Security Paper Translate Skill

A Codex skill for translating cybersecurity and related academic paper PDFs into professional Chinese while preserving the original PDF layout. It is designed for papers where formatting matters: two-column layouts, figures, tables, captions, code listings, bold headings, colored citations, and terminology tables.

What it produces

For an input paper paper.pdf, the skill writes final outputs under translated-pdfs/paper/:

• paper-zh.pdf — Chinese-only content pages plus terminology pages.
• paper-zh-en-interleaved.pdf — Chinese page first, then matching English page, repeated for each source page, plus terminology pages.
• paper-terms.tsv — source-backed terminology and no-translate policy table.
• paper-generation-stats.json and paper-generation-stats.md — validation and generation summary.

Core guarantees

• Codex performs the translation itself; do not call external MT, other LLM APIs, or local LLMs.
• Preserve page size, columns, visual density, figures, tables, captions, bold/italic/color styling, and title hierarchy.
• Preserve code/listing/verbatim regions exactly: no translation, rewrapping, reindentation, line-break changes, or monospace-style loss.
• Use real font-level CJK bold/semi-bold for headings and emphasis. Do not fake bold with duplicate overlays, stroke text, shadows, or repeated offset drawing.
• Use local pdf2zh / PDFMathTranslate only as a layout engine when available, with its translator layer replaced by Codex-written local translations.
• Append a sourced terminology table and validate rendered sample pages before claiming completion.

Repository layout

security-paper-translate/
├── SKILL.md                         # Main skill instructions
├── agents/openai.yaml               # Agent UI metadata
├── references/
│   ├── font-bold-rendering.md       # Real-font bold rendering and validation rules
│   ├── layout-preservation-notes.md # Layout, style, code-block, and validation rules
│   ├── pdf2zh-layout-integration.md # Local pdf2zh monkeypatch workflow
│   ├── terminology-schema.md        # Terms TSV schema
│   └── translation-standards.md     # Academic/security translation standards
├── scripts/
│   ├── check_environment.py         # Dependency and local extension checks
│   ├── check_pdf_bold_integrity.py  # Detect duplicate/overlapped/stroked bold artifacts
│   ├── paper_pdf_tools.py           # Inspect, terms, interleave, validate, finalize helpers
│   ├── pdf2zh_capture_chunks.py     # Capture pdf2zh chunks without MT
│   ├── pdf2zh_codex_driver.py       # Run pdf2zh with local Codex translations
│   └── pdf_postprocess_template.py  # Generic postprocess scaffold
└── .github/workflows/validate.yml   # Basic repository validation

Install for Codex

Clone or copy this folder into your Codex skills directory:

git clone <repo-url> ~/.codex/skills/security-paper-translate

Then ask Codex to use $security-paper-translate on an English academic paper PDF.

Runtime dependencies

The helper scripts are intentionally lightweight. Minimum useful environment:

python -m pip install -r requirements.txt

Recommended system tools:

• pdftoppm from poppler, for render-based validation.
• pdffonts from poppler, for checking embedded fonts.
• pdf2zh, optional but recommended for strict layout preservation.
• A CJK font such as Source Han Serif CN or Noto CJK; a real CJK Bold/SemiBold face is strongly recommended.

Check the local environment with:

python scripts/check_environment.py --json

Typical agent prompt

Use $security-paper-translate to translate /path/to/paper.pdf. Follow the preflight confirmation, terminology policy, layout-preservation, code-block preservation, real-font bold rendering, validation, and finalize workflow in SKILL.md. Generate Chinese-only and Chinese-first interleaved PDFs under translated-pdfs/<stem>/.

Development validation

Run these checks before committing:

python -m py_compile scripts/*.py
python scripts/check_environment.py --json >/tmp/security-paper-translate-env.json
python /path/to/skill-creator/scripts/quick_validate.py .

When bold labels were patched in a generated PDF, also run:

python scripts/check_pdf_bold_integrity.py /path/to/paper-zh.pdf --fail-on-issue