feat: Discord integration + multi-channel setup wizard

[unforsook]

Summary

Builds on top of the TypeScript conversion (#4) to add full Discord DM integration and an interactive setup wizard that lets users configure their preferred messaging channel and Claude model on first run.

• New Discord client (src/discord-client.ts) — a Discord.js bot that listens for DMs, queues messages through the shared queue-processor, handles typing indicators, respects Discord's 2000-char limit with smart message splitting, and supports !reset//reset to clear conversation history
• Configurable model selection — queue-processor reads from .tinyclaw/model config to switch between Sonnet (fast) and Opus (smartest) at runtime
• Interactive first-run setup — on first ./tinyclaw.sh start, users are prompted to choose their channel (Discord / WhatsApp / Both) and model preference, persisted to .tinyclaw/channel and .tinyclaw/model
• Dynamic tmux layouts — the daemon adapts its pane layout based on active channels: 4-pane grid for single channel, 5-pane layout when running both simultaneously
• ./tinyclaw.sh setup — reconfigure channel and model at any time without editing files
• Discord token validation — startup checks .env for DISCORD_BOT_TOKEN and fails fast with a clear error if missing

How it works

User DM ──► Discord Client ──► Queue (incoming/) ──► Queue Processor ──► Claude CLI
                                                                              │
User DM ◄── Discord Client ◄── Queue (outgoing/) ◄────────────────────────────┘

Both WhatsApp and Discord clients write JSON messages to .tinyclaw/queue/incoming/ with a channel prefix (discord_ / whatsapp_). The queue-processor picks them up FIFO, calls claude -c -p with the configured model, and writes responses to .tinyclaw/queue/outgoing/ where the originating client picks them up and replies.

Files changed

File	What changed
src/discord-client.ts	New — full Discord DM bot with typing indicators, message splitting, and queue I/O
src/queue-processor.ts	Added model config reading (MODEL_IDS map) and getModelFlag() for runtime model selection
tinyclaw.sh	Major overhaul — setup wizard, multi-channel support, dynamic tmux layouts, Discord validation, setup command
package.json	Added discord.js + dotenv dependencies, discord npm script
.gitignore	Added .tinyclaw/channel, .tinyclaw/model, .env
.claude/hooks/session-start.sh	Added Discord mention to startup banner

🤖 Generated with Claude Code

[unforsook]

I realise Claude changed a bit more than I thought I had steered it to, I am not a big fan of complexity, but I guess adding multiple channels requires some extra conditionals. Let me know what you think @jlia0

[jlia0]

I realise Claude changed a bit more than I thought I had steered it to, I am not a big fan of complexity, but I guess adding multiple channels requires some extra conditionals. Let me know what you think @jlia0

It's pretty good overall, although it gets messy for the shell script with different channels, do you usually use agents with multiple channels or just single channel like discord

[jlia0]

Closing this PR after merging #12

[unforsook]

I realise Claude changed a bit more than I thought I had steered it to, I am not a big fan of complexity, but I guess adding multiple channels requires some extra conditionals. Let me know what you think @jlia0

It's pretty good overall, although it gets messy for the shell script with different channels, do you usually use agents with multiple channels or just single channel like discord

yeah I usually use different channels, but tbh I think for simplicity, Discord DMs is completely fine here. Having tried it, perhaps I should remove the different channels code. You down for that?

[jlia0]

I'd really appreciate it if you could test #13!

I made the multi channel support easier.