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 +1,118 @@
+# Mixed Inherited/Explicit Font Size Fix (Document-Wide)
+
+## Problem (2026-07-01 生育友好宣传阵地建设协议)
+
+Source document has **mixed font sizing** in body text:
+- Some runs have explicit `sz=24` (12pt) — e.g., section headings, specific clauses
+- Other runs have **no explicit sz** — inherit from Normal style (`sz=21` / 10.5pt)
+- WB INS runs mostly got `sz=24` correctly, but the mix of explicit + inherited in **original** runs creates visual inconsistency
+
+Doro complaint: "文字大小不一致，修改" — the rendered result shows mixed sizes.
+
+## Root Cause
+
+- `docDefaults` / Normal style = 10.5pt (sz=21)
+- Many body runs (P12+) have explicit sz=24 (from original author or conversion)
+- ~72 original runs have NO explicit sz → inherit 10.5pt → render smaller
+- OnlyOffice renders the mix faithfully → visible inconsistency
+
+## Diagnosis
+
+```python
+from docx import Document
+from collections import Counter
+
+doc = Document('file.docx')
+print(f'Normal style sz: {doc.styles["Normal"].font.size}')  # If 133350 EMU = 10.5pt
+
+sizes = Counter()
+for p in doc.paragraphs[BODY_START:BODY_END]:
+    for run in p.runs:
+        if run.text.strip():
+            sizes[run.font.size.pt if run.font.size else 'inherited'] += 1
+
+# If both 'inherited' and explicit size (e.g. 12.0) appear → mixed problem
+print(sizes.most_common())
+```
+
+## Fix Pattern (Full Body Range)
+
+Unlike the INS-only sweep, this fix targets ALL runs in the body text range:
+
+```python
+import zipfile, re
+from lxml import etree
+
+WNS = '{http://schemas.openxmlformats.org/wordprocessingml/2006/main}'
+
+# 1. Identify body range (skip title/preamble and signature)
+BODY_START = 12   # First body content paragraph index
+BODY_END = 56     # Last body paragraph (exclusive)
+TARGET_SZ = '24'  # From explicit runs in body (majority value)
+
+# 2. Fix ALL runs in body range
+for pidx in range(BODY_START, min(BODY_END, len(paras))):
+    p = paras[pidx]
+    
+    # Plain runs
+    for r in p.findall(f'{WNS}r'):
+        t_elem = r.find(f'{WNS}t')
+        if t_elem is None or not (t_elem.text or '').strip():
+            continue
+        rpr = r.find(f'{WNS}rPr')
+        if rpr is None:
+            rpr = etree.SubElement(r, f'{WNS}rPr')
+            r.insert(0, rpr)
+        sz = rpr.find(f'{WNS}sz')
+        if sz is None:
+            sz = etree.SubElement(rpr, f'{WNS}sz')
+        sz.set(f'{WNS}val', TARGET_SZ)
+        szCs = rpr.find(f'{WNS}szCs')
+        if szCs is None:
+            szCs = etree.SubElement(rpr, f'{WNS}szCs')
+        szCs.set(f'{WNS}val', TARGET_SZ)
+    
+    # INS runs
+    for ins in p.findall(f'{WNS}ins'):
+        for r in ins.findall(f'{WNS}r'):
+            # same logic as above
+            ...
+    
+    # DEL runs (for visual consistency in markup view)
+    for d in p.findall(f'{WNS}del'):
+        for r in d.findall(f'{WNS}r'):
+            # same logic
+            ...
+```
+
+## Key Distinctions from INS-Only Fix
+
+| Aspect | INS-only sweep | Full body range fix |
+|--------|---------------|---------------------|
+| Scope | Only WB INS runs | ALL runs (plain + INS + DEL) |
+| Trigger | INS runs missing sz | Doro reports "文字大小不一致" |
+| Root cause | add_clause/tracked_replace gaps | Source document mixed inheritance |
+| Target sz | From neighboring runs | From majority explicit sz in body |
+
+## When to Apply
+
+- Doro says "文字大小不一致" on a delivered file
+- `wb-ins-font-verify.py` passes (INS runs OK) but rendered output still shows mixed sizes
+- Diagnostic shows body runs split between `inherited` and explicit sz
+
+## Important: Don't Change Preamble/Signature
+
+- Title/header (e.g., P0-P2): larger sz by design (22pt/sz=44) — don't touch
+- Party info (P3-P10): may use different sz — don't touch unless in body range
+- Signature area (P56+): often sz=21 (10.5pt) — don't touch
+- Only fix the **body text range** where sz should be uniform
+
+## Relationship to 格式保留铁律
+
+This fix does NOT violate "格式保留铁律" (don't change original formatting) because:
+- The original document's **intent** is uniform 12pt body text (evidenced by majority explicit sz=24)
+- The missing sz is a **formatting omission** (author forgot to set explicit sz on some runs)
+- The fix makes the document render as the original author intended
+- This is different from "changing 仿宋_GB2312 to 仿宋" (that changes the actual format choice)
+
+BUT: if the original document intentionally uses different sizes in body (e.g., smaller text for notes, larger for headings), don't blindly unify. Check the pattern first.