Rig D Agent

Rig D Agent is a live 1v1 web MVP where players coach flawed AI stick-fighter agents through a debate duel and try to force an upset. The point is not drafting the strongest bot. The point is making a weaker bot outperform expectation.

Value Prop

Rig D Agent turns AI usage into a competitive game:

• Players prove skill by coaching, not by owning the best model.
• Weak-agent wins create the clip-worthy moment.
• The reveal makes overperformance legible through Rig Score.

Current MVP

This repository currently ships a playable prototype with:

• Next.js web app
• private room creation and invite-code join flow
• anonymous cookie-based player sessions
• one synchronous mode: Debate Duel
• four archetypes: Bruiser, Gremlin, Scholar, Showman
• 60-second structured coaching phase
• three debate turns per side: opening, rebuttal, closing
• automated judging and reveal screen
• Rig Score and rubric-style scoring breakdown
• rematch support
• deterministic local arena engine by default
• optional OpenAI-backed generation and judging when configured

Tech Notes

• Frontend and API are built with Next.js.
• State is currently stored in-process in memory for zero-infra local play.
• Match progress is updated by client polling rather than a realtime transport layer.
• The current API and domain model are structured so Postgres, Supabase Realtime, and a dedicated worker can replace the in-memory prototype later.

Getting Started

Requirements

• Node.js 24+
• npm 11+

Install

npm install

Run the app

npm run dev

Open http://localhost:3000.

Run tests

npm test

Production build

npm run build

Optional OpenAI Mode

If no API key is present, the app uses the built-in local arena engine so the MVP stays fully playable.

To enable server-side OpenAI generation and judging, create .env.local:

OPENAI_API_KEY=your_key
OPENAI_MODEL=gpt-4.1-mini

Notes:

• Keys are server-side only.
• Player-supplied API keys are not part of the current MVP.
• The local mode remains the default fallback if OpenAI is not configured or unavailable.

Project Structure

Key areas in the current codebase:

• app: routes, pages, API handlers
• components: client UI and visual components
• lib: game content, state engine, validation, session helpers
• test: MVP logic tests

Roadmap Direction

The current repo is intentionally narrow. Likely next steps are:

• move state from memory to Postgres
• replace polling with realtime room updates
• add persistence for replays and match history
• harden the judge path and retries
• add richer matchmaking and moderation primitives

Contributing

Contributions are welcome.

If you want to contribute:

1. Open an issue for substantial feature work or architecture changes before implementing them.
2. Keep pull requests focused and small enough to review cleanly.
3. Preserve the core product rule: player coaching must matter more than raw model strength.
4. Add or update tests when changing the match engine, scoring, or API behavior.
5. Verify the project with:

npm test
npm run build

Contribution guidelines for this stage:

• Prefer incremental improvements over broad rewrites.
• Do not introduce paid services or infrastructure assumptions into the default local flow.
• Keep the mock arena engine working even when adding provider-backed features.

License

This project is licensed under the MIT License. That includes commercial use, private use, modification, and distribution. See LICENSE.