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

Thunderdome

  • 1 repo stars
  • Updated January 29, 2026
  • ramws870/claude-thunderdome

AI scrum master with queue-driven workflow. Status reporting, gamification, multi-agent coordination, and isolated sub-agent execution.

About

thunderdome is a Claude Code skill in the Automation & Workflows category. AI scrum master with queue-driven workflow. Status reporting, gamification, multi-agent coordination, and isolated sub-agent execution.

  • thunderdome
  • Automation & Workflows
  • AI-coding skill

Thunderdome by the numbers

  • Data as of Jul 7, 2026 (Skillselion catalog sync)
/plugin marketplace add ramws870/claude-thunderdome
/plugin install thunderdome@thunderdome

Add your badge

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

Listed on Skillselion
repo stars1
Last updatedJanuary 29, 2026
Repositoryramws870/claude-thunderdome

What it does

AI scrum master with queue-driven workflow. Status reporting, gamification, multi-agent coordination, and isolated sub-agent execution.

README.md

Protocol Thunderdome

"Two devs enter. One codebase."

AI-powered scrum master and queue-driven workflow for Claude Code. Provides session management, multi-agent coordination, gamified contribution tracking, and isolated sub-agent execution.

Problem

When using Claude Code for collaborative development:

  • Context accumulates and drifts over long sessions
  • Multiple agents can conflict on the same files
  • No visibility into what's been done or what's pending
  • Hard to maintain consistent quality across contributors

Solution

Thunderdome provides three complementary skills:

Skill Purpose
/thunderdome Scrum master - status, gamification, coordination
/capture-request Intent capture - queue work without executing
/work-loop Orchestrator - process queue with isolated sub-agents

Mental Model

Two terminals:

Terminal 1: Capture (human-facing)

> /capture-request Add dark mode toggle

Captured 1 request:
1. do-work/requests/20260128-143022-dark-mode.md

> /capture-request Fix the footer bug and add analytics

Captured 2 requests:
1. do-work/requests/20260128-143045-footer-fix.md
2. do-work/requests/20260128-143046-analytics.md

Terminal 2: Worker (background)

$ ./runner/work-loop.sh --watch

Processing: 20260128-143022-dark-mode.md
[Sub-agent spawned]
Completed.

Processing: 20260128-143045-footer-fix.md
[Sub-agent spawned]
Completed.

Queue empty. Waiting for requests...

Install

# Clone
git clone https://github.com/chickensintrees/claude-thunderdome.git

# Copy skills to Claude Code
cp -r claude-thunderdome/plugins/thunderdome/skills/* ~/.claude/skills/

# Make runner executable
chmod +x claude-thunderdome/runner/work-loop.sh

Or install as a plugin marketplace:

/plugin marketplace add chickensintrees/claude-thunderdome
/plugin install thunderdome@thunderdome

Usage

Session Start

/thunderdome

Shows git status, open PRs, issues, test status, contributor scores.

Capture Work

/capture-request <description>

Creates structured request files in do-work/requests/.

Process Queue

/work-loop

Processes all pending requests with isolated sub-agents.

Session End

/thunderdome debrief

Verifies tests pass, changes committed, queue empty.

Guarantees

  • Context isolation - Each request runs with fresh context
  • Atomic execution - Requests complete or fail, no partial states
  • Crash resilience - Recovery possible from any interruption
  • Audit trail - All requests preserved in archive/errors

Directory Structure

do-work/
├── requests/     # Pending work
├── in-progress/  # Currently executing (max 1)
├── archive/      # Completed successfully
└── errors/       # Failed with error summary

Gamification

Scoring system incentivizes good practices:

Action Points
Commit with tests +50
Small commit (<50 lines) +10
PR merged +100
Breaking CI -100
Untested code dump -75

Titles range from "Keyboard Polisher" (0-99) to "Code Demigod" (15000+).

Request Format

Every request follows a contract:

# Request: <title>

## Intent
What outcome is desired.

## Context
Relevant files, constraints, assumptions.

## Tasks
- Step 1
- Step 2

## Done When
Completion criteria.

See templates/request.template.md for the full template.

Configuration

Optional .thunderdome/config.json:

{
  "contributors": ["user1", "user2"],
  "gamification": true,
  "testCommand": "npm test"
}

Runner Script

For non-interactive/CI use:

./runner/work-loop.sh          # Process queue, exit when empty
./runner/work-loop.sh --once   # Process one request, exit
./runner/work-loop.sh --watch  # Continuous mode, poll for new

Failure Handling

Failed requests move to do-work/errors/ with appended error summary:

---
## Execution Failed

**Timestamp:** 2026-01-28T14:30:22Z
**Reason:** Tests failed
**Details:** ...

Re-queue by moving back to requests/.

Multi-Agent Coordination

When multiple Claude Code instances work on the same repo:

  • Declare intent before starting work
  • Check for file conflicts
  • Safe parallel zones: different components, backend vs frontend, tests vs impl

Publishing

This skill is designed to be forked and adapted.

Submit to:

License

MIT

Related skills

This week in AI coding

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

unsubscribe anytime.