pdf-translation

name: pdf-translation description: Use when processing PDF rulebooks, extracting raw markdown and image artifacts, or preparing source materials before chapter-split, init-doc, or translation setup. user-invocable: true disable-model-invocation: true

PDF Translation Workflow

Overview

Convert a source PDF into extracted markdown and image artifacts ready for chapter planning and translation setup.

Core principle: Extract cleanly and verify raw artifacts before chapter planning starts.

The Process

Step 1: Extract PDF

Run:

uv run python scripts/extract_pdf.py data/pdfs/<filename>.pdf

Expected outputs in data/markdown/:

• <name>.md
• <name>_pages.md
• images/<name>/

Step 2: Validate Raw Outputs

Validate:

• expected files exist
• page markers are present in _pages.md
• image manifest exists when image extraction is enabled
• extraction text is readable enough for downstream planning

Step 3: Re-extract or Fix Source if Needed

If extraction is garbled or page markers are broken:

• adjust source PDF
• re-run extraction
• do not continue until raw outputs are usable

Step 4: Handoff

Hand off extracted outputs to:

• chapter-split for chapter/file planning and navigation generation
• init-doc when project-level decisions and terminology bootstrap are still pending

Progress Sync Contract (Required)

1. Track extraction and raw validation steps in TodoWrite.
2. Mark extraction complete only after raw output validation.

When to Stop and Ask for Help

Stop when:

• extraction output is unreadable/garbled
• page markers are missing or unreliable
• repeated extraction attempts still fail

When to Revisit Earlier Steps

Return to Step 1 when:

• source PDF changes
• extraction options change

Red Flags

Never:

• hand-build chapter mapping here when chapter-split is the correct next skill
• skip raw output validation before handoff

Next Step

Continue with chapter-split or /init-doc.