discord-bot

name	discord-bot
description	Drive Mercury's brain-pipeline via a Discord bot. Use when the user wants to drop a video into a Discord channel and get a 3D cortical activation map plus three-tier narration as a reply, or any other "agent-as-bot" Discord workflow. Composes Mercury's existing gateway/discord with brain-viz / fmri-overlay / cortex-bridge.
version	0.1.0
author	Mercury
license	MIT
metadata	{"mercury":{"tags":["discord","bot","gateway","brain-viz","neuroscience","demo"],"category":"platforms","related_skills":["brain-viz","fmri-overlay","cortex-bridge","tailnet"]}}
prerequisites	{"files":["D:/mercury/gateway/discord"],"env":["DISCORD_BOT_TOKEN","DISCORD_GUILD_ID"],"python_packages":["cortex","discord.py>=2.5"]}

Discord Bot — Brain Pipeline as Slash Commands

Mercury already has a generic Discord gateway in gateway/discord/. This skill is the knowledge document for using that gateway specifically for the brain-response pipeline (the Jemma flow from D:/TRIBEV2/bot/).

The TRIBEV2 standalone bot will be retired in favor of Mercury's gateway — one process serves all platforms (Discord, WhatsApp, Telegram, Slack, etc.) instead of a separate bot per surface. This skill documents the user-facing slash commands the agent should expose.

When to Use

• User says "make a Discord bot that..." — first reach for Mercury's gateway, not a new bot
• User asks "can I drop a video in Discord and get the brain scan back?"
• User asks "how do I demo Jemma over Discord?" — yes, that's now Mercury
• User wants a slash command for any composed brain-pipeline flow

Do NOT use to build a generic Discord bot — that's just Mercury's gateway. This skill is specifically the brain-pipeline / hackathon flow.

Slash Commands Exposed

Command	Args	What it does
/brain	<attachment: video/audio>	Full pipeline: media_gate → brain_scan → narrate → reply with 3D viewer URL + toddler-tier narration in chat. The marquee command.
/scan	<attachment>	Same as /brain but no narration — fastest path to the visual
/explain	<scan_id> <tier>	Re-narrate an existing scan at a different tier (toddler / clinician / researcher)
/cortex-state	—	Returns current GPU state from cortex_bridge.cortex_state() — useful for queue debugging
/recent	—	Lists the user's last 5 scans with viewer URLs (per-user via Tailscale-User-Login derived from Discord user ID)

All commands are gated on:

1. The user having posted in the configured guild (DISCORD_GUILD_ID)
2. Cortex being reachable (cortex_state() != "unavailable")
3. The attachment being ≤ 50 seconds (per TRIBE v2's duration_trs lock)

How It Works

                Discord user posts /brain + attachment
                            │
                            ▼
            Mercury gateway (gateway/discord) receives the slash command
                            │
                            ▼ tool call
                       brain-viz skill
                            │ ↓ calls
                ┌───────────┼─────────────┐
                ▼           ▼             ▼
       cortex-bridge     three-js-      cortex-bridge
       .media_gate       component      .narrate
       .brain_scan        (existing      (×3 tiers)
                          webapp URL)
                            │
                            ▼
            Mercury gateway formats reply:
              • Embed with viewer URL
              • Toddler-tier text in chat
              • Buttons: "Clinician" / "Researcher" → /explain
                            │
                            ▼
                Discord user clicks viewer → opens
                https://brain.redteamkitchen.com/scan/abc123
                (Cloudflare-Tunnel-routed back to RTX 5090)

Step-by-Step Setup

Step 1 — Configure the gateway

# ~/.mercury/config.yaml
gateways:
  discord:
    bot_token_env: DISCORD_BOT_TOKEN
    guild_ids:    ["1234567890"]                  # optional restriction
    skills:       ["brain-viz", "fmri-overlay"]   # which skills the bot exposes

Step 2 — Start the gateway

mercury gateway up discord
# or as a long-running daemon
mercury gateway run --background discord

Step 3 — Slash-command registration

The first time the bot connects, it auto-syncs slash commands to the configured guild (instant) or globally (~1 hour propagation).

Step 4 — User flow (from the user's POV)

1. User joins the configured Discord guild
2. User types /brain and attaches a 20-second clip
3. Bot replies with "Scanning… (this takes 4–7 min)" and a progress indicator (the gateway updates the message via interaction.followup.edit)
4. On completion, the message is edited to include:
   • The viewer URL
   • Toddler-tier narration (≤50 words, in the chat directly)
   • Two buttons: "Clinician detail" and "Researcher detail"

Migration from D:/TRIBEV2/bot/

The old standalone Jemma Discord bot (D:/TRIBEV2/bot/bot.py) had its own RBAC, rate-limiting, and analysis-thread features. Migration map:

TRIBEV2/bot/ thing	Where it lives now
bot.py (Discord client)	gateway/discord/ (Mercury's existing gateway)
cat_gate.py (content gate)	cortex.media_gate.py (called via cortex-bridge.media_gate)
gemma.py (narration)	cortex-bridge.narrate
pipeline.py (TRIBE inference)	cortex.gpu_scheduler + cortex_bridge.brain_scan
RBAC	Mercury's Tailscale-User-Login identity middleware (Discord user → Mercury identity via pairing skill)
Rate-limiting	mercury cron schedules + per-user request queue (built-in)
Analysis threads	Discord's native threads — gateway creates one per /brain invocation

The retired bot stays in D:/TRIBEV2/bot/ as reference until the migration ships. Do not run both bots against the same Discord token — they'll fight for slash-command registration.

Hackathon Submission Notes (Nous Creative 2026)

For the demo recording, Discord is the delivery surface of choice:

1. The judge's POV is "user-friendly" — Discord is universally familiar, slash commands are obvious, the bot's reply showcases the multi-modal output beautifully
2. The recording captures the agent doing things across surfaces — the slash-command flow shows Mercury Agent's gateway discipline (one process, many platforms)
3. Kimi K2.6 does the planning step inside the gateway (visible in the bot's typing indicator: "thinking with kimi-k2.6…") — qualifies for the Kimi Track $5K bonus

Trademark line: "Gemma is a trademark of Google LLC." — include in the bot's /about slash-command reply.