LearningOS — powered by Drona

 ·  129 agents · 779 skills · 122 roles

A modular, config-driven GitHub Copilot Agent framework for learning, teaching, research, and technical career growth. One master orchestrator (Drona, the guru), a set of specialist sub-agents, and reusable skills — all in the official GitHub Copilot format so they work in VS Code, VS Code Insiders, and the GitHub Copilot CLI.

🌐 Live site: https://arasanirohithreddy.github.io/LearningOS/  ·  ⬇ Download: latest release  ·  🤝 Contribute: open an issue · start a discussion · CONTRIBUTING.md

LearningOS is not a single chatbot prompt. It is a small platform: agents compose reusable skills, behavior is shared through a single constitution (AGENTS.md), and new role-agents are built from tiny YAML config files instead of hand-written prompts.

Highlights

• Teaches, not just answers — first-principles, trade-offs, pitfalls, and a signature Learning Footer on every substantive reply (AGENTS.md).
• 129 agents · 779 skills · 122 role configs — Drona routes to specialist mentors and reusable /-skills across every technical domain.
• Visual-first learning — Mermaid, tables, KaTeX and step-traces by default, and Drona asks your preferred learning style up front (/visual-explainer).
• 361 no-subscription labs + run code in 90+ languages with nothing installed; practice AWS / Azure / GCP / Oracle locally via floci.
• 414 curated free resources — incl. 275 YouTube channels — plus dated tech news from 190+ RSS/Atom feeds and 130+ roadmap.sh paths turned into personalized plans (resources · news).
• Competitive programming & DSA — LeetCode / Codeforces-style practice: /competitive-programming-drill, /dsa-patterns-coach, /dynamic-programming-coach, /graph-algorithms-coach.
• Own MCP server (mcp/) — exposes the whole catalog (search skills & agents, resources, roadmaps, news, run code) to Claude Desktop, VS Code and Cursor.
• Config-driven & multi-host — VS Code, Insiders, Copilot CLI, Claude, Cursor, Gemini; add a role in YAML with /role-composer.

---

Why this exists

Most "learning agents" are one giant prompt that is hard to extend and inconsistent across tools. LearningOS instead uses the same primitives GitHub Copilot itself uses:

• Agents (.github/agents/*.agent.md) — personas with their own tools and responsibilities.
• Skills (.github/skills/<name>/SKILL.md) — on-demand workflows, invocable with /.
• Constitution (AGENTS.md) — shared teaching behavior, auto-loaded everywhere.
• Role configs (.github/roles/*.role.yml) — compose new specialist agents from config.

Everything optimizes for teaching, not just answering. See the teaching principles and the signature "Learning Footer" in AGENTS.md.

Architecture

flowchart TD
    U([Learner]) --> D{{Drona · Master Orchestrator}}

    D -->|delegates| CM[Coding Mentor]
    D -->|delegates| RN[Research & News Analyst]
    D -->|delegates| IC[Interview Coach]
    D -->|delegates| EC[Exam & Certification Coach]
    D -->|delegates| CA[Career Mentor]
    D -->|delegates| MP[Meeting & Presentation Coach]
    D -->|delegates| RO["122 Role-Agents<br/>Software · Languages · Web · Data · AI/ML · Cloud · DevOps · Security · Design · Product · Emerging · Enterprise"]

    subgraph Skills["779 Reusable Skills — invoke with /"]
      S1["Learn · Plan · Practice<br/>concept-explainer · socratic-tutor · worked-example · analogy-generator<br/>learning-roadmap · career-ladder · quiz-generator · mock-exam · exam-blueprint"]
      S2["Code & engineering<br/>code-review-coach · debugging-coach · refactoring-coach · test-writer<br/>sql-coach · git-coach · dockerfile-coach · regex-explainer · code-optimizer"]
      S3["Architecture & AI/data<br/>architecture-diagram · api-design-review · data-modeling-drill · threat-model<br/>prompt-optimizer · rag-designer · eval-designer · dataset-explorer"]
      S4["Research · Writing<br/>research-brief · daily-digest · official-docs-finder · paper-summarizer<br/>readme-generator · adr-writer · runbook-writer · changelog-writer"]
      S5["Career & communication · Build<br/>resume-tailor · cover-letter · linkedin-optimizer · coding-interview-drill<br/>slide-outline · demo-script · role-composer"]
    end

