fix(telegram): include the replied-to / quoted message in the prompt

[hoangqnguyen]

Problem

When a user replies to or quotes a message in Telegram, Claude never sees the quoted content — it responds as if the reply context didn't exist.

The Telegram API delivers the original message in reply_to_message (and, for partial highlights, a top-level quote field). But handleMessage() only captured the id and sender:

reply_to_message?: { message_id?: number; from?: TelegramUser };

…and the prompt builder only ever added the user's new message:

const promptParts = [`[Telegram from ${label}]`];
...
} else if (text.trim()) {
  promptParts.push(`Message: ${wrapUntrusted("user-message", text)}`);
}

So the replied-to/quoted text was silently dropped before Claude ever ran.

Fix

• Extend the TelegramMessage type to include reply_to_message.text / .caption and a quote.text field.
• When present, prepend the quoted context to the prompt as untrusted input (capped at 2000 chars, consistent with voice transcripts and other external content):

const repliedText = message.reply_to_message?.text ?? message.reply_to_message?.caption;
const quotedExcerpt = message.quote?.text;
if (quotedExcerpt || repliedText) {
  const fromBot = botId != null && message.reply_to_message?.from?.id === botId;
  const who = fromBot
    ? "your own earlier message"
    : `a message from ${message.reply_to_message?.from?.first_name ?? "someone"}`;
  promptParts.push(`In reply to ${who}: ${wrapUntrusted("replied-message", quotedExcerpt ?? repliedText!, 2000)}`);
}

quote.text (the highlighted excerpt when a user quotes only part of a message) takes precedence over the full reply_to_message body; the from-vs-bot check makes the phrasing natural in both DMs (replying to the bot) and groups (quoting another user).

Behavior

• Before: "reply to a message + ask about it" → Claude answers without the quoted text.
• After: the quoted text is included as context, so Claude can act on it.
• Media-only replies with no text/caption are unaffected (nothing to include — this covers text and captions, not re-reading an old attachment's bytes).

Testing

• bunx tsc --noEmit clean.
• bun test: only the pre-existing sessionFiles/sessions POSIX-path tests fail on a Windows dev box — unrelated to this change.

Notes

Independent of #250 (queue-on-busy) and #251 (argv-limit); any merge order works. Version bumped to 1.0.43 to avoid clashing with those PRs — happy to renumber/rebase if you'd prefer.

[TerrysPOV]

Clean, well-scoped fix. Including the replied-to / quoted message gives Claude the context it was missing, and — importantly — the quoted content is correctly fenced with wrapUntrusted("replied-message", …), so an untrusted reply body can't smuggle instructions into the prompt. Verified botId is in scope (module-level, set from getMe), the text/caption fallback and the partial-quote.text case are handled, and it merges clean against master.

Maintainer-committed the version housekeeping in a9a8066: the branch had bumped to 1.0.43, which equals current master, so the version guard would fail (base == head). I merged current master in and re-bumped to 1.0.44. Please run bun run bump:plugin-version + bump:marketplace-version yourself and target one past master for future PRs.

Approving.