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 +1,60 @@
+# Split-Run Numbering in docx XML
+
+## Problem
+Contract numbering like `（5）` is often split across multiple `<w:r>` runs in the XML:
+```xml
+<w:r><w:t>（</w:t></w:r>
+<w:r><w:t>5</w:t></w:r>
+<w:r><w:t>）委托方</w:t></w:r>
+```
+
+A naive `tracked_replace("（5）", "（6）")` searching for the complete string in a single `<w:t>` will **silently fail** — no match, no error, no renumbering.
+
+## Solution: Multi-run concatenation + split
+
+### Algorithm
+```python
+def tracked_replace_split_number(p, old_num, new_num):
+    """Handle （old_num） spread across multiple runs."""
+    target = f'（{old_num}）'
+    new_target = f'（{new_num}）'
+    
+    # 1. Collect all plain runs (not inside w:ins or w:del)
+    plain_runs = [(index, run, text) for each child of p]
+    
+    # 2. Slide a window: concatenate adjacent run texts until target is found
+    for start in range(len(plain_runs)):
+        concat = ""
+        for end in range(start, start+4):  # max 4 runs for a number
+            concat += plain_runs[end].text
+            if target in concat:
+                # Found! Extract before/after text around the number
+                runs_to_wrap = plain_runs[start:end+1]
+                # ...proceed to replace
+    
+    # 3. Remove original runs, insert:
+    #    - [before_run if text before number]
+    #    - DEL element with delText=target
+    #    - INS element with t=new_target  
+    #    - [after_run if text after number, e.g. "委托方"]
+    
+    # 4. Set rsid attributes: rsidDel on DEL runs, rsidR on INS runs
+```
+
+### Critical: Process order
+**Always renumber from bottom to top** (last paragraph first) to avoid index shifting:
+```python
+# CORRECT
+renumber = [(P113, '10', '11'), (P112, '9', '10'), (P111, '7', '8')]
+
+# WRONG - P112 was already renumbered when we get to it
+renumber = [(P111, '7', '8'), (P112, '9', '10'), (P113, '10', '11')]
+```
+
+### Edge cases encountered (2026-06-08)
+- `（` + `10）` (two runs, not three) — the closing `）` merged with the digit
+- `（` + `5` + `）委托方` — closing `）` merged with following text, must split run to preserve "委托方"
+- Copy `w:rPr` from original runs to all new DEL/INS runs to preserve font/size
+
+## Lesson
+This was the root cause of a terminal review failure where 3 new clauses were inserted without numbering, and subsequent numbering was not renumbered. The `tracked_replace` function matched nothing because it expected `（5）` as a single text node.