    D -.invokes.-> Skills
    CM -.invokes.-> Skills
    RN -.invokes.-> Skills
    IC -.invokes.-> Skills
    EC -.invokes.-> Skills
    CA -.invokes.-> Skills
    MP -.invokes.-> Skills
    RO -.invokes.-> Skills

    Skills -.reads.-> AG[[AGENTS.md — shared constitution]]
    D -.reads.-> AG

Loading

Repository layout

Learning Agents/
├── AGENTS.md                      # Shared teaching constitution (auto-loaded everywhere)
├── README.md                      # This file
├── LICENSE                        # MIT
├── CONTRIBUTING.md                # How to add roles / skills / agents / docs
├── CODE_OF_CONDUCT.md
├── docs/                          # The framework "brain" — how & why it works (20 guides)
│   ├── Architecture.md  Agents.md      Skills.md       Roles.md      Standards.md
│   ├── PluginSDK.md     Marketplace.md Extension.md    Install.md    Customize.md
│   ├── CodeExecution.md LocalPractice.md Floci.md      MCP.md        Memory.md
│   ├── News.md          Sources.md     Security.md     Testing.md    Roadmap.md
│   └── mcp.sample.json            # Copy servers into .vscode/mcp.json to enable MCP
├── .github/
│   ├── agents/                    # 129 custom agents (personas)
│   │   ├── drona.agent.md         # ⭐ Master orchestrator
│   │   ├── coding-mentor · research-analyst · interview-coach
│   │   ├── exam-coach · career-mentor · meeting-prep       # 6 core mentors
│   │   └── + 122 role-agents (python/rust/scala-developer, data-engineer, quantum-computing-engineer, …)
│   ├── skills/                    # 779 reusable workflows (slash commands), across 27 groups
│   │   ├── Learn: concept-explainer socratic-tutor misconception-buster analogy-generator worked-example glossary-builder mind-map note-generator teach-back knowledge-graph cheat-sheet
│   │   ├── Plan/Assess: learning-roadmap onboarding-plan career-ladder progress-tracker spaced-repetition-scheduler project-mentor practice-generator quiz-generator flashcards mock-exam skill-assessment gap-analysis rubric-grader exam-blueprint
│   │   ├── Code: code-review-coach debugging-coach refactoring-coach test-writer pair-programmer regex-explainer sql-coach git-coach dockerfile-coach code-optimizer algorithm-visualizer complexity-analyzer code-walkthrough system-design-drill
│   │   ├── Architecture/AI-data: architecture-diagram api-design-review data-modeling-drill threat-model tech-comparison estimation-coach prompt-optimizer rag-designer eval-designer dataset-explorer
│   │   ├── Research/Writing: research-brief daily-digest feed-curator official-docs-finder engineering-blog-finder github-repo-finder paper-summarizer literature-review technical-writing-coach readme-generator adr-writer runbook-writer documentation-planner changelog-writer
│   │   └── Career/Build: resume-tailor cover-letter linkedin-optimizer salary-negotiation star-story-builder portfolio-reviewer coding-interview-drill whiteboard-explainer slide-outline demo-script case-study role-composer
│   └── roles/                     # 122 config-driven role definitions — add your own!
│       ├── _TEMPLATE.role.yml
│       ├── Software · Languages (Py/Java/C#/Go/Rust/C++/TS/Kotlin/Scala/PHP/Swift/Elixir/R/Dart/Haskell/Clojure) · Web
│       ├── AI/ML · MLOps · Deep Learning · LLMOps · CV · NLP
│       ├── Data & BI · Streaming · Big Data · Snowflake · Tableau · Looker · Governance
│       ├── Cloud (Azure/AWS/GCP) · Platform · Network · Terraform · Serverless
│       ├── DevOps · SRE · Observability · FinOps · Linux/Windows admin · Chaos · Perf
│       ├── Security (App/Cloud/SOC/DevSecOps/IAM/GRC/Privacy) · QA/SDET · Architecture
│       ├── Emerging (AR/VR · Quantum · Edge · Graphics · Bioinformatics · Quant · GIS · HPC)
│       └── Design · Docs · DevRel · Product · EM · TPM · BA · Scrum · Business/Support · Enterprise (SF/ServiceNow/SAP/D365)
├── data/                          # Curated catalogs: 193 news feeds, 134 roadmaps, 414 free learning resources
├── mcp/                           # LearningOS MCP server (learningos-mcp) — 9 tools · 2 resources · 3 prompts
├── extension/                     # VS Code extension (@drona + dashboard + Tech News / Roadmaps views)
├── templates/                     # Optional MCP config templates (e.g., floci.mcp.json)
├── marketplace/                   # Generated pack index (registry.json + CATALOG.md)
├── scripts/build-registry.mjs     # Scans agents/skills/roles → rebuilds the marketplace index

