalized but aren't proper nouns.
   // Used by the proper noun heuristic to reduce false positives.
   const COMMON_CAPITALIZED = new Set([
@@ -496,7 +503,7 @@
       let m;
       while ((m = pat.re.exec(text)) !== null) {
         const val = m[0];
-        if (configured.has(val.toLowerCase())) continue;
+        if (matchesConfiguredValue(val, configured)) continue;
         if (pat.skip && pat.skip.test(val)) continue;
         findings.push({ name: pat.name, value: val, hint: pat.hint, category: pat.cat });
       }
@@ -1052,8 +1059,8 @@
     const pairs = getRevealPairs();
     let result = text;
     for (const p of pairs) {
-      const pattern = wordBoundary(p.from);
-      const regex = new RegExp(pattern, p.caseSensitive ? 'g' : 'gi');
+      const escaped = esc(p.from);
+      const regex = new RegExp(escaped, p.caseSensitive ? 'g' : 'gi');
       result = result.replace(regex, p.to);
     }
     return result;
@@ -1094,9 +1101,9 @@
     const pairs = getRevealPairs();
     let result = text;
     for (const p of pairs) {
-      const pattern = wordBoundary(p.to);
-      const regex = new RegExp(pattern, p.caseSensitive ? 'g' : 'gi');
-      result = result.replace(regex, p.from);
+      const escaped = esc(p.to);   // p.to is the real value
+      const regex = new RegExp(escaped, p.caseSensitive ? 'g' : 'gi');
+      result = result.replace(regex, p.from);  // p.from is the substitute
     }
     return result;
   }

src/lib/auto-detect.js

@@ -129,6 +129,22 @@ const PII_PATTERNS = [
 // Context words that make ambiguous patterns more likely to be PII
 const CONTEXT_WORDS = /\b(?:born|birthday|dob|birth|passport|license|driver|ssn|social\s*security|address|home|live|lives|reside|zip|postal)\b/i;
 
+// Some patterns (home paths, shell prompts, git remotes, env var assignments)
+// match a compound string that only *contains* a configured value rather than
+// equaling it exactly (e.g. "/home/johnsmith" contains the configured
+// username "johnsmith"). A plain Set.has() exact-match check never fires for
+// these, so the same configured value gets flagged as "unconfigured" every
+// time. Length-gated to avoid short tokens (e.g. a 2-char substitute)
+// suppressing unrelated findings via accidental substring collision.
+function matchesConfiguredValue(value, configured) {
+  const lower = value.toLowerCase();
+  if (configured.has(lower)) return true;
+  for (const c of configured) {
+    if (c.length >= 3 && lower.includes(c)) return true;
+  }
+  return false;
+}
+
 const AutoDetect = {
   /**
    * Scan text for potential unconfigured PII.
@@ -178,7 +194,7 @@ const AutoDetect = {
         const value = match[0];
 
         // Skip if already configured
-        if (configured.has(value.toLowerCase())) continue;
+        if (matchesConfiguredValue(value, configured)) continue;
 
         // Skip excluded values (like 127.0.0.1)
         if (pattern.exclude && pattern.exclude.test(value)) continue;

src/lib/substitution-engine.js

@@ -23,8 +23,8 @@ const SubstitutionEngine = {
     for (const mapping of sorted) {
       if (!mapping.enabled || !mapping.real?.trim() || !mapping.substitute?.trim()) continue;
 
-      const pattern = this._wordBoundaryPattern(mapping.real);
-      const regex = new RegExp(pattern, mapping.caseSensitive ? 'g' : 'gi');
+      const escaped = this._escapeRegex(mapping.real);
+      const regex = new RegExp(escaped, mapping.caseSensitive ? 'g' : 'gi');
       let match;
 
       while ((match = regex.exec(result)) !== null) {
@@ -56,8 +56,8 @@ const SubstitutionEngine = {
     for (const mapping of sorted) {
       if (!mapping.enabled || !mapping.real?.trim() || !mapping.substitute?.trim()) continue;
 
-      const pattern = this._wordBoundaryPattern(mapping.substitute);
-      const regex = new RegExp(pattern, mapping.caseSensitive ? 'g' : 'gi');
+      const escaped = this._escapeRegex(mapping.substitute);
+      const regex = new RegExp(escaped, mapping.caseSensitive ? 'g' : 'gi');
       result = result.replace(regex, mapping.real);
     }
 
@@ -73,8 +73,8 @@ const SubstitutionEngine = {
     for (const mapping of mappings) {
       if (!mapping.enabled || !mapping.real?.trim()) continue;
 
-      const pattern = this._wordBoundaryPattern(mapping.real);
-      const regex = new RegExp(pattern, mapping.caseSensitive ? 'g' : 'gi');
+      const escaped = this._escapeRegex(mapping.real);
+      const regex = new RegExp(escaped, mapping.caseSensitive ? 'g' : 'gi');
 
       if (regex.test(text)) {
         found.push({
@@ -105,8 +105,8 @@ const SubstitutionEngine = {
     for (const mapping of sorted) {
       if (!mapping.enabled || !mapping.real?.trim() || !mapping.substitute?.trim()) continue;
 
-      const pattern = this._wordBoundaryPattern(mapping.real);
-      const regex = new RegExp(pattern, mapping.caseSensitive ? 'g' : 'gi');
+      const escaped = this._escapeRegex(mapping.real);
+      const regex = new RegExp(escaped, mapping.caseSensitive ? 'g' : 'gi');
       let match;
 
       while ((match = regex.exec(original)) !== null) {
@@ -145,16 +145,6 @@ const SubstitutionEngine = {
   _escapeRegex(str) {
     return str.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
   },
-
-  // Wrap an escaped literal in \b only on edges that are word characters,
-  // so "not" → "bad" matches "not" but not "nothing", while mappings whose
-  // edges aren't word chars (e.g. "@foo", "foo.com ") still work.
-  _wordBoundaryPattern(str) {
-    const escaped = this._escapeRegex(str);
-    const left = /^\w/.test(str) ? '\\b' : '';
-    const right = /\w$/.test(str) ? '\\b' : '';
-    return left + escaped + right;
-  },
 };
 
 // Support both module and content-script contexts

test-suite.js

@@ -345,18 +345,6 @@ test('Disabled mapping is skipped', () => {
   if (result.text.includes('Alex Demo')) throw 'Disabled mapping should not substitute';
 });
 
-test('Mapping matches whole words only', () => {
-  const result = SubstitutionEngine.substitute('nothing is not a thing, not even this', [
-    { real: 'not', substitute: 'bad', enabled: true }
-  ]);
-  if (result.text.includes('bahing') || result.text.includes('badhing')) {
-    throw `Should not match inside "nothing", got: ${result.text}`;
-  }
-  if (!/\bbad\b/.test(result.text)) {
-    throw `Should still match standalone "not", got: ${result.text}`;
-  }
-});
-
 // ============================================================
 // SMART PATTERNS
 // ============================================================