hebbian 0.8.2 → 0.9.0

package/dist/bin/hebbian.js

@@ -2373,12 +2373,15 @@ function extractCorrections(messages) {
   const corrections = [];
   for (const text of messages) {
     if (corrections.length >= MAX_CORRECTIONS_PER_SESSION) break;
+    const trimmed = text.trim();
     if (text.length < MIN_CORRECTION_LENGTH) continue;
-    if (/^[\/!]/.test(text.trim())) continue;
-    if (text.trim().endsWith("?")) continue;
-    if (/^<[a-zA-Z]/.test(text.trim())) continue;
-    if (/^Base directory for this skill:/i.test(text.trim())) continue;
-    if (/^[•·▸▶\-\*]\s/.test(text.trim())) continue;
+    if (/^[\/!]/.test(trimmed)) continue;
+    if (/[?？]\s*[!！]?\s*$/.test(trimmed)) continue;
+    if (/^<[a-zA-Z]/.test(trimmed)) continue;
+    if (/^Base directory for this skill:/i.test(trimmed)) continue;
+    if (/^[•·▸▶\-\*]\s/.test(trimmed)) continue;
+    if (/<[a-zA-Z][a-zA-Z-]*>/.test(trimmed) && /<\/[a-zA-Z]/.test(trimmed)) continue;
+    if (isNarrativeKorean(trimmed)) continue;
     const correction = detectCorrection(text);
     if (correction) {
       corrections.push(correction);
@@ -2386,12 +2389,43 @@ function extractCorrections(messages) {
   }
   return corrections;
 }
+function isNarrativeKorean(text) {
+  const NARRATIVE_MARKERS = [
+    /이유는/,
+    // "the reason is..."
+    /예를\s*들면/,
+    // "for example..."
+    /그래서/,
+    // "so/therefore..."
+    /그런데/,
+    // "but then..."
+    /왜냐하면/,
+    // "because..."
+    /거든/,
+    // "...you see" (explanatory)
+    /있었[던거어]/,
+    // "there was..." (past narrative)
+    /중인데/,
+    // "...in the middle of" (ongoing situation)
+    /거 같은데/,
+    // "it seems like..." (speculation)
+    /어떻게\s*생각/
+    // "what do you think?" (asking opinion)
+  ];
+  const markerCount = NARRATIVE_MARKERS.filter((p) => p.test(text)).length;
+  return markerCount >= 2;
+}
 function detectCorrection(text) {
   const isNegation = NEGATION_PATTERNS.some((p) => p.test(text));
   const isMust = MUST_PATTERNS.some((p) => p.test(text));
   const isWarn = WARN_PATTERNS.some((p) => p.test(text));
   const isAffirmation = AFFIRMATION_PATTERNS.some((p) => p.test(text));
   if (!isNegation && !isMust && !isWarn && !isAffirmation) return null;
+  const categories = [isNegation, isMust, isWarn, isAffirmation].filter(Boolean).length;
+  const koreanRatio = (text.match(/[\uAC00-\uD7AF]/g) || []).length / Math.max(text.length, 1);
+  if (koreanRatio > 0.3 && categories < 2) {
+    if (text.length > 100) return null;
+  }
   let prefix;
   if (isNegation) prefix = "NO";
   else if (isMust) prefix = "MUST";
@@ -2590,24 +2624,21 @@ var init_digest = __esm({
       /\bnever\b/i,
       /\binstead\b/i,
       /^no[,.\s!]/i,
-      /\bdon[''\u2019]?t\s+use\b/i,
       /\bavoid\b/i,
-      // Korean negation — specific verb forms only to avoid matching incidental 않/대신 in explanations
-      /하지\s*마/,
-      /안\s*돼/,
-      /쓰지\s*마/,
-      /[가-힣]+지\s*마/,
-      /하지\s*않/,
-      // "do not" specifically
-      /쓰지\s*않/
-      // "do not use" specifically
+      // Korean negation — require AI-directed imperative context:
+      // "X하지 마" (don't X) — must have a verb object before 지 마
+      /[을를은는도이가]\s*[가-힣]+지\s*마/,
+      // "X 하면 안 돼" (must not X) — conditional + prohibition
+      /하면\s*안\s*돼/,
+      // "X 쓰지 마" (don't use X) — explicit "don't use"
+      /쓰지\s*마/
     ];
     AFFIRMATION_PATTERNS = [
-      /\balways\b/i,
       /\bshould\s+always\b/i,
       /\buse\s+\w+\s+instead\b/i,
-      // Korean affirmation
-      /항상/
+      // Korean affirmation — require directive context
+      /항상\s*[가-힣]+[해하]/
+      // "항상 X해" (always do X)
     ];
     MUST_PATTERNS = [
       /\bmust\b/i,