Documentation

The docs/ folder is the framework's brain — how and why it works:

Doc	Topic
Architecture	The five primitives, request flow, design principles
Agents	Persona roster + how the "100+ agents" vision is served
Skills	Skill catalog, authoring, and progressive loading
Roles	The config-driven role system + full 122-role catalog
PluginSDK	Packaging & distribution model
Install	Install across every host (Copilot CLI, VS Code/Insiders, Desktop, Claude, Cursor, Gemini) + one-command install.ps1
Platforms	Individual-level cross-platform setup (Claude Code/Desktop, Cursor, Gemini, M365 Cowork) — zip import, per-host MCP wiring, opt-in Azure CLI auth
Extension	Drona as an installable .vsix (@drona chat participant) — no Marketplace account needed
CodeExecution	Run code in 90+ languages with no local install (self-hosted Piston / onlinecompiler.io key) — setup + where to get a key
Customize	Add / edit / disable your own skills & agents; scaffolders + validate.mjs
Marketplace	Enterprise plugin marketplace: pack model + generated registry
LocalPractice	Practice every domain locally & free (no subscriptions) — the tool catalog
Floci — local cloud	Practice AWS, Azure, GCP & Oracle (OCI) locally & free with floci emulators — install, per-cloud endpoints, and the honest "AI-ready"/MCP story
MCP	Live docs/data via MCP + LearningOS's own MCP server (mcp/) for any client (Claude/VS Code/Cursor) + sample config
Memory	Learner profile, spaced repetition, RAG
News	News / RSS / research framework
Standards	Coding & teaching standards
Security	Security & Responsible AI guardrails
Testing	Structural + behavioral evaluation
Roadmap	Vision → status and what's next

Install & use

LearningOS is discovered automatically when this folder is your workspace/repo root. Nothing to build.

VS Code / VS Code Insiders (Agent Mode)

1. Open this folder in VS Code or VS Code Insiders (Copilot Chat + Agent Mode enabled).
2. In the Chat view, open the agent picker and choose Drona (or any specialist).
3. Ask naturally — e.g. "Teach me how Kubernetes scheduling works and quiz me."
4. Invoke a skill directly by typing / and picking one, e.g. /learning-roadmap, /quiz-generator.

Skill discovery locations are already enabled in .vscode/settings.json via chat.agentSkillsLocations (.github/skills is included). Skills under .github/skills/ and agents under .github/agents/ are picked up with no extra setup.

GitHub Copilot CLI

1. Install the CLI and sign in (copilot), then cd into this folder.
2. AGENTS.md is loaded automatically as shared guidance.
3. Custom agents in .github/agents/ and skills in .github/skills/ are available in the session.
4. Ask Drona to teach, plan, quiz, or research — the same behavior as in VS Code.

Use it in another project

Copy AGENTS.md and the .github/agents/, .github/skills/, and .github/roles/ folders (and optionally docs/) into any repository. That repo instantly gains the full LearningOS toolkit.

