feat: export core Hermes skills

2026-07-15 02:45:56 +00:00
parent a028b63eda
commit 54711fee2a
308 changed files with 41310 additions and 1 deletions
@@ -0,0 +1,144 @@
+# Layering WB Revisions on High-Density Tracked Changes Documents
+
+## Problem
+When a document already has extensive tracked changes from another author (e.g., 华诚-Z with 170+ INS and 90+ DEL), ContractEditor's `tracked_replace` frequently fails with `ValueError: Element is not a child of this node` because the paragraph structure is heavily fragmented with interleaved `w:ins`/`w:del`/`w:r` elements.
+
+## Solution: Direct lxml Operations
+
+### Strategy
+Use zipfile + lxml to directly manipulate the XML instead of ContractEditor library. Three operation types:
+
+### 1. Append text to existing paragraph end
+Find the paragraph, locate the last content element, and append a `w:ins` after it.
+
+```python
+# Find the last non-pPr child element in the paragraph
+last_content = None
+for child in p:
+    if child.tag != f'{WNS}pPr':
+        last_content = child
+
+# Create INS element
+ins = etree.SubElement(p, f'{WNS}ins')
+ins.set(f'{WNS}id', str(next_id))
+ins.set(f'{WNS}author', 'WB')
+ins.set(f'{WNS}date', '2026-07-02T00:00:00Z')
+
+r = etree.SubElement(ins, f'{WNS}r')
+# Clone rPr from nearby run
+rpr = get_reference_rpr(p)  # see below
+if rpr is not None:
+    r.insert(0, copy.deepcopy(rpr))
+
+t = etree.SubElement(r, f'{WNS}t')
+t.set('{http://www.w3.org/XML/1998/namespace}space', 'preserve')
+t.text = "追加的文字内容"
+```
+
+### 2. Insert new paragraph (全段INS)
+Clone neighboring paragraph's pPr, create a new `w:p` with all content inside `w:ins`.
+
+```python
+# Clone pPr from reference paragraph
+ref_p = paras[target_idx]  # the paragraph after which to insert
+new_p = etree.Element(f'{WNS}p')
+
+# Clone pPr
+ref_ppr = ref_p.find(f'{WNS}pPr')
+if ref_ppr is not None:
+    new_p.append(copy.deepcopy(ref_ppr))
+
+# Create INS wrapping all content
+ins = etree.SubElement(new_p, f'{WNS}ins')
+ins.set(f'{WNS}id', str(next_id))
+ins.set(f'{WNS}author', 'WB')
+ins.set(f'{WNS}date', '2026-07-02T00:00:00Z')
+
+r = etree.SubElement(ins, f'{WNS}r')
+rpr = get_reference_rpr(ref_p)
+if rpr is not None:
+    r.insert(0, copy.deepcopy(rpr))
+t = etree.SubElement(r, f'{WNS}t')
+t.set('{http://www.w3.org/XML/1998/namespace}space', 'preserve')
+t.text = "新增条款全文"
+
+# Insert after reference paragraph
+ref_p.addnext(new_p)
+```
+
+### 3. Character-level replacement within high-density paragraph
+When text to replace is inside an existing `w:ins` from another author (e.g., 华诚-Z), you need to split that ins element.
+
+```python
+# Find the ins element containing target text
+for ins_elem in p.findall(f'{WNS}ins'):
+    for r in ins_elem.findall(f'{WNS}r'):
+        t = r.find(f'{WNS}t')
+        if t is not None and t.text and old_text in t.text:
+            # Split: keep text before, add WB del+ins for changed part, keep text after
+            pos = t.text.index(old_text)
+            before = t.text[:pos]
+            after = t.text[pos + len(old_text):]
+            
+            # Modify existing t to keep only 'before'
+            t.text = before + after.replace(old_text, new_text)  # simplified
+            # Or split into multiple elements...
+```
+
+### Getting reference rPr
+```python
+def get_reference_rpr(p):
+    """Get rPr from first non-del run in paragraph, or from 华诚-Z ins"""
+    # Try plain runs first
+    for r in p.findall(f'{WNS}r'):
+        rpr = r.find(f'{WNS}rPr')
+        if rpr is not None:
+            return rpr
+    # Try non-WB ins elements
+    for ins in p.findall(f'{WNS}ins'):
+        if ins.get(f'{WNS}author') != 'WB':
+            for r in ins.findall(f'{WNS}r'):
+                rpr = r.find(f'{WNS}rPr')
+                if rpr is not None:
+                    return rpr
+    # Try previous paragraph
+    prev = p.getprevious()
+    if prev is not None:
+        return get_reference_rpr(prev)
+    return None
+```
+
+## Critical: Post-save sz fix
+
+When INS runs clone rPr from paragraphs that lack explicit `w:sz` (relying on style inheritance), the INS will render at wrong size. **Always run a post-save sweep:**
+
+```python
+# Determine dominant body sz from neighboring paragraphs
+# Then fix all WB INS runs missing sz
+for ins in body.iter(f'{WNS}ins'):
+    if ins.get(f'{WNS}author') != 'WB':
+        continue
+    for r in ins.findall(f'{WNS}r'):
+        rpr = r.find(f'{WNS}rPr')
+        if rpr is not None:
+            sz = rpr.find(f'{WNS}sz')
+            if sz is None:
+                sz = etree.SubElement(rpr, f'{WNS}sz')
+                sz.set(f'{WNS}val', dominant_sz)  # e.g., '24' for 12pt
+                szCs = etree.SubElement(rpr, f'{WNS}szCs')
+                szCs.set(f'{WNS}val', dominant_sz)
+```
+
+## Author Unification
+
+After Doro reviews and confirms, unify all authors to WB:
+```bash
+python scripts/unify-author-wb.py input.docx [output.docx]
+```
+
+## Lesson Learned (2026-07-02)
+- Doro will edit the files in OnlyOffice after upload. Always download Doro's version before doing further work.
+- "你自己要满意再给我" = self-verify before delivery, don't ask user to check.
+- "认真做" = thoroughness signal. Read full contract text, verify each modification landed correctly.
+- When Doro says "看看是否还有需要调整的" = compare your version vs Doro's, identify what Doro changed, assess if further work needed.
+- Unifying author is a standard final step — use the script, don't hand-code each time.