Language Tutor

Language Tutor — Foreign Language Learning Full Pipeline

An agent team collaborates to deliver level testing, curriculum design, lesson generation, quiz creation, and review management.

Execution Mode

Agent Team — 5 agents communicate directly via SendMessage and cross-verify each other's work.

Agent Roster

Workflow

Phase 1: Preparation (Orchestrator performs directly)

1. Extract from user input:
   • Target language: Which foreign language the learner wants to study
   • Native language: The learner's native language (default: English)
   • Learning experience: Prior study experience and self-assessed current level
   • Learning goal: Travel / work / exam / immigration / hobby, etc.
   • Available time: Weekly hours available for study
   • Existing materials (optional): Prior study materials, exam scores, etc.
2. Create a _workspace/ directory at the project root
3. Organize the input and save it to _workspace/00_input.md

Phase 2: Team Assembly and Execution

Inter-agent communication flow:

• level-assessor completes -> sends skill-area levels, strengths/weaknesses, learning goals, and available time to curriculum-designer
• curriculum-designer completes -> sends Week 1 topics, grammar, and vocabulary to lesson-tutor; sends milestone criteria to quiz-master
• lesson-tutor completes -> sends key learning items to quiz-master; sends learning items to review-coach
• quiz-master -> sends results to review-coach -> weakness reinforcement + review schedule generated

Continuous learning cycle: Lessons, quizzes, and reviews run iteratively. When the user requests more lessons:

1. lesson-tutor generates the next lesson (03_lesson_02.md, 03_lesson_03.md, ...)
2. quiz-master creates a quiz for that lesson (04_quiz_02.md, ...)
3. review-coach updates the review plan + generates a progress report

Phase 3: Progress Reporting

review-coach periodically generates progress reports:

• _workspace/06_progress_report.md — Overall progress, skill-area growth, quiz trends, study patterns

Task-Scale Modes

Automatic level detection: If the user does not know their level, provisionally estimate it from the expressions they use in conversation and suggest a diagnostic test.

Data Transfer Protocol

File naming convention: {order}_{deliverable}_{number}.{extension}

Error Handling

Test Scenarios

Normal Flow

Prompt: "I want to get better at English speaking. I travel abroad often for work, and I can study about 5 hours per week." Expected Results:

• Level assessment: Adaptive diagnostic test administered; CEFR level determined per skill area
• Curriculum: 12-week business English plan focused on conversation + emails + presentations
• First lesson: Business greetings and self-introduction patterns + 20 key vocabulary items
• Quiz: 15 items based on Lesson 1 content
• Review plan: Spaced repetition schedule + weakness reinforcement activities

Existing File Flow

Prompt: "Build a weakness-focused curriculum based on this TOEIC score report" + score report attached Expected Results:

• Score report copied as reference material in _workspace/01_level_assessment.md
• level-assessor determines skill-area levels based on the score report (streamlined diagnostic test)
• curriculum-designer + lesson-tutor + quiz-master + review-coach deployed

Error Flow

Prompt: "Maybe I will try learning a language" Expected Results:

• Target language not specified -> suggest English by default + offer alternatives
• Learning goal unclear -> propose 3 goals (travel / work / hobby) and ask user to choose
• After selection, execute full pipeline

Agent Extension Skills

Extension skills that enhance each agent's domain expertise: