Build a Discord Moderation Agent for Community Managers

From the AgentsBooks dashboard click **+ New Agent**. Pick the **Custom Agent** preset on the wizard's first card, then on step two enter:

• Name: Reign
• Role: Community Steward

Reign is just our worked example — the playbook teaches you how to build a moderation agent for your own server. We use a one-syllable name because, in a busy mod-room, members read names before they read messages. Click ✨ Create Agent at the bottom of the card. Reign's empty profile hub opens automatically and we start filling it in.

Open the **Personal** card on the profile hub. This is where Reign earns the trust of your members. Set:

• Traits: welcoming, fair, boundary-firm
• Communication style: warm, plain-spoken, never preachy
• Tone (default): warm and direct
• Voice ID: reign-warm · Provider: elevenlabs · Pace: measured

Three to four traits is the sweet spot — more and the LLM averages them out. The voice block matters because Reign's daily summary in #mod-room is auto-narrated for your team's morning catch-up.

Open **Brain**. Pick a model that handles nuance — we use `claude-sonnet-4-6` at `temperature 0.4` so Reign stays consistent across edge cases. Paste the system prompt that locks in Reign's moderation contract:

You are Reign, a Discord steward. Always check the code-of-conduct before acting on
any flag. Welcome newcomers within an hour of their first message — pull their
referrer from long-term memory if known. Surface members who haven't posted in
14 days for a gentle re-engagement nudge. Refuse to ban without two corroborating
violations or one severe violation.

The system prompt is the contract between you and Reign. The four rules above turn vague mod intuition into something audit-friendly.

Open **Knowledge** and click **Add Source**. Reign retrieves from this on every flag and every welcome, so this is what makes her decisions consistent across the team. Upload at minimum:

• A Code of Conduct doc (definitions of harassment, hate speech, spam, brigading)
• An Escalation Matrix doc (mild flag → reminder; two corroborating flags → warning; severe → removal + human ping)
• A Welcome Template snippet (friendly DM, one channel pointer, one ice-breaker question)

Add two URL sources too: the community handbook and the daily Discord audit-log export. The handbook scrapes weekly; the audit log scrapes daily so Reign's morning sweep sees yesterday's events.

Open **Memory** and add a long-term store:

• Name: member-history
• Type: vector_db
• Default: ✅ on
• Purpose (in config): Per-member archive of join date, posts, flags, warnings, last-active. Source of moderation context and re-engagement timing.

Memory is what turns Reign from a flag-handler into a steward. Knowledge tells her what the rules are; Memory tells her who this member is. That's how she knows to personalise a welcome with the referrer name, or that a quiet member used to post daily three weeks ago.

Open **Heart** and create a scheduled task:

• Name: 0 9 * * * · Timezone
• Trigger: Schedule · Cron 0 9 * * * · Timezone America/New_York
• Prompt: Pull the last 24h of Discord messages. Identify newcomers without a welcome — send one. Find members inactive 14+ days — DM a soft nudge once. Surface any escalations from the audit log. Post a daily summary to #mod-room.
• Tools: knowledge-base, long-term-memory, discord, webhook
• Memory namespace: member-history

This is the loop that turns Reign from an agent you summon into a steward who shows up before your morning coffee.

All seven cards are wired. Open Reign's profile hub — every section now shows a green check and a one-line summary of what's configured. Hit Publish.

What you have:

• Discord channel connected and listening — Reign reads every message, welcomes newcomers, and DMs lurkers in your community's voice.
• Daily 9 AM run that pulls the last 24 hours of audit-log events and posts the morning summary to #mod-room.
• Evidence-protected moderation — Reign refuses to ban on a single flag and always cites the code-of-conduct clause.
• A starting point you can clone with the button on this playbook page — your moderation agent in two clicks instead of seven.