Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
MarioNitzke avatar

Olympians

  • Updated April 6, 2026
  • MarioNitzke/claude-olympians

olympians is a Claude Code skill in the AI & Agent Building category. Pre-built agent team roles with interactive planning and spawning for Claude Code Agent Teams.

Key points

  • olympians
  • AI & Agent Building
  • AI-coding skill

Olympians by the numbers

  • Data as of Jul 7, 2026 (Skillselion catalog sync)
/plugin marketplace add MarioNitzke/claude-olympians
/plugin install olympians@olympians

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
Last updatedApril 6, 2026
RepositoryMarioNitzke/claude-olympians

What it does

Pre-built agent team roles with interactive planning and spawning for Claude Code Agent Teams.

README.md

Olympians

Olympians Banner

We used to write scripts and save code snippets. Today we write prompts and save agent roles. The craft has changed — the mindset hasn't. Build your toolkit, share what works, iterate on what doesn't.

A Claude Code plugin for planning, creating, and spawning Agent Teams.


What is Olympians?

Three skills:

  • /olympians:spawn-team — Spawn a team right now. Give it roles + task, it handles the rest.
  • /olympians:plan-team — Plan a complex team with phases, roles, and order. Calls spawn-team when done.
  • /olympians:create-olympian — Create your own custom role from a template.

8 generic roles included (architect, backend-developer, frontend-developer, qa-reviewer, db-specialist, researcher, devops, test-writer). Work with any stack.

8 example roles in examples/ — my personal roles for E2E testing, Stripe payments, chaos testing, multi-model research. Use them as inspiration to build your own.


Prerequisites

  • Claude Code
  • tmux or iterm2 (optional) — for split-pane visual monitoring of all teammates simultaneously
  • Example roles may require additional tools (Stripe CLI, agent-browser, Docker) — see each role's Requires field

Installation

# Local
claude --plugin-dir ./olympians

# Marketplace
/plugin marketplace add MarioNitzke/olympians
/plugin install olympians

Enable Agent Teams in ~/.claude.json (global) or .claude/settings.json (project):

{ "env": { "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1" } }

Example roles are opt-in — copy them to your project or to the plugin references:

# Project-local (this project only, overrides plugin defaults)
mkdir -p .claude/olympians
cp examples/*.md .claude/olympians/

# Global (all projects)
cp examples/*.md skills/spawn-team/references/

Recommended: Install tmux for split-pane mode

With tmux, each teammate gets its own terminal pane — you can see all agents working simultaneously. Without it, teammates run hidden and you navigate with Shift+Down.

# macOS
brew install tmux

# Then set in ~/.claude.json:
{ "teammateMode": "tmux" }

Also works with iTerm2 (it2 CLI + Python API enabled).


When to Use What

Spawn Team — quick, mid-work

You're working and need help now:

/olympians:spawn-team review auth changes with qa-reviewer
/olympians:spawn-team build payment feature with backend-developer and frontend-developer
/olympians:spawn-team investigate booking bug with researcher and backend-developer

It reads role references, infers phases, applies best practices, shows preview, launches. See spawn-team workflow diagram.

Plan Team — structured, complex tasks

Plan Team Workflow

You're tackling a big problem and want to plan properly (plan mode on):

/olympians:plan-team

Wizard: task → phases → AI proposes team table → you modify → Team Lead instructions → auto-calls spawn-team.

Create Olympian — build your own roles

/olympians:create-olympian

Shows examples for inspiration, reads your project structure, then walks you through: name → tools → model → system prompt → saved. See create-olympian workflow diagram.


Generic Roles (active)

Role Model Tools Isolation
architect opus all none
backend-developer opus all worktree
frontend-developer sonnet all worktree
qa-reviewer sonnet read-only none
db-specialist opus all worktree
researcher sonnet read-only none
devops sonnet all worktree
test-writer sonnet all worktree

Example Roles (in examples/, opt-in)

These are MY roles — they show what custom roles look like. Copy what fits, modify it, or build your own.

Role What it does
admin-tester E2E tests admin panel via agent-browser
customer-tester E2E tests customer flows (logged-in or anonymous)
api-tester Tests REST API with curl, monitors logs + DB
email-tester Tests outgoing emails via GreenMail/Docker
stripe-tester Tests payments via Stripe CLI
angry-customer Chaos tester — breaks the app, consults Gemini + Codex
plan-agent Autonomous planner — God's Plan without questions
ultra-researcher Multi-model research across Claude + Gemini + Codex + Reddit

Example: How ultra-researcher works

/olympians:spawn-team research SignalR vs WebSockets with ultra-researcher

The ultra-researcher launches 4 parallel research streams:

  1. Claude — WebSearch + WebFetch for articles and docs
  2. Reddit — mcp__reddit__search_reddit for real user experiences
  3. GeminiSkill("gemini:rescue") for Gemini's perspective
  4. CodexSkill("codex:rescue") for Codex's perspective

Then synthesizes: consensus (all agree) → majority (3/4) → disagreements → unique findings → Reddit reality check. Output is a structured report with confidence levels and citations.


Build Your Own Roles

The generic roles are a starting point. Custom roles tailored to YOUR project are what make agent teams powerful.

/olympians:create-olympian

Or copy an example and modify it:

# Save to project (recommended)
mkdir -p .claude/olympians
cp examples/stripe-tester.md .claude/olympians/my-tester.md

# Save to plugin (global)
cp examples/stripe-tester.md skills/spawn-team/references/my-tester.md

Every role is a markdown file with: name, tools, model, isolation, system prompt, spawn example. See any file in references/ or examples/ for the format.

Role Resolution Order

When spawning a team, roles are resolved in this order (first match wins):

  1. .claude/olympians/ — project-specific roles (can override plugin defaults)
  2. skills/spawn-team/references/ — plugin's built-in generic roles

This means you can customize any built-in role per project by placing a file with the same name in .claude/olympians/.


Best Practices

Spawn-team enforces best practices from official Claude Code docs, community research (Reddit, YouTube), and production experience. Full guide in best-practices.md. Key points:

  • Contract-first — foundation agents define schemas before implementation starts
  • File ownership — no two agents on the same file
  • Delegate mode (Shift+Tab) — lead coordinates, doesn't code
  • 3-5 teammates, 5-6 tasks each
  • Read-only reviewers — can't break code
  • Competing hypotheses — for debugging, spawn agents with different theories
  • Cross-model validation — different models for builder vs reviewer

Contributing

Share your custom roles! Submit a PR to examples/. We're building a community library of proven agent roles.

  1. Create your role → 2. Test it → 3. PR to examples/

See CONTRIBUTING.md for guidelines and role file format.


License

MIT

Related skills

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.