nachafrommars/atydam-epub-v2.0/atydam-epub-v2-0

Name: atydam-epub-v2-0
Author: nachafrommars

Claude Code and agent skill by nachafrommars/atydam-epub-v2.0

Trust tier 3

community

Source check pending

Checking that the SKILL.md file is still in the repository.

Automated PDF→EPUB conversion pipeline v2.0 (base version). Smart 9-phase pipeline + closed-loop audit to produce premium EPUB from any PDF book (especially Vietnamese + Pali Buddhist texts with broken font encoding). NO GPT-STYLE bold emphasis. For bold version, use atyDam-epub v3.0. Aliases: pdf-epub-v2, pdf2epub-v2, pdf-to-epub-v2, atyDam-epub.

Install

npx vskill@latest install nachafrommars/atydam-epub-v2.0/atydam-epub-v2-0

SKILL.md

atyDam-epub v2.0 — Automated PDF→EPUB Pipeline (Base Version)

When to use this skill

When user wants to convert a PDF book to premium EPUB format without bold emphasis:

PDF with broken font encoding (Vietnamese + Pali Buddhist texts)
Books with complex tables, figures, footnotes
Books needing professional EPUB output (cover, hierarchy, italic markup)
No GPT-style bold — plain clean text

For GPT-style bold emphasis version, use atyDam-epub v3.0 instead.

Pipeline Overview — 9 Phases + Audit Loop

PDF input
  │
  ├─► Phase A: EXTRACT (PyMuPDF)
  │     • Extract images, render cover, detect chapters/TOC/blank pages
  │     • Output: extract_meta.json
  │
  ├─► Phase 1: TEXT RECOVERY (Gemini Flash + REST API)
  │     • Split PDF to 4500-char chunks  
  │     • Fix Vietnamese diacritics + Pali subset font damage
  │     • CRITICAL: thinkingBudget=0 to avoid truncation
  │     • Output: recovered/chunk_NNNN.txt
  │
  ├─► Phase B2: PARAGRAPH RECONSTRUCT (regex)
  │     • Join physical lines into real paragraphs
  │     • Detect headings, lists, captions
  │     • Fix orphan letters (page break artifacts)
  │
  ├─► Phase B3: SMART LIST SPLIT (regex)
  │     • Split inline "1)... 2)... 3)..." patterns
  │     • Mark tables with TABLE_START/END for B9
  │
  ├─► Phase B6: AI STRUCTURE DETECTION (Gemini)
  │     • Hierarchy: CHƯƠNG > Roman > Letter > Number
  │     • Heading + body merge logic
  │     • Inline list to proper markdown
  │
  ├─► Phase B7: PERFECT STRUCTURE (Gemini)
  │     • Semantic paragraph boundaries (8 RULES vàng)
  │     • Merge broken lines, split on topic shift
  │
  ├─► Phase B9: TABLE VISION (Gemini Vision)
  │     • Render PDF page → image
  │     • Gemini Vision reads table layout from image
  │     • Output: HTML <table> with rowspan/colspan
  │
  ├─► Phase B5: PROOFREAD (Gemini)
  │     • Spelling fixes + Phật học terms
  │     • Dictionary + AI suggestions
  │
  ├─► Phase C-D: MARKUP + BUILD (Python + Pandoc)
  │     • Add YAML front matter
  │     • Pali italic wrapping
  │     • Image embeds + caption
  │     • Cover image, premium CSS
  │     • EPUB 3.2 structure
  │
  └─► AUDIT LOOP (closed-loop, max 5 iterations)
        • Layer 1: Structural (inline headings, paragraph)
        • Layer 2: Semantic (AI - orphan fragments)
        • Layer 3: Spelling (dict + AI suggestions)
        • Layer 4: Tables (validation, empty cells)
        • Layer 5: Images (file verify)
        • Layer 6: Cross-ref (numbered gaps)
        • Layer 7: AI meta-review (deep read)
        • Loop until 0 issues OR max iterations

Step-by-step Execution

Step 1: Setup workspace

bash

WORKDIR="/root/.openclaw/workspace/<book-name>-epub"
mkdir -p "$WORKDIR/{chunks,recovered,images}"
cd "$WORKDIR"
cp scripts/* .  # Copy all phase scripts

Step 2: Run pipeline

bash

export GEMINI_API_KEY="..."
 
# Phase A: Extract
python3 phase_a_extract.py
 
# Phase 1: Recovery (paid tier required, ~$0.50 for 400-page book)
python3 pipeline_parallel.py  # 5 workers, thinkingBudget=0
 
# Phase B2-B3: Structure prep
python3 phase_b2_paragraphs.py
python3 phase_b3_smart_list.py
 
# Phase B6-B7: AI structure  
python3 phase_b6_smart_structure.py
python3 phase_b7_perfect_structure.py
 
# Phase B9: Vision tables
python3 phase_b9_tables_vision.py
 
# Phase B5: Proofread
python3 phase_b5_proofread.py
 
# Phase C-D: Build
python3 phase_c_markup.py
python3 phase_d_build.py
 
# Audit loop (closed)
python3 audit_loop.py
python3 audit_v2.py  # Phase A+B
 
# Final build
python3 phase_d_build.py  # Rebuild from cleaned text

Step 3: Validate

bash

java -jar /usr/share/java/epubcheck.jar TrietHoc-*.epub
# Expected: 0 fatals / 0 errors / 0 warnings

