the-arena（竞技场）

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

restricted tool access (fs.workspaceOnly: true, exec.security: "deny").

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

automatically. You must manually review and apply all gateway config patches.

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

files within the skill workspace. They make no network calls and write no files outside the skill directory. Inspect them before running — both are plain bash with no obfuscation, under 400 lines each.

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

skill directory (or at $DEBATE_SCOREBOARD_DB if set). It never writes outside the skill workspace.

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

message in that channel. This increases token usage and data exposure. Use requireMention: true for lower cost and reduced visibility.

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

Create a new Discord server or pick an existing one. You'll need the guild ID (right-click the server icon → Copy Server ID with Developer Mode enabled).

Step 2 — Choose a Moderator Persona

Pick the voice your moderator uses during debates. Default: Scholar.

Details in references/personas.md.

Step 3 — Choose a Default Format

The default format used when someone starts a debate without specifying one. Default: Campfire.

Details in references/formats.md.

Step 4 — Judging Criteria Weights

Customize how the moderator scores arguments. Defaults:

Weights must sum to 100%. Details in references/judging.md.

Step 5 — Configure Channels

Default channel names (all customizable):

Step 6 — Generate Config (for your review)

The agent generates config snippets and an AGENTS.md tailored to your choices. You must review and manually apply all gateway config changes. The agent does not apply config patches automatically. See references/setup-guide.md for the full config template and a step-by-step walkthrough.

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

After you create channels in Discord, the agent can post welcome messages from assets/welcome-messages.md — or you can copy-paste them manually.

Step 8 — Post Rules

The agent posts the full rules document in #rules.

Configuration Reference

Persona

Controls the moderator's voice and commentary style.

• 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

The format used when a debate is started without specifying one.

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

Judging Weights

Customize the four scoring criteria. Must sum to 100%.

• 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

All five channels can be renamed. Provide a mapping during setup.

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

How the moderator delivers the final ruling.

• 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.

CLI: scripts/scoreboard.sh. DB location: $DEBATE_SCOREBOARD_DB or ./data/scoreboard.db.

Debate Timeout

Hours before the moderator flags a stale debate. Default: 48.

Max Concurrent Debates

Maximum simultaneous debates in the arena. Default: 3.

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

In #propose-a-topic, post:

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

In #the-arena, the moderator:

1. Announces the topic and format
2. Assigns or confirms sides (except Roundtable/Hot Takes)
3. Posts the rules for that format
4. Calls for opening statements

Participants can also start directly in the arena:

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

During a Debate

The moderator's behavior depends on the format:

• 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

The moderator flags logical fallacies, tracks participation balance, and keeps the debate moving. In mention-only mode, participants @mention for moderator input.

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

The moderator evaluates using the configured judging weights and delivers the verdict in the configured style. The verdict is posted in both #the-arena and #hall-of-records.

Scoreboard

When the scoreboard is enabled, results are automatically recorded.

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,

canvas, sessions_*, subagents, memory_search, memory_get, tts, image

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

What this ensures:

• 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

All config changes are your responsibility. The agent generates templates; you review and apply them. See references/setup-guide.md for the full security configuration and a tested tool deny list.

File Reference