Anki Add Sentence

Anki Add Sentence

Overview

This skill provides a procedural workflow for generating i+1 example sentences for Anki learning targets. It specifically targets notes where the target field is currently empty, ensuring no data is overwritten. It ensures that generated sentences only use words already learned by the user, plus the single new target word. It handles Anki furigana formatting and updates both the sentence and its English translation.

Parameters

When this skill is triggered, respect the following parameters:

• --deck: (Required) The Anki Deck Name to process.
• --source-field: (Optional) The field containing the target word. Defaults to Expression.
• --target-field: (Optional) The field where the generated sentence will be stored. Defaults to Sentence.
• --target-field-english: (Optional) The field where the English translation will be stored. Defaults to SentenceEnglish.
• --limit: (Optional) Maximum number of notes to process in one run. Defaults to 20.
• --batch-size: (Optional) The number of words to include in a single generation request. Defaults to 5.
• --dry-run: (Optional) Show a list of words that would be processed without generating sentences or updating Anki.

Workflow

To generate sentences for an Anki deck, follow these steps:

1. Identify Learned Vocabulary

Before generating any sentences, you must know what the user has already learned.

1. Run the scripts/extract_learned_vocab.py script.
2. Capture the stdout output as the list of learned words.

2. Fetch and Filter Notes (Strictly Empty Target)

Query AnkiConnect for notes in the specified --deck where the --target-field is unset (empty).

AnkiConnect Query Example:

curl -s -X POST http://localhost:8765 -d '{
  "action": "findNotes",
  "version": 6,
  "params": {
    "query": "deck:\"'"$DECK"'\" \"'"$TARGET_FIELD"':\""
  }
}'

1. Fetch note details using notesInfo.
2. Sort by Frequency: Sort the resulting notes by the Frequency field (ascending) to prioritize common/important vocabulary.
3. Limit: Take only the first --limit from the sorted list.

3. Generate Sentences

Group the target words (from the --source-field) into batches of --batch-size.

Generation Instruction (Japanese Example):

"You are an expert Japanese linguist. Generate one 'i+1' example sentence for each of the following target words.

LEARNED VOCABULARY: [Insert content of extract_learned_vocab.py stdout here]

TARGET WORDS: [List of target words from batch]

CONSTRAINTS:

1. Use ONLY words from the learned list plus the target word.
2. Use a natural and conversational tone.
3. Bold the target word using <b> tags.
4. [Language Specific: e.g. For Japanese, use Anki furigana style: 漢字[ふりがな] with a leading space].
5. Provide output as a JSON object: {\"word\": {\"sentence\": \"...\", \"english\": \"...\"}}."

4. Apply Updates

Map the generated sentences and translations back to the original Note IDs. Update Anki using the multi action to bundle multiple updateNoteFields and addTags requests into a single network call.

• Field Mapping:
  • [--target-field]: The generated sentence with proper formatting and bolding.
  • [--target-field-english]: The English translation of the sentence.

Guidelines

• Smart Fill Rule: NEVER process notes that already have content in the --target-field. This is the primary filtering mechanism.
• Priority: Always sort by Frequency before limiting by --limit.
• Batching: Always respect the --batch-size to maintain quality and avoid context limits.
• Language Formatting:
  • Japanese: Strictly follow the 漢字[ふりがな] format with the preceding space.
  • Other Languages: Adjust formatting requirements (e.g. accents, gender agreement) in your internal instructions.
• i+1 Principle: The core value of this skill is the restricted vocabulary. Ensure this constraint is strictly followed.