Skip to content

ludovicschmetz-stack/olympus

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

⚡ Olympus

A divine council of AI agents reviewing every line of your code.

Olympus is a ready-to-use methodology kit for Claude Code that brings structured, multi-agent code review to your vibe coding workflow. Inspired by Bernard Lambeau's approach to building the Elo programming language with Claude Code.

No framework. No dependencies. No code to maintain. Just battle-tested prompts, a proven workflow, and 4 divine reviewers.


🏛️ The Pantheon

Agent Domain Role
Athena Security Detects vulnerabilities, injection flaws, exposed secrets, broken auth
Hermes Domain-Driven Design Guards ubiquitous language, bounded contexts, domain model integrity
Cassandra Edge Cases & Bugs Foresees production failures, missing error paths, race conditions
Apollo Simplicity & Clarity Fights complexity, over-engineering, and violations of SOLID/DRY/KISS

Each agent runs in its own isolated context window, with read-only tool access — they analyze without modifying your code.


🚀 Quick Start

1. Clone into your project

# Option A: Clone as a new project
git clone https://github.com/ludovicschmetz-stack/olympus.git my-project
cd my-project
./init.sh --template data-platform --name "My Project"

# Option B: Add to an existing project
git clone https://github.com/ludovicschmetz-stack/olympus.git /tmp/olympus
cp -r /tmp/olympus/.claude /tmp/olympus/tasks /tmp/olympus/CLAUDE.md .
rm -rf /tmp/olympus

2. Customize CLAUDE.md

Edit CLAUDE.md at your project root to describe your architecture, stack, and conventions. Use one of the provided templates in templates/ as a starting point.

3. Start working with Claude Code

claude
> Read CLAUDE.md and available agents.
> Then pick the first task in tasks/to-do/ and implement it.

Claude Code will automatically:

  • Load your project context from CLAUDE.md
  • Discover agents in .claude/agents/
  • Invoke the Pantheon for review after each implementation
  • Report findings with severity and fix suggestions

📂 Project Structure

olympus/
├── CLAUDE.md                          # Project context (customize this)
├── .claude/
│   ├── agents/
│   │   ├── athena.md                  # 🛡️ Security auditor
│   │   ├── hermes.md                  # 📝 DDD guardian
│   │   ├── cassandra.md               # 🔮 Edge case prophet
│   │   ├── apollo.md                  # ☀️ Simplicity arbiter
│   │   └── council.md                 # ⚡ Orchestrator (runs all 4)
│   └── safe-setup/
│       ├── Dockerfile                 # Alpine sandbox for autonomous mode
│       └── docker-compose.yml
├── tasks/
│   ├── to-do/
│   │   └── _template.md              # Task template
│   ├── in-progress/
│   ├── hold-on/
│   ├── analyze/
│   └── done/
├── templates/                         # Pre-built CLAUDE.md templates
│   ├── data-platform.md
│   ├── web-api.md
│   ├── dbt-project.md
│   └── generic.md
├── init.sh                            # Project scaffolding script
├── README.md
└── LICENSE

🎮 Three Modes of Operation

Following Bernard Lambeau's proven methodology:

Mode When How
Accept-it Simple, low-risk changes Claude proposes, you approve, agents review after
Plan mode Complex features, architecture decisions Claude asks design questions first, you agree on approach, then implement
Autonomous Batch operations, refactoring, tests Use --dangerously-skip-permissions inside Docker sandbox only

📋 Kanban Workflow

Tasks are managed as markdown files in the tasks/ directory:

tasks/
├── to-do/          → Ready to start
├── in-progress/    → Currently being worked on (max 1 at a time)
├── hold-on/        → Blocked, needs human decision
├── analyze/        → Needs investigation before implementation
└── done/           → Completed, reviewed by the Pantheon

Task Lifecycle

  1. Create task from template: cp tasks/to-do/_template.md tasks/to-do/003-my-task.md
  2. Claude picks up the task, moves it to in-progress/
  3. Claude implements the task
  4. The Pantheon reviews (all 4 agents)
  5. Fix all CRITICAL findings
  6. Address or explicitly accept WARNING findings
  7. Move to done/ with review summary appended

🔧 Customization

Adding Your Own Agents

Create a new .md file in .claude/agents/:

---
name: your-agent-name
description: When this agent should be invoked
tools: Read, Grep, Glob
model: sonnet
---

You are a [role description]...

## Scope
...

## Output format
...

Domain-Specific Templates

Use templates/ to pre-configure CLAUDE.md for common project types. Run:

./init.sh --template data-platform --name "My Data Project"

Available templates: data-platform, web-api, dbt-project, generic


⚔️ Olympus vs Anthropic's Claude Code Review

On March 9, 2026, Anthropic launched Code Review — a managed, multi-agent PR review service built into Claude Code. Here's how it compares to Olympus.

Olympus Claude Code Review
What it is Open-source methodology kit (prompts + workflow) Managed SaaS service
Cost per review $0 (uses your existing Claude Code subscription) $15–25 per PR
Access Any Claude Code user Team & Enterprise plans only
Integration Manual invocation in Claude Code Auto-triggers on GitHub PRs
Review axes 4 specialized: Security, DDD, Edge Cases, Complexity Logic errors + light security
DDD / Domain review Yes (Hermes agent) No
Complexity metrics Yes (Apollo agent, cyclomatic complexity, nesting) No dedicated axis
Scoring Deterministic 0–100 with explicit formula Color-coded severity (red/yellow/purple)
Customization Full — edit any agent prompt, fork, extend REVIEW.md + CLAUDE.md
Transparency 100% — every prompt is readable and versioned Black box
Workflow Kanban + 3 interaction modes + Docker sandbox PR review only
Setup Clone + init.sh (5 min) Admin toggle per repository

When to use which

Use Claude Code Review if you're an enterprise team generating high volumes of PRs and want zero-config, automated review on every push. You need GitHub-native integration and don't mind paying per review.

Use Olympus if you want full control over what gets reviewed and how, need DDD and complexity analysis, work as a freelancer or small team, or want a complete development methodology — not just a review tool. Olympus agents are educational assets: reading athena.md teaches OWASP, reading hermes.md teaches DDD.

Use both by aligning your CLAUDE.md and REVIEW.md across both tools. Claude Code Review catches logic errors on every PR automatically; Olympus provides deeper, structured review on milestone deliverables with a quality score you can share with stakeholders.


🏗️ For Freelancers

Olympus is designed to be part of your professional delivery:

"Every deliverable passes through a multi-agent review covering security (OWASP), domain consistency (DDD), edge cases, and complexity analysis. Clients receive a quality report with each delivery."

Include the Pantheon review summary in your delivery documentation to demonstrate engineering rigor.


📖 References


License

MIT — See LICENSE

About

A divine council of AI agents reviewing every line of your code. 4 specialized reviewers + Kanban workflow for Claude Code.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors