@heylemon/lemonade 0.1.9 → 0.1.10

package/dist/build-info.json

@@ -1,5 +1,5 @@
 {
-  "version": "0.1.9",
-  "commit": "148a3ecd94824b7bed57702ff16e416442b7d831",
-  "builtAt": "2026-02-21T19:41:57.574Z"
+  "version": "0.1.10",
+  "commit": "091a5fb31c13a8706120e95f8cc1161d21b1d842",
+  "builtAt": "2026-02-21T20:08:01.424Z"
 }

package/dist/canvas-host/a2ui/.bundle.hash

@@ -1 +1 @@
-b3ce500e7c71ed0a3415b4d2d8e472926cb0135cdb8a2f66ab05abb127c334ac
+ae5dd78f56173dbfc66b8b6f4a72fb917e3108dcb10745fd17c25536ee4bd709

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@heylemon/lemonade",
-  "version": "0.1.9",
+  "version": "0.1.10",
   "description": "AI gateway CLI for Lemon - local AI assistant with integrations",
   "publishConfig": {
     "access": "restricted"

package/skills/docx/SKILL.md

@@ -147,6 +147,27 @@ new Paragraph({
 ```
 
 Also note: use `italics: true` (with 's'), not `italic: true`. The latter silently does nothing in docx-js.
+
+### Prevent Duplicate Text Rendering (CRITICAL)
+
+**Use exactly one text source per paragraph.** In docx-js, a paragraph can get text from:
+- `text: "..."` on `new Paragraph(...)`, or
+- `children: [new TextRun({ text: "..." })]`
+
+If you set both, the same title/subtitle/date can render twice (common on cover pages).
+
+```javascript
+// ✅ Correct: one source (children only)
+new Paragraph({
+  spacing: { after: 160 },
+  children: [new TextRun({ text: "February 22, 2026", size: 28, color: COLORS.muted })]
+});
+
+// ❌ Wrong: duplicates text (text + children)
+new Paragraph({
+  text: "February 22, 2026",
+  children: [new TextRun({ text: "February 22, 2026", size: 28, color: COLORS.muted })]
+});
 ```
 
 ### Lists: Never Use Unicode Bullets
@@ -346,6 +367,7 @@ Or use a more sophisticated approach with field codes (requires XML manipulation
 - Use `HeadingLevel.HEADING_1`, etc., with `outlineLevel` for TOC
 - Use `children` with TextRun for headings, NOT the `run` property (Pages ignores `run`)
 - Use `italics: true` (with 's'), not `italic: true`
+- Use exactly one text source per paragraph: either `text` OR `children`, never both
 - Set `spacing.before` and `spacing.after` explicitly on headings
 
 **DON'T:**
@@ -359,6 +381,7 @@ Or use a more sophisticated approach with field codes (requires XML manipulation
 - Assume A4 page size (always set explicitly to US Letter)
 - Use `run` property on headings (use `children` with TextRun instead — Pages ignores `run`)
 - Use `italic: true` without the 's' (use `italics: true`)
+- Put the same literal content in both `text` and `children` on one Paragraph (duplicates output)
 - Mix font styles — stick to 2 fonts (heading + body)
 
 ---
@@ -450,6 +473,7 @@ The validator checks:
 - Paragraph length (warns if > 500 chars)
 - Empty headings
 - Manual bullet characters (should use List APIs)
+- Obvious cover/title duplication risks (`text` + `children` on same paragraph in generation code)
 
 ### 2. Fix Tables for Cross-Platform (MANDATORY if document contains tables)
 
@@ -495,6 +519,8 @@ pdftoppm -jpeg -r 150 report.pdf page
 # View: page-1.jpg, page-2.jpg, etc.
 ```
 
+During this visual pass, explicitly check the title block (title, subtitle, date) for ghosted/double text. If duplicated, inspect the corresponding `new Paragraph(...)` and remove either `text` or `children` so only one text source remains.
+
 ### 5. Fix & Re-validate
 
 If the validator reports issues or the PDF looks off:

package/skills/docx/references/templates.md

@@ -2,6 +2,17 @@
 
 Each template shows the structure and a docx-js code snippet for that document type.
 
+## Important Implementation Notes
+
+- Prefer the deterministic Python engine (`scripts/create_doc.py`) for production output when possible.
+- If using docx-js directly, use exactly one paragraph text source:
+  - `text: "..."` **or** `children: [new TextRun({ text: "..." })]`
+  - never both on the same `new Paragraph(...)`.
+- Run both post-processors after generation:
+  - `python scripts/fix_tables.py output.docx`
+  - `python scripts/fix_styles.py output.docx`
+- Run `python scripts/validate.py output.docx` before delivery to catch duplicate render patterns early.
+
 ## Report
 
 The most versatile template. Use for analytical, strategic, or informational documents.

package/skills/docx/scripts/validate.py

@@ -8,6 +8,7 @@ Checks for:
 - Paragraph length and structure
 - Empty headings
 - Manual bullet characters (should use List APIs)
+- Duplicate title/content rendering (same phrase repeated adjacently)
 - Table consistency
 - Page setup
 
@@ -18,6 +19,7 @@ import sys
 import os
 import zipfile
 import xml.etree.ElementTree as ET
+import re
 from pathlib import Path
 
 try:
@@ -90,10 +92,35 @@ def validate_document(docx_path):
 
     # Unicode bullet characters
     bullet_chars = {'•', '◦', '▪', '●', '○', '■', '–', '—', '-', '*'}
+    non_empty_paragraphs = []
+
+    def normalize_text(text):
+        """Normalize visible text for duplicate detection."""
+        # Case-insensitive, collapse whitespace, keep only alnum/space
+        cleaned = re.sub(r"[^a-zA-Z0-9\s]", " ", text.lower())
+        return " ".join(cleaned.split())
+
+    def find_adjacent_repeat_phrase(text, min_words=3, max_words=16):
+        """Detect adjacent repeated phrase in one paragraph.
+
+        Example: "The Great War ... The Great War ..."
+        """
+        words = normalize_text(text).split()
+        if len(words) < min_words * 2:
+            return None
+        upper = min(max_words, len(words) // 2)
+        for size in range(upper, min_words - 1, -1):
+            for i in range(0, len(words) - 2 * size + 1):
+                left = words[i:i + size]
+                right = words[i + size:i + 2 * size]
+                if left == right:
+                    return " ".join(left)
+        return None
 
     for p in doc.paragraphs:
         paragraph_count += 1
         text = p.text.strip()
+        norm_text = normalize_text(text)
 
         # Track empty paragraphs
         if not text:
@@ -102,6 +129,7 @@ def validate_document(docx_path):
             max_consecutive_empty = max(max_consecutive_empty, consecutive_empty)
         else:
             consecutive_empty = 0
+            non_empty_paragraphs.append((paragraph_count, text, norm_text))
 
         # Check headings
         if p.style.name.startswith("Heading"):
@@ -137,6 +165,14 @@ def validate_document(docx_path):
             word_count = len(text.split())
             warnings.append(f"Very long paragraph ({len(text)} chars, {word_count} words): \"{text[:50]}...\" — consider breaking up")
 
+        # Detect duplicated phrase inside the same paragraph
+        if len(text) >= 20:
+            repeated = find_adjacent_repeat_phrase(text)
+            if repeated:
+                issues.append(
+                    f"Possible duplicated text in paragraph {paragraph_count}: repeated phrase \"{repeated[:60]}\""
+                )
+
         # Check for excessive spacing/empty paragraphs
         if len(text) == 0 and paragraph_count > 1:
             # This is tracked separately
@@ -181,6 +217,15 @@ def validate_document(docx_path):
     if paragraph_count < 3:
         warnings.append("Document is very short — consider if this is complete")
 
+    # Detect duplicate consecutive non-empty paragraphs (common in title blocks)
+    for idx in range(len(non_empty_paragraphs) - 1):
+        curr_num, curr_raw, curr_norm = non_empty_paragraphs[idx]
+        next_num, next_raw, next_norm = non_empty_paragraphs[idx + 1]
+        if curr_norm and curr_norm == next_norm:
+            issues.append(
+                f"Consecutive duplicate paragraphs at {curr_num} and {next_num}: \"{curr_raw[:80]}\""
+            )
+
     # ─── Summary info ───
     info.append(f"Paragraphs: {paragraph_count}")
     info.append(f"Headings: {heading_count}")