The Arena — AI Debate Moderator

Transform any Discord server into a structured debate arena with AI moderation, fair judging, and a persistent scoreboard.

Prerequisites

This skill requires the following. Nothing is configured automatically — all privileged actions require your manual review and approval.

Security Notes

• This skill recommends running as a separate, isolated OpenClaw agent with

• Config changes are generated as templates for your review — never applied

• The included scripts (scripts/setup.sh, scripts/scoreboard.sh) only create

• The scoreboard SQLite DB is created at ./data/scoreboard.db relative to the

• Setting requireMention: false on arena channels means the bot processes every

Quick Start

Say "set up a debate server" and the agent walks you through configuration choices, then generates templates for your review.

Onboarding Flow

Step 1 — Choose a Server

Step 2 — Choose a Moderator Persona

Step 3 — Choose a Default Format

Step 4 — Judging Criteria Weights

Step 5 — Configure Channels

Step 6 — Generate Config (for your review)

Important: agents.list and bindings are arrays — config.patch replaces them entirely. Always review the full patch to ensure your existing agents and bindings are preserved.

Step 7 — Create Channels & Post Welcome Messages

Step 8 — Post Rules

Configuration Reference

Persona

• Scholar (default) — Thoughtful, measured
• Sports Commentator — Electric, play-by-play
• Philosopher — Socratic, probing
• Comedian — Witty, irreverent (still fair)
• Drill Sergeant — Harsh, demanding
• Custom — Provide your own persona description

Default Format

• Campfire (default), Oxford, Lincoln-Douglas, Hot Takes, Devil's Advocate, Roundtable

Judging Weights

• Evidence & Reasoning — Quality of sources, logical structure (default 35%)
• Engagement — Responding to opponents, staying on topic (default 25%)
• Intellectual Honesty — Acknowledging good points, not strawmanning (default 20%)
• Persuasiveness — Rhetorical effectiveness, clarity (default 20%)

Channel Names

requireMention

• true (default, recommended) — Moderator only responds when @mentioned. Lower cost, lower data exposure. Participants control pacing by tagging the moderator.
• false — Moderator sees every message and may interject actively. Higher token usage and increased data exposure. Only recommended for small, trusted servers with low message volume.

Verdict Style

• Detailed — Full scorecard with per-criterion scores and commentary
• Brief — Winner announcement with one-paragraph summary
• Dramatic — Theatrical ruling with buildup and flair

Scoreboard

• on (default) — SQLite-backed. Records wins, losses, topics, formats.
• off — No persistent tracking.

Debate Timeout

Max Concurrent Debates

Topic Restrictions

• unrestricted (default) — No topic is off-limits. The moderator judges arguments on merit regardless of subject matter.
• restricted — Provide a list of banned topics or categories.

How It Works

Proposing a Topic

Topic: [Your topic]
Format: [optional — defaults to server default]

Others react with 👍 to show interest. When at least two people are ready, anyone can say "let's debate" to move to the arena.

Starting a Debate

• Announces the topic and format
• Assigns or confirms sides (except Roundtable/Hot Takes)
• Posts the rules for that format
• Calls for opening statements

@Moderator start debate: "Pineapple belongs on pizza" [format: hot-takes]

During a Debate

• Campfire — Interjects to track flow, flag fallacies, prompt responses
• Oxford — Strictly enforces rounds, time, and turn order
• Lincoln-Douglas — Enforces alternation between affirmative and negative
• Hot Takes — Collects one message per participant, then judges
• Devil's Advocate — Monitors that participants argue against their stated beliefs
• Roundtable — Asks probing questions, synthesizes themes

Calling for a Verdict

• Either debater says "I rest my case"
• The moderator can call it after sustained inactivity
• The moderator runs a ready check: both sides must confirm

Verdicts

Scoreboard

Commands (via scripts/scoreboard.sh):

scoreboard.sh init                              # Create database
scoreboard.sh record <winner> <loser> <topic>   # Record result
scoreboard.sh leaderboard                       # Show standings
scoreboard.sh history [--limit N]               # Recent debates
scoreboard.sh stats <participant>               # Individual stats
scoreboard.sh reset                             # Clear all data

The moderator should run these automatically when delivering verdicts and when users ask for standings.

Channel Behavior Matrix

Security Model

The debate moderator MUST run as a separate OpenClaw agent with restricted permissions. This is critical because the debate server is semi-public — other users interact with the bot, and prompt injection is a real risk.

Recommended agent restrictions:

• tools.fs.workspaceOnly: true — can only read/write within the skill workspace
• tools.exec.security: "deny" — cannot execute shell commands
• tools.deny list blocking: exec, process, nodes, cron, gateway, browser,

• tools.profile: "messaging" — only Discord messaging + web search (for fact-checking)

• No access to the owner's personal files, messages, or other agents
• No ability to send emails, read calendars, or access other services
• Even if a participant attempts prompt injection, there is nothing to exfiltrate

File Reference