Key Insights from Production Use

Gemini Configuration (CRITICAL)

Model: gemini-2.5-flash (paid tier)
thinkingBudget: 0 ← ESSENTIAL or output truncates
temperature: 0.1 for factual fidelity
maxOutputTokens: 16384 safety net
REST API (not SDK 0.8.6) because old SDK doesn't support thinkingConfig

Cost per book (~420 pages)

Phase 1 Recovery: ~$0.54 (129 chunks)
Phase 2 Tables: ~$0.20 (Vision rebuild)
Phase 5 Proofread: ~$0.40
Phase 7 Structure: ~$0.42
Phase 8 Perfect: ~$0.42
Audit V1: ~$0.79
Audit V2: ~$0.38
Total: ~$3.15 (≈ 75,000 VNĐ)

Common Issues & Fixes

Output truncation → thinkingBudget: 0
Free tier 20 RPD limit → Need paid billing
Tables flat text → Vision API with PDF page render
Headers inline with text → Regex split at sentence boundary + \n\n
Pandoc auto-lists → Use markdown-fancy_lists-startnum format
Cell numbering becomes `<ol>` → Replace N. with (N) in cells
Caption duplicated in `<th>` → Strip first row of thead
Table click trắng xóa → Remove background-color, only borders
Reader theme override → Don't force color/background
Page break orphan letters → Merge with next paragraph

Quality Indicators (v2.0 target)

✅ Avg paragraph length 150-380 chars (proper semantic)
✅ 9 chapters proper hierarchy (not broken)
✅ All tables HTML with thead/tbody/th/td
✅ Italic Pali wrapping (~1,100+ terms)
✅ epubcheck: 0 errors / 0 warnings
✅ Images embedded (~30 figures)
✅ Cover from PDF page 1
✅ NO bold emphasis (plain clean text)

Difference: v2.0 vs v3.0

Feature	v2.0	v3.0
Text recovery	✅	✅
Structure detection	✅	✅
Tables rebuild	✅	✅
Proofreading	✅	✅
GPT-Style bold	❌ NO	✅ YES (~1,500 patterns)
Cost per book	~$3.15	~$3.53
Output size	1.4 MB	1.4 MB

Files Structure

<book>-epub/
├── chunks/              # PDF text chunks (Phase A)
├── recovered/           # Font-recovered chunks (Phase 1)
├── images/              # Extracted images (Phase A)
├── extract_meta.json    # All metadata
├── cover.png            # Rendered cover
├── style-v2-premium.css # Tiểu Tâm color palette
├── pipeline_parallel.py # Parallel recovery (5 workers)
├── phase_a_extract.py   # Smart PDF extraction
├── phase_b2_paragraphs.py  # Paragraph reconstruct
├── phase_b3_smart_list.py  # Inline list split
├── phase_b5_proofread.py   # AI proofreading
├── phase_b6_smart_structure.py  # AI hierarchy
├── phase_b7_perfect_structure.py # Semantic paragraphs
├── phase_b9_tables_vision.py    # Tables with Vision
├── phase_c_markup.py    # Markdown markup
├── phase_d_build.py     # Pandoc EPUB build
├── audit_loop.py        # 7-layer closed-loop audit
├── audit_v2.py          # Deep audit + auto-apply
├── audit-log.json       # Audit findings
└── TrietHoc-A-Ty-Dam-VN.epub  # Final output (NO BOLD)

Trigger Auto-run

When user provides a PDF path with intent to convert (standard aliases):

"convert PDF sang EPUB"
"pdf to epub"
"pdf2epub"
etc.

Em will run v2.0 by default. For GPT-style version, user must explicitly request v3.0.

Memory Note

atyDam-epub v2.0 is the base version without GPT-style bold. Produced from work done 16/05/2026 on Triết Học A-Tỳ-Đàm (Buddha Abhidhamma) — Dr. Mehm Tin Mon.

V10 (LANDMARK): v2.0 base pipeline
V11: v3.0 with GPT-style bold emphasis

For GPT-style version, see atyDam-epub v3.0 skill.

Security

Security scanPassed, 95 of 100Pattern checks on the skill's files.
Model reviewPassed

See the full security report

Quality evaluation

Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

See evaluation results

Source

SKILL.md in the repositoryCheckingSKILL.md

Popularity

7-day trend score: 5

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

Claude Code
Cursor
GitHub Copilot
Windsurf
Codex
Gemini CLI
Cline
Amp
Roo Code
Goose
Continue
Kilo Code
Junie
Trae
OpenHands
Qwen Code
AI Agent
Tabnine Chat
Sourcegraph Cody
Amazon Q
Firebase Genkit
Sweep
Devin
Phind
Mentat
Claude Desktop
Aider
GPT Pilot
Plandex
bolt.new
v0
Replit Agent
Anse
TypingMind
BetterChatGPT
LibreChat
Warp AI
Pieces
Fig AI

Embed badge

[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/nachafrommars/atydam-epub-v2.0/atydam-epub-v2-0/badge)](https://verified-skill.com/skills/nachafrommars/atydam-epub-v2.0/atydam-epub-v2-0)

Something wrong with this skill or its page? Tell us.