18 commands · 4 skills · 3 agents · 23,217 tokens measured
Research prediction markets and crypto with Claude Code: find the edge, size it with Kelly, backtest your calls, and run a live Polymarket and crypto terminal. Research-first, read-only by default.
14-day refunds · works on 3 devices · plain Markdown you can read
slash commands
skills, auto-loaded
specialist agents
tokens, measured at pack time
Command explorer
Nothing hidden behind a demo call. This is the complete /tk command list, the same files that land in your .claude directory.
/tk alertTrack & proveSet a read-only whale or price alert: poll the data on an interval and notify when a threshold is crossed. No trading.
/tk arbResearchFind the same event priced differently across venues: devig both, show the gap, and the edge it implies.
/tk backtestTrack & proveScore your forecast log: ROI, win rate, Brier, the calibration table, max drawdown, and an honest sample-size call.
/tk botTerminal & botGenerate the autonomous trading bot from TRADER.md: scanner, bull/bear/judge consensus, and a dry-run-default executor.
/tk calibrationTrack & proveCalibration report on your forecast log: the reliability table, Brier, and an honest skilled-or-lucky verdict.
/tk cryptoTerminal & botScan crypto: rank movers by 24h volume, read one symbol's ticker and book, and hand back the single best setup.
/tk decideDecide & sizeMake a full trade decision on one market: research a probability, find the edge, size it with Kelly, or say pass.
/tk holdersResearchSee who holds one market, which side they're on, and how concentrated the smart money is at the top.
/tk journalTrack & proveLog a forecast or trade to trades.csv, fill in outcomes when markets resolve, and show a running calibration snapshot.
/tk portfolioDecide & sizeYour open positions vs your risk caps: total exposure, per-market risk percent, and any cap breaches flagged loud.
/tk reportTrack & proveWeekly performance report a stakeholder can read: ROI, calibration verdict, best and worst calls, and what changed.
/tk researchResearchDeep-research one event into a probability q, with the edge vs the live market and the sources behind it.
/tk scanResearchScan live markets for edge opportunities, rank them by a transparent score, and surface the single best setup.
/tk setupDecide & sizeFirst-run setup: create TRADER.md, build the venv, check the live data APIs are reachable, and run a first scan.
/tk strategyTerminal & botTurn a described crypto strategy into a backtesting.py Strategy, run it on real OHLCV, and report ROI, drawdown, and Sharpe.
/tk terminalTerminal & botLaunch the live DEEPMIND-RUNTIME trading terminal from your own markets, save a PNG snapshot, and recap the strongest signal.
/tk watchlistResearchManage your watchlist in TRADER.md and scan only those markets for edge and whale context.
/tk whalesResearchTrack the biggest live trades, group them by market, and flag where smart money is piling in one-sided.
Flagship: /tk decide
Take one market from 'should I bet this?' to a sized, risk-capped decision. It pulls the live price, researches a probability with a bull/bear/judge debate, computes the edge and EV, then sizes the bet with quarter-Kelly, or tells you to pass.
It pulls the live Polymarket or Kalshi price, the order book, the whale trades over $3,000, and the largest holders. You see what the crowd is paying and which side the big wallets are on before you commit to a view.
A bull case and a bear case each get a number, then a judge weighs them against base rates and the live positioning. Your probability starts at the market price and only moves when you can state the reason in one sentence.
Edge is your probability minus the price. Full Kelly has roughly a one in three chance of halving your bankroll before it doubles, so the kit stakes a quarter of Kelly and caps every bet to your limits. No edge means the price is right and you pass, which is the answer most of the time.
Under the hood
Skills are knowledge files Claude Code loads automatically when the topic comes up. Agents are read-only specialists a command can dispatch for research or review without polluting your main context.
Auto-loaded context: the rules, benchmarks, and playbooks the commands cite. Plain Markdown you can open and read.
tk-contextShared memory: TRADER.md holds your venues, watchlist, bankroll, risk caps, and mode (research, paper, or live).
tk-edgeThe quant brain: correct prediction-market Kelly, edge and EV, vig removal, Brier calibration, and bull/bear/judge consensus.
tk-terminalThe live trading terminal engine: read-only Polymarket and crypto data, the scanner, whale flow, and the dashboard renderer.
tk-backtestProve an edge is real, not luck: backtest a forecast log for ROI, calibration, and a Kelly bankroll curve with drawdown.
Dispatched by commands for focused work. Read-only by design: they research and report, they never edit your files.
tk-risk-auditorRead-only review of a trade or the current book against your risk caps. Catches oversizing, breaches, and correlated bets.
tk-market-researcherRead-only deep research on one market into a probability, with base rates, recent news, positioning, and sources.
tk-calibration-criticRead-only honest review of your track record: separates skill from luck, checks the sample size, finds the miscalibration.
Install
Authenticate the CLI with your license key, then install. The files land in ./.claude (or ~/.claude with -g) and every /tk command shows up immediately.
$ ck install tradekit/plugin install flow; both paths ship the same files.Use cases
/tk-scan ranks every live Polymarket and Kalshi market by an edge score: your model's probability against the price the market is paying. /tk-whales shows the biggest trades and flags where one wallet is loading a single side, and /tk-holders shows how concentrated the smart money is. The markets where your number and the crowd's price disagree are where you look first.
Kelly stakes a fraction of your bankroll equal to your edge divided by the payout if you win. The kit uses the version that is correct for 0-to-1 prediction-market contracts, then takes a quarter of it, because full Kelly has about a one in three chance of halving your bankroll before it doubles. Every stake is also capped to the per-trade and per-day limits in your TRADER.md.
Not reliably, and any tool that promises it is selling you something. On-chain data shows roughly 84% of Polymarket wallets lose money, and a real edge fades once enough people find it. What TradeKit gives you is a sharper probability and disciplined sizing, with you in the loop on every decision. It is a research desk, not a money printer.
Yes. /tk-crypto reads movers, tickers, and order books across 100-plus exchanges through ccxt, all read-only. /tk-strategy turns a described idea into a backtesting.py run on real candles and reports return, max drawdown, and Sharpe. It also tells you plainly that a backtest assumes perfect fills and ignores slippage and fees, so live results come in worse.
Pricing
Both come with 14-day refunds and licenses that cover 3 devices.
One-time, the kit as shipped
$99once
Every kit, every update, while active
$49.99/mo
FAQ
18 slash commands, 4 skills, and 3 agents, written as plain Markdown with small Python engines, measured at 23,217 tokens. The engines read live data from Polymarket, Kalshi, and crypto with no API keys, and the tk-edge skill carries the Kelly, edge, expected value, vig-removal, and Brier-score math. Install it with ck install tradekit or as a Claude Code plugin.
The other kits