Atomic Watch

atomic-watch — long-running auto-collect for pump.fun creator fees

src/watch-collect.js is a daemon that polls a coin's coinCreatorVault PDA on a schedule and fires src/collect-jito.js (atomically: collect + drain to DESTINATION) when the vault crosses MIN_COLLECT_SOL.

It's the script you deploy to Railway / Fly / a screen session when you want creator fees swept to a safe wallet automatically.

When to invoke

Pick this skill when the user wants to:

• Deploy an auto-collect bot to a persistent host (Railway, Fly, bare VM, screen session).
• Avoid manually running collect-jito.js every time the vault fills up.
• Drain on a threshold (e.g. "only collect when ≥ 0.05 SOL has accumulated, otherwise the Jito tip eats the proceeds").

Skip this skill if the user only wants:

• A one-shot collect right now → use skills/collect/.
• A full drain (vault + creator + funder) → use skills/consolidate/.
• Notifications without collection → use tools/check-balances.ts polled by your own cron.

Required environment

RPC_URL=...                     # paid provider strongly recommended for long-running
FUNDER_SECRET=<base58>          # or FUNDER_KEYPAIR=./funder.json
CREATOR_PUBKEY=<base58>         # pubkey only; the script doesn't need the secret for polling
CREATOR_SECRET=<base58>         # needed to sign the collect ix when threshold is met
MINT=<base58>                   # the pump.fun coin's mint
DESTINATION=<base58>            # safe wallet for the drained SOL
MIN_COLLECT_SOL=0.05            # vault threshold (SOL) before firing collect
POLL_INTERVAL_SECONDS=30        # default 30; lower if your RPC supports it
JITO_TIP=0.005                  # raise in busy markets

CREATOR_PUBKEY lets you keep the creator secret offline most of the time — the poller only needs the pubkey. The secret is loaded at collect-fire time. This is the less wrong option for a long-running daemon; the right option is a remote signer (out of scope).

Usage

Foreground:

npm run watch
# or
node src/watch-collect.js

Background (Railway / Fly / systemd):

# In Railway's `Start Command`:
node src/watch-collect.js

The script logs every poll cycle, every threshold crossing, and every collect attempt.

Operational characteristics

• Poll cadence: POLL_INTERVAL_SECONDS (default 30s). Faster polling costs more RPC requests; slower polling means fees sit in the vault longer.
• Exit behavior: any uncaught exception exits the process. Use a supervisor (PM2, systemd, Railway's restart policy, Fly's machine restart).
• Memory: ~100 MB resident. No leaks observed over multi-day runs.
• RPC rate limits: Helius free tier sustains ~60s polls indefinitely. Public RPC will rate-limit within minutes.
• Tip auction: if collect bundles return Invalid, the script logs and retries with the same tip. Raise JITO_TIP if this becomes persistent.

Failure modes and fixes

Worked example — Railway deployment

1. Fork or clone this repo into your Railway project.
2. Railway → Variables: set every *_SECRET, MINT, DESTINATION, MIN_COLLECT_SOL, JITO_TIP, RPC_URL.
3. Railway → Settings → Start Command: node src/watch-collect.js.
4. Deploy. Logs should show "Polling vault..." every 30s.
5. After the vault crosses MIN_COLLECT_SOL, you should see a "Collecting..." line followed by a Solscan link.

The destination wallet's balance should grow over time. Verify with npm run check-balances -- <destination>.

Reference

• Script: src/watch-collect.js
• Per-script docs: docs/scripts/watch-collect.md
• Atomic primitive: skills/collect/ (the one-shot collect that this daemon wraps)