Quick start prompts

• @Drona teach me the CAP theorem from first principles, then give me 5 quiz questions.
• /learning-roadmap 90-day plan to become an Azure AI Engineer, ~1 hour/day
• /research-brief latest changes in the Model Context Protocol (MCP) spec, official sources only
• @Interview Coach run a 30-minute system design mock for a URL shortener and score me
• /code-review-coach review this function and teach me what to improve
• /flashcards 20 spaced-repetition cards on Kubernetes networking

Practice cloud & tech locally (no subscription needed)

LearningOS includes 361 hands-on labs (*-lab skills) so you learn by doing — per language (Python, JS, TS, Go, Rust, Java, C#), framework (React), data (pandas, SQL, Spark, Kafka, PyTorch), and tooling (Git, Docker/Kubernetes).

For AWS / Azure / GCP / Oracle (OCI), you can do the cloud labs entirely offline against the free, open-source Floci local emulators — docker compose up, no cloud account, token, or paid tier. The dedicated skills walk you through setup:

• /floci-aws-local-lab — AWS locally via Floci (http://localhost:4566)
• /floci-azure-local-lab — Azure locally via Floci AZ (http://localhost:4577)
• /floci-gcp-local-lab — GCP locally via floci-gcp (http://localhost:4588)
• /floci-oracle-local-lab — Oracle (OCI) locally via floci-oci (http://localhost:4599)

Point your existing SDK / CLI / Terraform at the local endpoint and run the aws-*-lab, azure-*-lab, and gcp-*-lab exercises for free. (Floci is an emulator for learning/dev/testing — verify behavior against the official cloud docs before production.)

And it's not just cloud — you can practice databases (Postgres, MySQL, MongoDB, Redis, SQLite), messaging (Redpanda/Kafka, RabbitMQ, NATS, MQTT), Kubernetes (minikube, kind, k3d), observability (Prometheus + Grafana, Jaeger), auth (Keycloak), and even LLMs & RAG (Ollama + Chroma/Qdrant/pgvector — no API key or cost) entirely on your laptop. The full, verified catalog with the matching labs is in docs/LocalPractice.md.

Extend it — build a new role-agent from config

Instead of writing a new prompt, describe the role in YAML and let LearningOS build the agent:

1. Copy .github/roles/_TEMPLATE.role.yml to .github/roles/my-role.role.yml and fill it in.
2. Run the role-composer skill: /role-composer .github/roles/my-role.role.yml.
3. A ready-to-use .github/agents/<my-role>.agent.md is generated in the correct format.

122 ready-made role configs ship across every domain (Software, Programming Languages, Web & Frameworks, AI/ML, Data/BI, Cloud & Platform, DevOps/SRE, Security, QA & Testing, Design/Docs, Architecture, Product & Management, Emerging & Specialized Tech, Business/Support, Enterprise Platforms) — see the full catalog in docs/Roles.md. Two good starting points: data-engineer.role.yml and azure-ai-engineer.role.yml.

Roadmap / where to take it next

The core is intentionally small and high-quality, and designed to grow. Phases 1–3 are done — orchestrator, shared constitution, 779 skills, 129 agents (Drona + 6 mentors + 122 role-agents), config-driven roles + composer, and learner templates. Phase 4 is in progress — the roster is well past the 100+ agent milestone and a first-cut Enterprise Plugin Marketplace now ships (generated registry + catalog) (see docs/Roadmap.md):

• Breadth toward 100+ agents / 500+ skills — added in reviewed, high-quality batches across every domain (each is one role config or one skill folder; no stubs). 129 agents / 779 skills and counting.
• Live MCP integrations (when available in your client): GitHub, Fetch/Web, Playwright, Microsoft Learn, RSS/arXiv, Memory, SQLite — to pull live docs, feeds, and papers into lessons.
• Learning analytics & evaluation: progress / gaps / streaks over the learner profile, and a golden-prompt evaluation harness.

License

MIT. Built to be cloned, extended, and shared.