smart-money

TraderSpy follows thousands of top-ranked accounts on four exchanges and records their positions as they open, change and close. This skill turns that feed into answers about positioning and about individual traders. Positions are observations of what other people did with their own money — present them as evidence, never as instructions.

Tools

Always pass source with a trader id. get_trader_profile and get_trader_position_history default to Binance; a Hyperliquid address (0x…) or an OKX id looked up on the wrong exchange returns nothing. Take source from the row you got the id from.

Reading the data

Symbols differ per exchange. Hyperliquid rows use bare coins (BTC, ZEC) and prefix tokenized stocks with xyz: (xyz:AVGO); Binance, Bybit and OKX rows use BTCUSDT. When the user asks about "BTC positions", pass symbol: "BTC" — the filter matches both BTC and BTCUSDT exactly — and read the symbol field before summing anything. Tokenized stocks (xyz:AVGO) are not reachable through the filter; fetch without symbol and pick them out.

Position rows: side LONG / SHORT, size in coins, entryPrice, markPrice, leverage, unrealizedPnl (open rows) or pnl + roi (closed rows), openTime / closeTime, isOpen, lastEvent (increase, partial_close, or null when the last change was the open itself). Notional ≈ size × markPrice; use notional, not row count, when you say "whales are net long".

Freshness. Positions are read from each exchange on a schedule: the top-ranked traders about once a minute, the rest of the tracked set every 5–15 minutes. There is no live fill stream, so a new position can show up a few minutes after it opened. On a free-tier key the feed is delayed 15 minutes — if get_positions returns nothing newer than that, say the feed is delayed rather than "no activity".

Leaderboard rows (get_top_traders) carry the exchange's own ranking snapshot: roi, pnl, assets (account size), winRate (null on Hyperliquid — the exchange does not publish it), rankings[] with the same trader across every window, and smartScore when TraderSpy has scored them. traderName is a shortened address when the account has no public name. A trader who is #1 on 30-day ROI with assets of $5k is a different animal from #19 with $3.4M — quote both.

Smart score (get_elite_leaderboard, 0–100, recomputed daily): realized PnL 30% (all-time blended with the last 30 days), effective win rate 22%, ROI edge 18%, consistency 14%, longevity 10%, trade depth 6%, plus a recency bonus (up to +9) and a staleness penalty (down to −12 after about three weeks idle). scoreBreakdown shows each part, rationale[] is a ready-made plain-language justification, metrics has the raw counts (closedTrades, profitFactor, realizedPnl30d, openPositionCount, worstDayPnl). reliabilityMultiplier below 1 means a short history dragged the score down — say so when a 6-day-old account ranks top-5.

Market stats aggregate every tracked position, thousands of them, so realizedPnl can be a large negative number even when the leaders are winning; the tracked universe includes traders who fell off the leaderboard. Use it for counts and the exchange split, not as "smart money is losing".

Workflows

"What are whales doing in SOL?" → get_positions with symbol, status: open, limit 50. Sum notional by side, name the largest two or three accounts with entry and leverage, and note how recent the newest entries are. If open rows are few, add status: closed for the last day to show whether they have been exiting.

"Who are the best traders right now?" → get_elite_leaderboard first (it is cross-exchange and score-based), then get_top_traders for a specific exchange or window if the user wants raw ROI/PnL. Explain that 7-day ROI rankings reward one lucky week; the smart score is built to penalise exactly that.

"Research this trader" / "Is trader X any good?" → get_trader_profile + one or two pages of get_trader_position_history. From the history compute what the user actually needs: win rate from closed rows, typical hold time (closeTime − openTime), typical leverage, how concentrated in one coin, worst single trade, and whether the recent month looks like the all-time record. Present it as a profile, not as a reason to follow anyone: this connector cannot follow, copy or trade for anyone.

"Are the pros long or short?" (whole market) → get_positions status: open on the majors plus get_market_stats; for exchange-wide long/short ratios of ALL accounts use get_derivatives from technical-analysis instead, which answers that directly.

Presenting

Positions: one row per position —

| Trader | Exchange | Side | Size (notional) | Entry | Mark | Lev | uPnL | Opened |

Traders: name, exchange, score or rank, window, ROI, PnL, account size, win rate (or "n/a" on Hyperliquid), then the two-line rationale. Lead with the number that answers the question, keep addresses shortened as returned, and always say which window a ROI belongs to.

Conduct

• Positions are what other traders did, with their capital, their hedges and their exits, which you cannot see in full. Present them as evidence about positioning, never as a recommendation to mirror them; a leader's long can be one leg of a hedge.
• Past ROI and win rates describe a past sample — do not present a leaderboard as a forecast of who will win next. Point out short histories and small samples.
• Nothing in this connector trades or copies. There is no order, follow, transfer or withdrawal tool, by design. If asked to copy or execute a trade, say so plainly; the decision and the trade stay with the user.
• Quote only what the tools returned; say when data is delayed or a field is null.
• When the answer is about a specific trade idea, end with one plain sentence that crypto derivatives are high-risk and this is market information, not financial advice.

For exchange-specific quirks and refresh cadences read references/exchange-notes.md.