stream Redstone replies via sendMessageDraft + fix self-killing server_down

Two related changes to how Redstone behaves in chat.

Streaming: replies now animate word-by-word into Telegram using the new
sendMessageDraft endpoint (Bot API 9.5, March 2026) via the @grammyjs/stream
plugin, with @grammyjs/auto-retry on the API layer to swallow 429s
transparently. The previous editMessageText-based approach is gone —
sendMessageDraft is designed for this and animates natively on the client
without hitting the 1/sec-per-chat edit limit. Pace lives in
STREAM_WORD_DELAY_MS=50; tunable in one spot.

server_down footgun: the MCP tool was running `compose down` with no
service arg, which tore down the whole project — including the bot
container running the tool call, which then got SIGTERM mid-conversation
and didn't come back (unless-stopped doesn't restart on a clean compose
down). Behaviour now matches the existing /stop slash command:
compose stop minecraft autossh, leaving the bot up. Pass an explicit
service to stop just that one.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

bot/bot.ts

@@ -1,4 +1,8 @@
 import { Bot, Context, GrammyError, HttpError, InlineKeyboard } from "grammy";
+import { autoRetry } from "@grammyjs/auto-retry";
+import { stream, type StreamFlavor } from "@grammyjs/stream";
+
+type BotContext = StreamFlavor<Context>;
 import { Database } from "bun:sqlite";
 import { readFileSync, mkdirSync } from "node:fs";
 import { dirname } from "node:path";
@@ -314,7 +318,14 @@ async function maybeSendSticker(ctx: Context, event: string) {
 }
 
 // ---- Bot ----
-const bot = new Bot(TOKEN);
+const bot = new Bot<BotContext>(TOKEN);
+
+// Auto-retry handles 429 Too Many Requests from Telegram automatically (waits the
+// retry_after window then re-sends). The stream plugin uses the new
+// sendMessageDraft endpoint (Bot API 9.5, March 2026) to natively animate
+// progressively-revealed text on the client — feels like an LLM typing into chat.
+bot.api.config.use(autoRetry());
+bot.use(stream());
 
 // Capture incoming + outgoing for the audit log; track usernames + seen users.
 bot.use(async (ctx, next) => {
@@ -1205,8 +1216,16 @@ bot.on("message:text", async (ctx) => {
   }
 
   const out = trim(reply, 3500);
-  const sent = await ctx.reply(out, { reply_parameters: { message_id: ctx.message.message_id } });
-  Q.insMsg.run(chatId, sent.message_id, null, "out", "text", out, ctx.message.message_id);
+  // Stream the reply word-by-word so the message animates into the chat like
+  // an LLM typing. @grammyjs/stream uses sendMessageDraft (Bot API 9.5) to push
+  // each delta as a native, animated draft and finalizes with sendMessage.
+  const messages = await ctx.replyWithStream(streamWords(out), {}, {
+    reply_parameters: { message_id: ctx.message.message_id },
+  });
+  const finalMessage = messages[messages.length - 1];
+  if (finalMessage) {
+    Q.insMsg.run(chatId, finalMessage.message_id, null, "out", "text", out, ctx.message.message_id);
+  }
 
   // Emoji → sticker swap. Scan the reply for an emoji that has a matching
   // sticker in the library and, if found, fire it as a follow-up. We don't
@@ -1224,6 +1243,19 @@ bot.on("message:text", async (ctx) => {
   }
 });
 
+// Word-by-word delta generator for ctx.replyWithStream. ~50ms between words
+// gives a brisk LLM-typing feel without dragging on long replies. The split
+// regex keeps whitespace so the rendered text reconstructs cleanly.
+const STREAM_WORD_DELAY_MS = 50;
+async function* streamWords(text: string): AsyncGenerator<string> {
+  const parts = text.split(/(\s+)/);
+  for (const p of parts) {
+    if (!p) continue;
+    yield p;
+    if (STREAM_WORD_DELAY_MS > 0) await new Promise((r) => setTimeout(r, STREAM_WORD_DELAY_MS));
+  }
+}
+
 // Walk the reply text and return the file_id of the first sticker-library
 // emoji that appears in it, or null if none match. We rank by where the emoji
 // occurs in the text (earliest wins) rather than by sticker frequency, so the

bot/package.json

@@ -7,6 +7,8 @@
     "start": "bun run bot.ts"
   },
   "dependencies": {
+    "@grammyjs/auto-retry": "^2.0.2",
+    "@grammyjs/stream": "^1.0.1",
     "grammy": "^1.30.0"
   },
   "devDependencies": {