New:Socket for Asana Is Now Available.Learn more
Get Started

vibelearn

Package Overview
Dependencies
Maintainers
1
Versions
5
Alerts
File Explorer

Advanced tools

Socket logo

Install Socket

Detect and block malicious and high-risk dependencies

Install

vibelearn

Developer learning tool — analyze coding sessions, extract concepts, generate quiz questions

latest
Source
npmnpm
Version
0.1.6
Version published
Weekly downloads
3
-40%
Maintainers
1
Weekly downloads
 
Created
Source


VibeLearn

Learn from every session. Built for Claude Code.

License Version Node

Quick StartHow It WorksVL CLIConfigurationPrivacyTroubleshooting

VibeLearn is a Claude Code plugin that silently watches what you build, extracts the concepts you encounter, and turns them into quiz questions — so you actually retain what you learn while coding.

What It Does

Every time you end a Claude Code session, VibeLearn automatically:

  • Detects your tech stack — reads package.json, pyproject.toml, go.mod, etc.
  • Analyzes your code changes — identifies patterns: custom hooks, API routes, TypeScript types, design patterns
  • Extracts learning concepts — a single LLM call produces a session summary and a list of concepts you encountered
  • Generates quiz questions — a second LLM call creates targeted questions per concept across 7 formats: multiple_choice, code_reading, spot_the_bug, fill_in_blank, open_ended, true_false, and ordering — selected based on concept difficulty (junior / mid / senior)
  • Syncs to vibelearn.dev — your learning profile is stored securely (optional, requires vl login)

Then run vl quiz to review what you learned.

Quick Start

Install the plugin in a Claude Code session:

/plugin marketplace add anergcorp/vibelearn
/plugin install vibelearn

Restart Claude Code. VibeLearn will start capturing learning data automatically from your next session.

Optional — connect to vibelearn.dev:

vl login <your-api-key>

Get your API key at vibelearn.dev.

vl CLI

The vl command lets you review and interact with your learning data:

vl quiz              # Interactive quiz — all pending questions
vl quiz --session    # Quiz questions from the last session only

vl status            # Sessions analyzed, top concept categories, mastery stats
vl gaps              # Concepts you haven't mastered yet (mastery < 50%)

vl login <api-key>   # Connect to vibelearn.dev
vl login --status    # Check login status

Example session:

$ vl quiz

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  VibeLearn Quiz — 3 questions
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Q1/3 (intermediate) [React Server Actions]

  Code:
    'use server'
    export async function createPost(data: FormData) { ... }

  What does the 'use server' directive tell Next.js?

  A) Run this function in a Web Worker
  B) Execute this function on the server, not the client
  C) Cache the function result server-side
  D) Mark the function as async-only

  Your answer (A/B/C/D): B

  ✓ Correct!

  Explanation: 'use server' creates a Server Action — a function that
  runs exclusively on the server. The client receives only the result.

Adaptive Difficulty

VibeLearn tracks your mastery per concept and adjusts difficulty automatically:

  • 3 correct in a row → promoted to the next level (junior → mid → senior)
  • 1 wrong answer → dropped back a level
  • Follow-up questions are inserted mid-quiz when you nail a junior question, to probe deeper understanding
  • Open-ended answers at senior level are evaluated by a 5-dimension rubric (accuracy, depth, trade-offs, practical reasoning, completeness)

How It Works

5 Lifecycle Hooks

SessionStart    → Worker starts, session initialized
UserPromptSubmit → Session linked to user prompt
PostToolUse     → File edits/writes/bash commands captured
Stop (Summary)  → 5-step analysis pipeline runs
SessionEnd      → Session finalized

Analysis Pipeline (runs at session end)

1. StackDetector   — reads package.json/config files → vl_stack_profiles
2. StaticAnalyzer  — regex/AST patterns on code changes (hooks, routes, types…)
3. ConceptExtractor — LLM call → session summary + concept list → vl_concepts
4. QuizGenerator   — LLM call → quiz questions per concept → vl_questions
5. UpstreamSync    — HMAC-signed POST to api.vibelearn.dev (queued offline if unavailable)

Worker Service

An Express HTTP server on port 37778, managed by Bun. Hooks talk to it over localhost. It handles all database writes and the analysis pipeline.

Database

SQLite at ~/.vibelearn/vibelearn.db. Key tables:

TablePurpose
vibelearn_session_summariesHuman-readable session narratives
vl_conceptsExtracted concepts per session
vl_questionsGenerated quiz questions
vl_quiz_attemptsYour answers (HMAC-signed before sync)
vl_developer_profileMastery score per concept
vl_stack_profilesDetected tech stack per session
vl_sync_queueOffline retry queue

Configuration

Settings are auto-created at ~/.vibelearn/settings.json on first run.

Key settings:

{
  "VIBELEARN_WORKER_PORT": "37778",
  "VIBELEARN_DATA_DIR": "~/.vibelearn",
  "VIBELEARN_LOG_LEVEL": "INFO",
  "VIBELEARN_PROVIDER": "claude",
  "VIBELEARN_GEMINI_API_KEY": "",
  "VIBELEARN_OPENROUTER_API_KEY": "",
  "VIBELEARN_AUTO_SYNC": "true",
  "VIBELEARN_EXCLUDED_PROJECTS": ""
}

AI Provider for Analysis

The analysis pipeline (concept extraction + quiz generation) uses your configured LLM provider. Priority order:

  • Gemini — set VIBELEARN_GEMINI_API_KEY (free tier available)
  • OpenRouter — set VIBELEARN_OPENROUTER_API_KEY
  • Anthropic — uses ANTHROPIC_API_KEY from environment (claude-haiku-4-5)

Excluding projects:

{
  "VIBELEARN_EXCLUDED_PROJECTS": "/path/to/skip,~/personal/*"
}

Privacy

Wrap any content in <private> tags to prevent it from being stored or synced:

Please review <private>my-secret-api-key: sk-...</private> configuration

Everything inside <private> is stripped at the hook layer before reaching the worker or database.

What is never stored:

  • Absolute file paths (only basenames are sent upstream)
  • Raw file contents (only short snippets from the analysis)
  • Full user prompts

Anti-tamper: Quiz attempt records are HMAC-signed with your API key before syncing. The server recomputes your streak from accepted attempt records — local SQLite data cannot be used to fake progress.

System Requirements

  • Node.js: 18.0.0 or higher
  • Claude Code: Latest version with plugin support
  • Bun: JavaScript runtime (auto-installed if missing)

Windows Notes

If you see npm : The term 'npm' is not recognized:

Install Node.js and restart your terminal. Bun is auto-installed by the plugin setup script.

Build

npm install
npm run build-and-sync   # Build + sync to marketplace + restart worker

Built outputs land in plugin/scripts/:

  • worker-service.cjs — the worker daemon
  • mcp-server.cjs — MCP tools
  • vl-cli.cjs — the vl binary

Contributing

  • Fork the repository
  • Create a feature branch
  • Make your changes with tests
  • Update documentation
  • Submit a Pull Request

License

GNU Affero General Public License v3.0 (AGPL-3.0).

See the LICENSE file for full details.

Support

Built with Claude Agent SDK | Powered by Claude Code | Made with TypeScript

Keywords

claude

FAQs

Package last updated on 26 Mar 2026

Related posts