VibeLearning

Learn while vibe coding - Spaced repetition learning system for AI coding agents

Turn your coding sessions into learning opportunities. VibeLearning automatically asks you questions about the concepts you're using, helping you truly understand the code you write with AI assistance.

Features

- Automatic Learning Triggers: Just install and it works across all sessions
- SM-2 Algorithm: Scientific spaced repetition (like Anki)
- 5-Level System: Recognition → Understanding → Comparison → Edge Cases → Architecture
- Fatigue Management: 15-min cooldown, auto-pause after consecutive skips
- Unknown Unknowns: Tracks concepts you don't know you don't know
- Report Export: Save learning reports as markdown files

Supported Platforms

| Platform | Auto Install | Features |
|----------|--------------|----------|
| OpenCode | ✅ | Plugin + MCP + /learn commands |
| Claude Code | ✅ | Hooks + MCP + /learn commands |
| Cursor | 📋 Manual | MCP server only |
| Cline | 📋 Manual | MCP server only |

Installation

$3

``bash


Install VibeLearning for OpenCode

npx vibe-learning install opencode

Restart OpenCode to apply changes

`

What gets installed:
- MCP server in ~/.config/opencode/opencode.json
- Plugin: vibe-learning-opencode (auto-installed via npm)
- Command: ~/.config/opencode/command/learn.md (for autocomplete)
- Plugin handles all /learn commands via chat.message hook

$3

`bash


Install VibeLearning for Claude Code

npx vibe-learning install claude-code

Restart Claude Code to apply changes

`

What gets installed:
- MCP server in ~/.claude.json
- Command: ~/.claude/commands/learn.md
- Hooks: ~/.vibe-learning/hooks/ (activity tracking + learning trigger)
- Hook configuration in ~/.claude/settings.json

$3

`bash


Auto-detect and install for all supported platforms

npx vibe-learning install

Check installation status

npx vibe-learning status
`

$3

Add to your MCP configuration:

`json
{
"mcpServers": {
"vibe-learning": {
"command": "npx",
"args": ["-y", "vibe-learning-mcp"]
}
}
}
`

How It Works

1. Request a task from your AI coding agent
2. AI asks 2-3 questions about your approach (Senior mode, default)
3. Explain your reasoning - why this approach? what alternatives?
4. AI implements after you demonstrate understanding
5. Progress is tracked using spaced repetition for optimal retention

Usage

$3

Both OpenCode and Claude Code support the /learn command:

#### Status & Control

| Command | Description |
|---------|-------------|
| /learn | Show current status |
| /learn on | Full learning mode (senior + after, default) |
| /learn off | Turn off learning (recording continues) |
| /learn senior | Enable senior mode (pre-implementation questions) |
| /learn senior off | Disable senior mode |
| /learn after | Enable after mode (post-implementation questions) |
| /learn after off | Disable after mode |
| /learn pause | Pause for 1 hour |
| /learn focus [area] | Focus questions on a specific area |

#### Statistics & Reports

| Command | Description |
|---------|-------------|
| /learn stats | Learning statistics dashboard |
| /learn report | Weekly learning report |
| /learn report month | Monthly learning report |
| /learn report --save | Save report as markdown file |
| /learn unknowns | Unknown unknowns dashboard |
| /learn unknowns --save | Save unknowns as markdown file |
| /learn review | Start reviewing due concepts |
| /learn interview | Interview practice mode |

$3

| Level | Name | Question Style |
|-------|------|---------------|
| 1 | Recognition | "Do you know what X is?" |
| 2 | Understanding | "Can you explain how X works?" |
| 3 | Comparison | "When would you use X instead of Y?" |
| 4 | Edge Cases | "What are the pitfalls of X?" |
| 5 | Architecture | "How would you design X at scale?" |

$3

When asked a learning question, respond naturally. The AI will evaluate your answer as:
- correct - You understood it well → Level up possible
- partial - Partial understanding → Stay at current level
- incorrect - Wrong answer → May level down
- skipped - Too busy right now → Asked again later

$3

VibeLearning uses two independent toggles that can be combined:

| Toggle | Default | Behavior |
|--------|---------|----------|
| Senior Mode | ✅ On | Pre-implementation questioning (2-3 rounds) |
| After Mode | ✅ On | Post-task learning questions |

Mode Combinations:

| Senior | After | Result |
|--------|-------|--------|
| ✅ | ✅ | Full learning (default) - Questions before AND after |
| ✅ | ❌ | Senior only - Deep pre-implementation review |
| ❌ | ✅ | After only - Quick post-task questions |
| ❌ | ❌ | Off - No questions (recording continues) |

#### Senior Mode Details

Senior mode simulates a strict senior developer reviewing your decisions:

`
[VibeLearning Senior Mode]
[Round 1/3] Why this approach?

[Round 2/3] What alternatives did you consider? Tradeoffs?

[Round 3/3] (if needed) Hint: ... What do you think?

Good thinking! [Summary]. Implementing now.
`

- Minimum 2 rounds required - Won't skip to implementation on first answer
- Strict evaluation - Only proceeds when you demonstrate understanding of alternatives and tradeoffs
- Educational summary - Provides brief explanation before implementing
- Learning recorded - Discussions are recorded for spaced repetition review

MCP Tools Reference

For direct MCP integration or custom implementations:

| Tool | Description |
|------|-------------|
| should_ask_question | Check if it's a good time to ask |
| get_concept_level | Get current level (1-5) for a concept |
| record_learning | Record learning result |
| record_unknown_unknown | Register a concept the user might not know |
| get_stats | Get learning statistics |
| get_report_data | Get comprehensive report data |
| get_unknown_unknowns | Get list of unexplored concepts |
| mark_explored | Mark a concept as explored |
| get_due_reviews | Get concepts due for review |
| get_mode / set_mode | Manage learning mode |
| save_report | Save report to markdown file |
| save_unknowns | Save unknowns to markdown file |

Data Storage

- Database: ~/.vibe-learning/learning.db (SQLite)
- Reports: ~/.vibe-learning/*.md
- Custom path: Set VIBE_LEARNING_DB_PATH environment variable

Uninstallation

$3

`bash
npx vibe-learning uninstall
`

This removes:
- OpenCode plugin and command files
- Claude Code command and hook files

$3

#### OpenCode

1. Remove plugin from ~/.config/opencode/opencode.json:
`json
{
"plugin": ["vibe-learning-opencode"] // Remove this line
}
`

2. Remove MCP server from ~/.config/opencode/opencode.json:
`json
{
"mcp": {
"vibe-learning": { ... } // Remove this block
}
}
`

3. Delete command file:
`bash
rm ~/.config/opencode/command/learn.md
`

#### Claude Code

1. Remove MCP server from ~/.claude.json:
`json
{
"mcpServers": {
"vibe-learning": { ... } // Remove this block
}
}
`

2. Delete command file:
`bash
rm ~/.claude/commands/learn.md
`

3. Remove hooks from ~/.claude/settings.json:
- Remove VibeLearning entries from hooks.PostToolUse and hooks.Stop

4. Delete hook files:
`bash
rm -rf ~/.vibe-learning/hooks/
`

$3

To also remove all learning data:

`bash
rm -rf ~/.vibe-learning/
`

Development

`bash


Clone the repository

git clone https://github.com/your-repo/vibe-learning.git
cd vibe-learning

Install dependencies

npm install

Build

npm run build

Run tests

npm test

Lint

npm run lint
`

Architecture

`
┌─────────────────────────────────────────────────────────┐
│ VibeLearning │
├─────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ OpenCode │ │ Claude Code │ │ Cursor │ │
│ │ Plugin │ │ (Hooks) │ │ (MCP) │ │
│ └──────┬──────┘ └──────┬──────┘ └──────┬──────┘ │
│ │ │ │ │
│ ▼ ▼ ▼ │
│ ┌─────────────────────────────────────────────────┐ │
│ │ MCP Server (Core) │ │
│ │ - SM-2 Algorithm │ │
│ │ - SQLite Database │ │
│ │ - 13 Tools │ │
│ └─────────────────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────┘
``

License

MIT