Hyperliquid whale alerts in Telegram

Build an agent workflow that runs every 30 minutes on a cron trigger and posts curated Hyperliquid whale activity to a configured Telegram chat. The point of the agent is judgement: vanilla whale webhooks fire on every $1M move and the channel becomes unreadable, so this agent should rank and contextualize moves and stay silent when nothing meaningful happened.

On each run, the agent should pull two things from CoinGlass:

1. Get Hyperliquid Whale Alert — recent $1M+ position events on Hyperliquid.

2. Get Hyperliquid Whale Positions — the current open whale book, used as context for how each wallet was already positioned.

After fetching, the agent should:

• Deduplicate against the previous run. Persist the set of alert IDs (or address + coin + side + timestamp signatures) it has already covered, and drop anything it has already posted about. The agent's persistent state is the right place for this.

• Filter out noise: very small notional relative to that wallet's existing book, addresses already covered in the last few runs, and alerts older than the current 30-minute window.

• Rank what is left by (a) notional value in USD and (b) newness versus that wallet's prior exposure, using the current whale positions data as the baseline. A wallet flipping from flat to a $20M long is more interesting than a wallet adding $2M to an existing $200M position.

• Take the top 1 to 3 moves. If there are zero moves that clear the bar, send nothing — silence is the correct output.

For each surviving move, write a one-paragraph blurb that includes: the coin, the side (long or short), size in USD and as a percent of that wallet's prior exposure on that coin, leverage, and a short plain-English read on whether it looks like a fresh directional bet, a hedge against another open position the wallet holds, or a top-up of an existing position. Keep each blurb tight, a few sentences, written for a trader skimming Telegram on their phone.

Send each blurb as its own Telegram message via the Telegram Send Message operation, posted to a chat identifier the user configures at setup (their own account's Saved Messages, a private chat, a group, or a channel they have access to).

Constraints worth respecting: CoinGlass returns a standard envelope where code "0" means success, so check that before reading data. Honor rate limits and back off on 429s. For Telegram, send at roughly one message per second per chat and back off on FloodWaitError. If the CoinGlass call fails for the run, skip the run quietly rather than posting an error to the chat.

Inputs the workflow should expose at setup: the Telegram chat identifier (peer) to post into, an optional minimum notional threshold (default $1M), and an optional list of coins to focus on (default: all).