The assistant that writes code directly to your workspace. 21 specialist agents, 37 skills, and full project intelligence — built directly into GitHub Copilot Chat.
Install the extension. Open Copilot Chat. Type @shastra. That’s the entire setup.
Installation
- Open VS Code Extensions panel
Ctrl+Shift+X (Windows/Linux) or Cmd+Shift+X (Mac)
- Search AI Shastra
- Click Install
- Open GitHub Copilot Chat
Ctrl+Alt+I
- Type
@shastra and start
No terminal. No npm. No config files. No folders to copy. The extension is fully self-contained.
How It Works
AI Shastra runs entirely inside VS Code through GitHub Copilot Chat. There is nothing external to install or initialise. Every agent, skill, workflow, and intelligence module is bundled inside the extension and activates on demand.
@shastra Open chat Shastra greets you and explains options
@shastra /scan Read your entire project understands it in one pass
@shastra /create a sidebar component Build it directly in your workspace
@shastra /debug auth not working Root-cause analysis and auto-fix
@shastra /agent security-auditor Invoke a specific specialist directly
@shastra /plan add a payment flow Plan with your actual stack in mind
@shastra /plan View the project roadmap
🚀 Direct File Editing (v1.1+)
Unlike standard chat assistants, Shastra can read, create, and modify files in your workspace directly. When an agent proposes a change, it can execute the write command to apply that change to your project immediately.
You can also press F1 (Command Palette) and search Shastra to access every command without typing in chat.
/scan Full Project Understanding in One Command
Run /scan once. AI Shastra reads your workspace and instantly knows:
- What the project is SaaS, e-commerce, weather service, API, CLI tool, documentation site
- What it delivers the actual user-facing value proposition
- Tech stack language, frontend framework, backend, styling, testing libraries
- Infrastructure databases, cloud providers, auth, payments, CI/CD pipelines
- Routes and pages Next.js App Router, Pages Router, Express, Hono
- Environment variables extracted from
.env.example or .env.sample
- Third-party integrations Stripe, OpenAI, Twilio, SendGrid, Cloudinary, Sentry
- Architecture pattern monolith, microservices, serverless edge, monorepo
After scanning, every response is tailored to your actual project. No more generic answers.
Self-Evaluation and Real-Time Adaptation
AI Shastra adjusts and improves with every message in your session.
Session memory Shastra tracks every topic you discuss and your expertise level (auto-detected from your vocabulary). By turn 5, it knows whether to explain concepts step-by-step or go straight to the technical detail.
Real-time self-evaluation after every response, Shastra scores itself on completeness and stack-relevance. If it detects a vague or incomplete answer, it surfaces a suggestion automatically.
Self-healing on correction when you say "no that's wrong", "actually it's...", or "you missed...", Shastra:
- Detects the correction pattern automatically
- Records the correction in session memory
- Applies it to every answer for the rest of the session
- Confirms the healing inline
Expertise adaptation Shastra detects beginner vs expert signals in how you phrase questions. It continuously adjusts its response depth and vocabulary.
This is per-session adaptation within your conversation. It resets on a new session or when you run Shastra: Reset Session Memory from the Command Palette.
21 Specialist Agents
Use @shastra /agent <name> to invoke any specialist:
| Agent |
Expertise |
orchestrator |
Coordinates multiple agents for complex tasks |
codebase-intelligence |
Reads and maps your entire project in one pass |
frontend-specialist |
React, Next.js, Tailwind, accessibility, performance |
backend-specialist |
Node.js, Python, APIs, serverless, edge functions |
database-architect |
Schema design, SQL, Prisma, Drizzle, migrations |
devops-engineer |
Docker, CI/CD, cloud deployment, infrastructure |
security-auditor |
OWASP, vulnerability scanning, secure code review |
test-engineer |
Jest, Vitest, Playwright, testing strategy |
debugger |
Systematic root-cause analysis |
performance-optimizer |
Core Web Vitals, bundle analysis, profiling |
project-planner |
Task breakdown, dependency graphs, sprint planning |
documentation-writer |
README, API docs, JSDoc, architecture docs |
code-archaeologist |
Legacy code, refactoring, technical debt |
api-architect |
REST, GraphQL, tRPC design, OpenAPI specs |
data-scientist |
Data analysis, ML patterns, visualisations |
ux-researcher |
Heuristic evaluation, a11y audits, user flows |
green-stack-developer |
Full-stack cross-platform, React Native |
mcp-builder |
Model Context Protocol server development |
cursor-rules-agent |
IDE configuration, AI rule files, productivity |
innovation-catalyst |
Creative ideation, unconventional problem solving |
explorer-agent |
Codebase mapping, pattern recognition |
Slash Commands (12 Workflows)
| Command |
What It Does |
/scan |
One-shot full project intelligence |
/brainstorm |
3+ options with pros, cons, and a recommendation |
/create |
Full app build with coordinated agent handoffs |
/debug |
Reproduce, isolate, root-cause, fix, regression test |
/deploy |
Pre-flight TypeScript + lint + audit + bundle checks |
/enhance |
Add features to an existing codebase safely |
/orchestrate |
Run multiple specialists in parallel on one request |
/plan |
Task breakdown with file structure and dependency graph |
/preview |
Explore and document unfamiliar code |
/status |
Project health, open issues, progress summary |
/test |
Generate tests matched to your existing patterns |
/ui |
Design-first interface engineering for your actual stack |
Command Palette (F1 Search "Shastra")
Every feature is accessible from the VS Code Command Palette with zero typing in chat:
| Command |
Action |
Shastra: Scan Workspace |
Run the full project intelligence scan |
Shastra: Open Documentation |
Launch the interactive docs panel |
Shastra: Activate Agent |
Quick-pick any of the 21 specialist agents |
Shastra: Insert Workflow |
Quick-pick any slash command into chat |
Shastra: Show Skill Details |
Browse all 37 knowledge skills |
Shastra: Reset Session Memory |
Clear the self-eval context |
Shastra: Open Chat |
Open chat with @shastra pre-filled |
The AI Shastra icon in the VS Code activity bar opens three panels:
- Agents all 21 specialists listed; click any to activate in chat
- Skills 37 knowledge modules by category
- Workflows 12 slash commands with step breakdowns; click to run
Settings
{
// Default agent for open-ended questions (no slash command)
"shastra.defaultAgent": "orchestrator",
// Automatically scan the workspace when VS Code opens a folder
"shastra.autoScanOnOpen": false,
// Show self-evaluation hints after responses
"shastra.selfEvalEnabled": true
}
Requirements
- VS Code 1.85 or newer
- GitHub Copilot Chat (active subscription)
No other dependencies. No external packages. No initialisation step.
AI Shastra By Nitin Bhaiya
Install. Open chat. Type @shastra. Everything else is already there.
| |