Vocabulary Builder
Build and review vocabulary from books, podcasts, and daily encounters.
- Rating
- 4.1 (352 reviews)
- Downloads
- 7,534 downloads
- Version
- 1.0.0
Overview
Build and review vocabulary from books, podcasts, and daily encounters.
Complete Documentation
View Source →
Vocabulary Builder
This skill reads and writes the following files (paths relative to workspace):
- Vocabulary tracker:
memory/vocabulary.md— all word data lives here - Audio clips directory:
docs/tts-fr/— read-only; user places pronunciation audio files here
Reading Workflow
When the user wants to read/practice reading a book:
- Ask what book they're reading
- Check the vocabulary tracker for words from that book — read the END of the Active Words section (use
tailor read the last entries) to find the actual last word. Note:memory_searchmay return partial/ranked results, so also verify by reading the file directly when checking the latest entry. - Tell them the last word added + page number so they know where to continue
- Ask if they want to: practice pending words, or keep reading and add new ones
Adding New Words
When the user gives a new word:
- Give pronunciation (IPA, American English)
- Give meaning — clear, simple
- Give synonyms — similar words they might know
- Add to tracker after confirming with the user (e.g., "Adding [word] to your vocabulary — ok?")
- Show the word card after adding
## Active Words section, at the END (before --- separator for Long-Term Review).Word Entry Format
### [word]
- **Type:** noun/verb/adj/adv
- **Learned:** YYYY-MM-DD HH:MM TZ
- **Book:** [source name]
- **Page:** [number]
- **Pronunciation:** /IPA/
- **Meaning:** [explanation]
- **Synonyms:** [similar words]
- **Context:** "[sentence from source]"
- **Practice History:**
- YYYY-MM-DD HH:MM TZ: Step N ✓/✗ (notes)
French Words
- Context: French sentence only
- English Translation: Separate field with English translation
- Do NOT mix French and English in the same field
- User places audio clips in
docs/tts-fr/→ add Audio field with the file path
3-Step Learning Process
Run all 3 steps in one conversation flow (not spread across hours):
- Step 1: Show the plain word + review count (e.g. "reviewed 0 times" or "reviewed 3 times") → ask "Do you know the pronunciation?" → user types word to confirm
- Step 2: Ask "What does it mean?" → check if correct/close enough
- Step 3: Ask user to write a sentence using the word
Spaced Repetition Schedule
After completing all 3 steps, review at:
- Next day → 3 days → 1 week → 2 weeks → 1 month → 3 months
- After 3-month review: word is mastered
Word Progression
Words move through three sections in the tracker:
- Active Words — currently learning
- Long-Term Review Words — completed all steps, in spaced review
- Mastered Words — passed all reviews through 3 months
Quiz Rules
- One word per quiz — no rapid-fire
- No spam: If no reply to previous quiz today → don't send new one
- Reset next day: New day = can send quiz even if yesterday's unanswered
- Sleep hours: No messages 11 PM – 7 AM (user's timezone)
- Priority: due for review > newer words (incomplete steps) > refresher
Random Word Selection
- Count each word's total practice history entries (all steps and reviews combined)
- Sort all words by review count ascending (least reviewed first)
- Take the group with the minimum review count
- Within that group, randomize: count words (N), use
(current_timestamp_ms % N) + 1to pick position - Never pick the same word twice in a row. Check Quiz State for the last quizzed word and skip it.
On-Demand Quizzes
User can request specific quizzes anytime — these override normal priority and spam rules:
- "Quiz me" → random word
- "Quiz me on [word]" → specific word
- "Quiz me on words from [book]" → random from that book
- "Quiz me on words from this week" → last 7 days
- "Quiz me on [book] page [N] to [M]" → page range
- "Give me 3 quizzes" → run 3 words in a row
Quiz State Tracking
Keep a ## Quiz State section at the top of the tracker file:
## Quiz State
- **Pending quiz:** [word] ([review type], [step] sent)
- **Last quiz sent:** YYYY-MM-DD HH:MM TZ
Update after each quiz interaction.
Cron Setup
When the user asks to set up a scheduled vocabulary quiz:
- Ask for: frequency (default: every 1h), delivery channel + target
- Job name:
vocabulary-quiz-{agentName}(e.g.vocabulary-quiz-english) - Create a cron job with these settings:
- Session: isolated
- Timeout: 120 seconds
- Message: instruct the agent to use the
vocabulary-builderskill to run a quiz - Quiz constraints (include in the cron message):
- Check Quiz State — skip if pending quiz unanswered today
- Respect sleep hours: no messages 11PM–7AM (user's timezone)
References
Installation
openclaw install vocabulary-builder
💻Code Examples
- YYYY-MM-DD HH:MM TZ: Step N ✓/✗ (notes)
### French Words
- **Context:** French sentence only
- **English Translation:** Separate field with English translation
- Do NOT mix French and English in the same field
- User places audio clips in `docs/tts-fr/` → add **Audio** field with the file path
## 3-Step Learning Process
Run all 3 steps in one conversation flow (not spread across hours):
- **Step 1:** Show the plain word + review count (e.g. "reviewed 0 times" or "reviewed 3 times") → ask "Do you know the pronunciation?" → user types word to confirm
- **Step 2:** Ask "What does it mean?" → check if correct/close enough
- **Step 3:** Ask user to write a sentence using the word
Trust-based pronunciation — no voice/ASR check. User types word to confirm.
## Spaced Repetition Schedule
After completing all 3 steps, review at:
- Next day → 3 days → 1 week → 2 weeks → 1 month → 3 months
- After 3-month review: word is **mastered**
### Word Progression
Words move through three sections in the tracker:
1. **Active Words** — currently learning
2. **Long-Term Review Words** — completed all steps, in spaced review
3. **Mastered Words** — passed all reviews through 3 months
## Quiz Rules
- **One word per quiz** — no rapid-fire
- **No spam:** If no reply to previous quiz today → don't send new one
- **Reset next day:** New day = can send quiz even if yesterday's unanswered
- **Sleep hours:** No messages 11 PM – 7 AM (user's timezone)
- **Priority:** due for review > newer words (incomplete steps) > refresher
### Random Word Selection
1. Count each word's total practice history entries (all steps and reviews combined)
2. Sort all words by review count ascending (least reviewed first)
3. Take the group with the minimum review count
4. Within that group, randomize: count words (N), use `(current_timestamp_ms % N) + 1` to pick position
5. **Never pick the same word twice in a row.** Check Quiz State for the last quizzed word and skip it.
**Always show review count** when quizzing: e.g. "(reviewed 0 times)" or "(reviewed 3 times)".
### On-Demand Quizzes
User can request specific quizzes anytime — these override normal priority and spam rules:
- "Quiz me" → random word
- "Quiz me on [word]" → specific word
- "Quiz me on words from [book]" → random from that book
- "Quiz me on words from this week" → last 7 days
- "Quiz me on [book] page [N] to [M]" → page range
- "Give me 3 quizzes" → run 3 words in a row
On-demand refresher: If no words are due/pending, pick random word from all learned words. Still record quiz date.
## Quiz State Tracking
Keep a `## Quiz State` section at the top of the tracker file:### [word]
- **Type:** noun/verb/adj/adv
- **Learned:** YYYY-MM-DD HH:MM TZ
- **Book:** [source name]
- **Page:** [number]
- **Pronunciation:** /IPA/
- **Meaning:** [explanation]
- **Synonyms:** [similar words]
- **Context:** "[sentence from source]"
- **Practice History:**
- YYYY-MM-DD HH:MM TZ: Step N ✓/✗ (notes)## Quiz State
- **Pending quiz:** [word] ([review type], [step] sent)
- **Last quiz sent:** YYYY-MM-DD HH:MM TZTags
Quick Info
Ready to Install?
Get started with this skill in seconds
Related Skills
4claw
4claw — a moderated imageboard for AI agents.
Aap Passport
Agent Attestation Protocol - The Reverse Turing Test.
Acestep Lyrics Transcription
Transcribe audio to timestamped lyrics using OpenAI Whisper or ElevenLabs Scribe API.
Adaptive Suite
A continuously adaptive skill suite that empowers Clawdbot.