Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Watches NEAR's price across Binance, Bybit, and Coinbase and alerts the user on Telegram when the prices drift apart by ~1.5% or more — an early sign of an arbitrage gap or of trouble (paused withdrawals, broken liquidity) on one exchange. Runs every 2 hours, logs each snapshot to build a spread history, and stays silent while the exchanges agree.
.claude/skills/nearai-cross-exchange-spread-monitor/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 45% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 44% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-21 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-22 | ✗→✓ | ▲ Improved | -31% | 0% |
You watch NEAR's price across three exchanges and warn the user when they drift apart — an early signal of arbitrage or trouble on one exchange.
spreads/history.md with memory_read before writing, then append the new snapshot with memory_write. Never overwrite the file from scratch and never drop earlier snapshots. On the first run, if the file does not exist, create it with the current snapshot as the baseline and stop — no alert.time tool. Never guess it.HEARTBEAT_OK and stop — send no message.Create a routine that runs every 2 hours. The routine goal must contain these full steps as a self-contained prompt, because a routine does not keep any context from this conversation when it runs:
http tool:https://api.binance.com/api/v3/ticker/price?symbol=NEARUSDThttps://api.bybit.com/v5/market/tickers?category=spot&symbol=NEARUSDThttps://api.coinbase.com/v2/prices/NEAR-USD/spottime tool.spreads/history.md with memory_read, then append this snapshot: timestamp, the three prices, the widest gap and its pair.HEARTBEAT_OK and stop.Alert format:
⚡ NEAR Cross-Exchange Spread
Widest gap: ~[X]% between [exchange A] and [exchange B]
- Binance: $[price]
- Bybit: $[price]
- Coinbase: $[price]
💡 A gap this wide usually means one of two things: a real arbitrage window, or paused withdrawals / broken liquidity on one venue. A gap that can't be arbitraged away persists — check that exchange's status before trading.
Normal spread for NEAR is under ~0.5%.
🕐 [timestamp]show spread — fetch the three prices right now and show the current gapsshow spread history — list the recent snapshots from memory with their widest gaps| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 15,376 | 8,114 | -47% | 1 | 1 | 0% | 2,979 | 1,193 | -60% | 0 | 0 | — |
case-17 | fail→fail | 2,684 | 9,320 | +247% | 1 | 1 | 0% | 403 | 1,378 | +242% | 0 | 0 | — |
case-02 | fail→fail | 5,747 | 6,900 | +20% | 1 | 1 | 0% | 897 | 1,216 | +36% | 0 | 0 | — |
case-03 | fail→fail | 6,353 | 3,634 | -43% | 1 | 1 | 0% | 1,230 | 979 | -20% | 0 | 0 | — |
case-04 | pass→pass | 9,903 | 5,052 | -49% | 1 | 1 | 0% | 1,805 | 1,658 | -8% | 0 | 0 | — |
case-05 | pass→pass | 7,830 | 6,881 | -12% | 1 | 1 | 0% | 1,647 | 2,134 | +30% | 0 | 0 | — |
case-06 | pass→fail | 17,904 | 11,153 | -38% | 1 | 1 | 0% | 3,504 | 1,550 | -56% | 0 | 0 | — |
case-07 | fail→fail | 8,835 | 6,508 | -26% | 1 | 1 | 0% | 2,056 | 2,108 | +3% | 0 | 0 | — |
case-08 | fail→pass | 5,648 | 3,628 | -36% | 1 | 1 | 0% | 1,064 | 1,544 | +45% | 0 | 0 | — |
case-09 | fail→pass | 5,591 | 3,476 | -38% | 1 | 1 | 0% | 1,049 | 1,514 | +44% | 0 | 0 | — |
case-10 | fail→fail | 4,748 | 1,239 | -74% | 1 | 1 | 0% | 839 | 910 | +8% | 0 | 0 | — |
case-18 | fail→fail | 2,679 | 6,191 | +131% | 1 | 1 | 0% | 379 | 1,037 | +174% | 0 | 0 | — |
case-11 | fail→pass | 11,459 | 3,283 | -71% | 1 | 1 | 0% | 2,208 | 1,510 | -32% | 0 | 0 | — |
case-12 | pass→pass | 8,398 | 4,220 | -50% | 1 | 1 | 0% | 1,460 | 1,753 | +20% | 0 | 0 | — |
case-13 | pass→pass | 6,962 | 2,622 | -62% | 1 | 1 | 0% | 1,230 | 1,359 | +10% | 0 | 0 | — |
case-14 | pass→pass | 7,230 | 2,066 | -71% | 1 | 1 | 0% | 1,279 | 1,126 | -12% | 0 | 0 | — |
case-15 | pass→pass | 13,002 | 4,601 | -65% | 1 | 1 | 0% | 2,237 | 1,892 | -15% | 0 | 0 | — |
case-16 | pass→pass | 9,127 | 1,653 | -82% | 1 | 1 | 0% | 2,047 | 1,059 | -48% | 0 | 0 | — |
case-19 | pass→pass | 7,778 | 3,982 | -49% | 1 | 1 | 0% | 1,558 | 1,526 | -2% | 0 | 0 | — |
case-20 | pass→pass | 3,172 | 2,618 | -17% | 1 | 1 | 0% | 692 | 1,272 | +84% | 0 | 0 | — |
case-21 | fail→pass | 17,535 | 2,339 | -87% | 1 | 1 | 0% | 1,848 | 1,162 | -37% | 0 | 0 | — |
case-22 | fail→pass | 10,353 | 2,144 | -79% | 1 | 1 | 0% | 1,569 | 1,077 | -31% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 16 counted toward the lift figure. The other 6 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +18 percentage points is the difference between those two pass rates over the 16 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.