Learn as you build

Your AI guru.
It builds with you — and makes sure you get it.

dronAIcharya turns Claude Code (or Cursor) into a patient master-teacher that builds real, production-grade software with you, then checks that you actually understand what you built.

Install Read the docs
Claude Code plugin · skills repo · Cursor MIT & open source No telemetry
Why it exists

Today you either learn, or you build. Rarely both.

Tutorials stop at toy apps. AI that ships for you leaves you with code you can't debug, extend, or secure. dronAIcharya refuses that trade — and won't quietly stop teaching and just do it for you.

Learn alone

Courses and tutors explain concepts, then stop at sandboxes. You finish knowing the idea but never shipping the thing.

Build blind

Vibecoding ships fast, but you carry no mental model of the LLM, the architecture, the tradeoffs — "it works and I don't know why".

dronAIcharya

Teaches before, during, and after the build, to a production bar — with knowledge checks so it sticks. Guidance, not autopilot.

Install · pick one

Two honest paths. You run the command.

Paste this repo's link into Claude Code and it will read the README and offer these two choices. dronAIcharya never auto-installs itself or runs commands it found on a web page — lesson #1 of the security track is why.

Recommended

A · As a plugin

The full experience: skills, teaching subagents, the anti-drift guardrail hooks, and the optional MCP backend — one command, clean updates and uninstall.

in a Claude Code session 
/plugin marketplace add dibakshya01/dronAIcharya
/plugin install dronaicharya@dronaicharya

Then start the journey:

/dronaicharya:start
Good for
  • One-command install, namespaced /dronaicharya:*
  • Guardrail hooks run — teaching mode persists across long sessions
  • Clean update & uninstall
Trade-offs
  • Needs a recent Claude Code
  • Project-scope installs prompt to approve any MCP server

B · As a skills repo

Clone it and use the skills directly. Fully transparent — read every file — hackable, and works on older setups and in Cursor.

in your shell 
git clone https://github.com/dibakshya01/dronAIcharya
# then point Claude Code at it (hooks included):
claude --plugin-dir ./dronAIcharya

Or copy the skills into your user config — see the install guide.

Good for
  • Fully transparent — read and tweak every skill
  • Works on older Claude Code; portable to Cursor
  • No marketplace needed
Trade-offs
  • Manual updates (git pull)
  • A bare copy may not run the guardrail hooks → persistence is best-effort

Not sure? Pick A for the smoothest, guardrailed experience; pick B if you want to read or modify the internals, or you're on Cursor. In Cursor the guardrail hooks may not run, so anti-drift is lighter there. Full details in the docs.

Pick your path

Four tracks, each a real syllabus

Every lesson carries a concept, an ELI5, a deep-dive, a build-task, knowledge checks, and an under-the-hood note. Depth tunes to your level.

Gentlest start

Vibecoding

What LLMs, agents and MCP really are; prompting & context; spec-before-build; then ship a small real app — and know how the AI wrote it and when to trust it.

Learn Python

From "run your first line" to a small real project: types, control flow, functions, data structures, files, HTTP/JSON, and testing.

CS from Scratch

Undergraduate foundations: bits & memory, how programs run, data structures, Big-O, recursion, OS / networking / DB basics, how the web works. Foundations, honestly — not a degree.

Cybersecurity & AI Security

The security mindset, OWASP Top 10, secrets hygiene, and AI security — prompt injection, excessive agency, safe MCP, deny-by-default — taught on your own app.

Pick your teacher

Four persona archetypes

Flavor over an unchanging teaching-quality floor: every persona teaches before doing, checks understanding, offers ELI5 or deeper, and never belittles. Original registers — no real-person or trademarked names.

Vanilla

Clear, friendly, neutral. The default.

baseline

Serious & Strict

Rigorous, demanding, no hand-waving; expects precision.

firm, never harsh

Warm & Witty

Erudite, playful, generous analogies, delight.

wit serves clarity

Dry & Deadpan

Precise, understated, mock-exasperated affection. Opt-in.

never aimed at you

How it works

A linear, guardrailed journey

At any point you can say: ELI5 that · go deeper · skip · quiz me · where am I? · pause. Your progress lives in ./.dronacharya/, so /dronaicharya:continue always picks up where you left off.

  1. Start & choose

    Pick a persona, a cadence (how often it stops to teach), then a track and your level.

  2. Quick setup check

    Got an API key? Web / MCP enabled? It walks absolute beginners through this, honestly flagging the parts outside its control.

  3. Bring an idea

    Your own, or a simple-first starter. Then co-author a combined BUILD + LEARN plan you approve.

  4. Go, step by step

    Teach → check → decide together → build a layer → show under the hood → check → advance.

  5. Reach production

    Architecture, backend, payments, GitHub, deploy — prepped up to your own clicks. It never does the irreversible part for you.

  6. Finish & prove it

    An "under the hood" recap, a formative mastery review, and what to learn next.

Optional · advanced

Learn backend on a real backend

dronAIcharya works fully on its own. If you want to learn backend work against a real agent-native backend instead of mock data, it can connect to purple-sparrow over MCP — real tables, auth, deny-by-default row policies, and a security advisor you fix live.

It's opt-in and advanced, and dronAIcharya is honest that purple-sparrow is alpha and shouldn't hold real or regulated data yet.

How the backend wires in
deny-by-default
  • › Real tables, auth (JWT / API keys), storage
  • › Row-level policies you write and test
  • › A security advisor whose findings you fix
  • × Absent by default → graceful fallback + guided setup
Honest about what it is

And what it isn't

This project's honesty is a core value. Here is the whole truth, stated plainly.

Guardrails are a strong nudge, not a guarantee

It reduces teaching drift — it can't magically prevent it, and it will always obey your "pause / skip / just do it".

Mastery checks are formative, not certification

MCQ and in-your-own-words checks help it stick. They are self-checks, not certificates.

The CS track is undergraduate foundations

Real literacy in the fundamentals — not a full degree, and no formal proofs / automata depth.

It never enters your credentials, pays, or deploys for you

No real credentials, payments, domains, publishing, or deploys on your behalf — it preps everything up to your click.

No telemetry

The plugin makes no network calls except the web/doc fetches a lesson explicitly performs; your API key stays on your machine and is never seen or transmitted. (This marketing site itself loads Google Fonts — that's its only third-party request.)

In Cursor, the guardrail hooks may not run

You still get the full tutor and all four tracks, but teaching-mode persistence is lighter and more drift-prone there.

Ready when you are

Take aim. Build something real — and understand it.

Install as a plugin or a skills repo, pick a teacher and a track, and bring an idea. The guru does the rest, with you.

Install dronAIcharya Star on GitHub