Skip to content

xiapeli/elis

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ELIS — Multi-Agent Orchestrator

ELIS

Multi-agent orchestrator for Claude Code — 10+ agents across 6 layers, from idea to deployed product

MIT License Claude Code Agents Layers

Named after Elis Regina (1945-1982), Brazil's greatest singer. Like a maestro conducting an orchestra, ELIS orchestrates specialized AI agents to create world-class web applications.

Install · Architecture · Commands · PRD Loop · Contributing


What It Does

One prompt. Six layers. Ten+ agents. Full-stack delivery.

Describe your product idea. ELIS breaks it into phases and delegates across specialized agents — from market research through design, build, test, deploy, and monetization.

/elis full "A SaaS landing page for an AI writing tool with waitlist and Stripe billing"

ELIS runs Scout for market research, Designer + UXer in parallel for the visual system, Builder for production code, Tester for quality gates, Deployer for Vercel, and Stripe for payments. You get a deployed, monetized product.

The fundamental rule: ELIS never writes code directly. It plans, delegates, synthesizes, and enforces quality gates. Every agent has a structured contract ("partitura") defining inputs, outputs, and pass/fail criteria.

The 6 Layers

                          ┌──────────────────┐
                          │      ELIS        │
                          │   Conductor      │
                          │     (Opus)       │
                          └────────┬─────────┘
                                   │
    ┌──────────┬──────────┬────────┴────────┬──────────┬──────────┐
    ▼          ▼          ▼                 ▼          ▼          ▼
 DISCOVERY   DESIGN     BUILD           BACKEND    DEPLOY    BUSINESS

Layer 1: Discovery

Agent Model Purpose
Scout Haiku Market research, competitor analysis, viability scoring, tech stack recommendation
Analyzer Haiku Technical deep-dive, architecture assessment

Layer 2: Design

Agent Model Purpose
Designer Sonnet Design systems — HSL colors, typography scales, spacing, effects, animations
UXer Sonnet Nielsen heuristics (scored), WCAG audit, AIDA journey, conversion analysis
Animator Sonnet Framer Motion, CSS animations, micro-interactions, reduced-motion fallbacks

Layer 3: Build

Agent Model Purpose
Builder Sonnet Production code — HTML, Next.js, or Astro. Responsive, semantic, accessible
Tester Sonnet Playwright E2E, visual regression, accessibility audit, performance checks

Layer 4: Backend

Agent Model Purpose
Supabase Sonnet Schema design, Row Level Security policies, Edge Functions, auth, storage

Layer 5: Deploy

Agent Model Purpose
Deployer Sonnet Vercel config, GitHub Actions, environment variables, domain setup, preview deploys

Layer 6: Business

Agent Model Purpose
Stripe Sonnet Checkout sessions, subscriptions, webhook signature verification, Customer Portal
Launcher Sonnet SEO meta/OG tags, sitemap, robots.txt, analytics, launch checklist

Commands

Command What It Runs
/elis full <description> All 6 layers end-to-end
/elis analyze <url> Scout + Analyzer + Designer synthesis
/elis design <description> Scout, Designer + UXer in parallel, approval gate
/elis build <spec> Builder, Animator, Tester, quality gate
/elis deploy Deployer + Launcher
/elis monetize Stripe integration + testing

Options

--reference <url>       Reference site for Scout/Designer
--stack <html|next|astro>  Output framework (default: html)
--output <path>         Output directory
--with-backend          Activate Supabase layer
--with-payments         Activate Stripe layer
--style <minimal|bold|playful|professional>

Contract System

Every agent operates under a structured contract ("partitura"):

{
  "model": "sonnet",
  "input_contract": {
    "required": ["project_description", "target_audience"],
    "optional": ["reference_url", "style_preference"]
  },
  "output_contract": {
    "format": "json",
    "schema": { "design_tokens": {}, "color_palette": [], "typography": {} }
  },
  "quality_gates": ["wcag_aa_contrast", "mobile_responsive", "lighthouse_90"],
  "timeout_ms": 120000,
  "max_retries": 3
}

When quality gates fail: retry (up to 3x) → escalate with context → request human input → fallback → graceful degradation.

Quick Start

Install

git clone https://github.com/xiapeli/elis.git
cd elis

# Copy the skill (won't overwrite existing files)
mkdir -p ~/.claude/commands
cp -n elis.md ~/.claude/commands/

Use

# Open Claude Code and run:
/elis full "A portfolio site for a photographer with dark theme and contact form"

# Analyze an existing site:
/elis analyze https://example.com

# Design only:
/elis design "A meditation app landing page" --style minimal

Autonomous Mode (PRD Loop)

ELIS includes a PRD-driven autonomous loop for hands-off development. Define user stories in prd.json, and ELIS implements them one by one:

cd prd-loop
cp prd.json.example prd.json  # edit with your stories
./elis-loop.sh 10             # run up to 10 iterations

Each iteration reads the PRD, implements the next story, commits, and logs progress. Stops when all stories pass or max iterations reached.

Note: The loop uses --dangerously-skip-permissions for autonomous operation. Review prompt.md before running.

Design Standards

ELIS ships with opinionated 2026 design defaults:

  • Gradient orbs — 80-150px blur, 20-30% opacity
  • Bento grids — asymmetric layouts, 16-24px gaps
  • Expressive typography — 48-96px headlines, serif+sans pairing
  • Glass morphism — 8-16px blur, white/5-10%
  • Noise textures — SVG-based, 3-8% opacity
  • Micro-interactions — lift on hover, scale on press, glow on focus
  • Mandatory prefers-reduced-motion fallback on all animations

See docs/customization.md to override defaults.

Documentation

Requirements

  • Claude Code with API access
  • Claude Opus for the orchestrator
  • Claude Sonnet/Haiku for sub-agents

Contributing

Contributions welcome. Ideas:

  • New agents (Analytics, Copywriter, SEO specialist)
  • Design standard presets (brutalist, corporate, playful)
  • Example PRDs for common project types
  • Translations for agent prompts
  • Integration tests for the contract system

Open an issue or submit a PR. Follow conventional commits (feat:, fix:, docs:).

License

MIT — Phelipe Xavier, 2026

About

Multi-Agent Orchestrator for Claude Code — 10+ specialized agents, 6 execution layers, contract-based quality gates. Named after Elis Regina.

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages