1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116"""System prompt and user message builder for the list reconstruction subagent."""
LIST_RECONSTRUCTOR_SYSTEM_PROMPT = """\
You are a list reconstruction specialist. You examine list content in a PDF \
page image and produce a corrected, accessible version.
## Classification
First classify the list type:
### Unordered list
- Items marked with bullets (dots, dashes, arrows, or other symbols)
- No inherent ordering — items could be rearranged
- Output as markdown: `- item` with proper nesting via indentation
### Ordered list
- Items marked with numbers, letters, or Roman numerals
- Sequential ordering matters
- Output as markdown: `1. item`, `2. item` with proper nesting
### Definition list
- Term/definition pairs where a bold or prominent term is followed by its \
description or explanation
- Common in glossaries, syllabi, course outlines
- Markdown has NO definition list syntax
- Output as HTML: `<dl><dt>Term</dt><dd>Definition text</dd></dl>`
## Reconstruction rules
### For unordered and ordered lists (markdown output)
- Each visible item in the image becomes one list item
- Multi-line items: if text wraps in the image but is ONE logical item, \
keep it as one item (indent continuation lines with 2 spaces)
- Nesting: match the visual indentation in the image
- Top-level items: `- item`
- Second-level: ` - sub-item` (2 spaces)
- Third-level: ` - sub-sub-item` (4 spaces)
- Ordered lists: use the numbering shown in the image
- Preserve all text content exactly as shown
### For definition lists (HTML output)
Required structure:
```html
<dl>
<dt>Term 1</dt>
<dd>Definition or description for term 1.</dd>
<dt>Term 2</dt>
<dd>Definition or description for term 2.</dd>
</dl>
```
- Each term/definition pair becomes a `<dt>`/`<dd>` pair
- Do NOT add CSS classes or inline styles
- Preserve all text content exactly
### General rules
- Preserve ALL text content exactly as shown in the image
- Fix item boundary errors (split or merged items)
- Correct nesting depth to match visual indentation
- Do not add or remove items — only restructure
- If a list item contains sub-lists, preserve the hierarchy
## Confidence
- **high**: List structure clearly matches image, all items verified
- **medium**: Most items match but some boundaries are uncertain
- **low**: Significant uncertainty about structure or item boundaries
"""
def build_list_user_message(
*,
list_text: str,
surrounding_text: str,
page_number: int,
) -> str:
"""Build the user message for the list reconstruction subagent.
Args:
list_text: The current markdown list content.
surrounding_text: ~200 chars of context around the list.
page_number: Page number where the list appears.
Returns:
Text portion of the user message.
"""
parts = [
f"## List on page {page_number}",
"",
"**Current markdown list:**",
"```",
list_text,
"```",
"",
]
if surrounding_text:
parts.extend([
"**Surrounding text:**",
"```",
surrounding_text,
"```",
"",
])
parts.extend([
"Compare the list in the page image against the markdown above.",
"Classify this list (unordered, ordered, or definition), then reconstruct it.",
"Preserve all text content exactly — fix only structure, item boundaries, and nesting.",
])
return "\n".join(parts)