jlpt-study-pipeline

name: jlpt-study-pipeline description: Generate JLPT study packages with Obsidian Markdown, Anki cards, edge-tts audio assets, and immersive video outputs from a user-provided Japanese study topic.

JLPT Study Pipeline

Use this skill when the user asks to generate JLPT learning material, Japanese vocabulary packs, Obsidian notes, Anki cards, or immersive study videos.

Required Inputs

If any required input is missing, ask for it before generating files:

• Topic, such as JLPT N1/N2 擬聲詞 100 個.
• Output directory.
• Target JLPT level or levels.
• Item count.

Data Generation Rules

Generate source.json before running the pipeline. The JSON must contain metadata and entries.

Each entry must include:

• id
• term
• kana
• jlpt_level_estimate
• category
• zh_tw_meaning
• example_ja — complete Japanese sentence using the vocabulary
• example_ja_phrase (optional) — short phrase or collocation containing the vocabulary word, without forming a full sentence. Include this when a concise phrase is more useful for quick recall than a full sentence.
• example_zh_tw
• recall_prompt_zh_tw
• verification_status

Use Traditional Chinese for explanations. Use natural Japanese for examples. If the content is AI-generated and not checked against user-provided source material, set verification_status to needs_review and treat the JLPT level as an estimate.

Do not claim JLPT level authority unless the user provides verified source material. Include exam_risk_note_zh_tw when a level or usage point should be checked.

Output Workflow

After source.json exists, run:

python scripts/jlpt_pipeline.py build \
  --source path/to/source.json \
  --out path/to/output \
  --deck-name "JLPT Study" \
  --tts-provider edge \
  --voice ja-JP-NanamiNeural \
  --example-style sentence \
  --word-repetition 2 \
  --video

Use --example-style phrase to use the short example_ja_phrase field instead of the full sentence for TTS audio, video subtitles, and Anki card backs. If an entry has no example_ja_phrase, it falls back to example_ja automatically. Use --word-repetition <count> to configure how many times the Japanese vocabulary word is read out (default: 2). Use --video-words-per-short <count> with --video to create segmented YouTube short-style videos under shorts/short_*/video.mp4; omit it to keep the original single long video.mp4. Always ask the user which style, word repetition count, and video segmentation mode they prefer before running the pipeline.

The pipeline exports Obsidian Markdown, Anki .apkg, anki.csv, narration.txt, subtitles.ass, optional edge-tts audio, optional long video.mp4 or segmented shorts/short_*/video.mp4, and validation_report.md.

If edge-tts is unavailable or audio is not needed, use --tts-provider none so the user still gets Markdown, Anki, subtitles, narration, and silent video assets.

edge-tts

edge-tts is the default TTS provider. It does not require an API key, but it does require the edge-tts command from the Python package and network access to the Microsoft Edge TTS service.

Use dry-run before large generations to estimate character usage:

python scripts/jlpt_pipeline.py dry-run --source path/to/source.json --out path/to/output

List available voices with:

edge-tts --list-voices

Reporting

At the end, report generated file paths and summarize validation warnings. Treat needs_review as a normal review state, not as an error.