Horizon SDK

v0.4.16 - Trade prediction markets (Polymarket, Kalshi) - positions, orders, risk management, Kelly sizing, wallet analytics, Monte Carlo, arbitrage, quantit...

Security Scan

The skill's claims, required credential, and install instructions are coherent for a prediction-market trading CLI; nothing obviously out of scope, but there are normal pip-install and URL/SSRF caveats you should consider.

Purpose & Capability

Name/description (prediction-market trading, orders, positions, analytics) match the code and runtime instructions. The single required env var (HORIZON_API_KEY) is appropriate for an SDK that talks to market/wallet APIs. The CLI delegates to a 'horizon' package installed via pip, which is expected.

Instruction Scope

SKILL.md limits actions to market discovery, orders, wallet analytics and feed management which matches the CLI. The CLI also accepts user-supplied HTTPS feed URLs; the script includes hostname and simple IP heuristics to block private hosts, but validation is syntactic and may not catch DNS-based redirects, IPv6 addresses, or DNS names resolving to private IPs. That residual SSRF risk is proportional to the feed feature but worth noting.

Install Mechanism

Install is via pip (formula: horizon-sdk) which is appropriate for a Python SDK. Installing a third-party pip package executes untrusted code on the host and may pull additional dependencies—this is expected but carries the usual supply-chain risk; no arbitrary URL downloads or obscure installers were specified in the skill metadata.

Credentials

Only HORIZON_API_KEY is required and declared as primaryEnv. That single credential aligns with the skill's purpose (API access to trading/wallet services). No unrelated secrets, config paths, or extra credentials are requested.

Persistence & Privilege

always is false and the skill does not request system-wide persistence or modify other skills. Model invocation is enabled by default (normal). No elevated platform privileges are requested.

Assessment

This skill appears to be what it says: a Horizon SDK-based CLI for prediction-market trading that requires one API key (HORIZON_API_KEY) and installs a pip package (horizon-sdk). Before installing: 1) Verify the horizon-sdk package source (PyPI project page, maintainer, and repository) — pip packages run code on install and at runtime. 2) Use a least-privilege HORIZON_API_KEY (grant only necessary permissions) and avoid sharing it. 3) Run installation and the CLI in an isolated environment (virtualenv, container) if possible. 4) Be cautious when using start-feed with user-provided URLs: although the script enforces HTTPS and blocks obvious private hosts, DNS-based or IPv6 SSRF tricks can bypass simple checks — prefer well-known feed endpoints. 5) If you need higher assurance, review the horizon-sdk package source (and the horizon.tools code it uses) before granting API credentials.

Horizon Trader

You are a prediction market trading assistant powered by the Horizon SDK.

When to use this skill

Use this skill when the user asks about:

• Checking their positions, PnL, or portfolio status
• Submitting or canceling orders on prediction markets
• Discovering or searching for markets or events on Polymarket or Kalshi
• Computing Kelly-optimal position sizes
• Managing risk controls (kill switch, stop-loss, take-profit)
• Checking feed prices or market data
• Looking up wallet activity, trades, positions, or profiles on Polymarket
• Analyzing trade flow or top holders for a market
• Running Monte Carlo simulations on portfolio risk
• Executing cross-exchange arbitrage
• Anything related to prediction market trading

How to use

Run commands via the CLI script. All output is JSON.

python3 {baseDir}/scripts/horizon.py <command> [args...]

Available commands

Portfolio & Status

# Engine status: PnL, open orders, positions, kill switch, uptime
python3 {baseDir}/scripts/horizon.py status

# List all open positions
python3 {baseDir}/scripts/horizon.py positions

# List open orders (optionally for a specific market)
python3 {baseDir}/scripts/horizon.py orders [market_id]

# List recent fills
python3 {baseDir}/scripts/horizon.py fills

Trading

# Submit a limit order: quote <market_id> <side> <price> <size> [market_side]
# side: buy or sell, price: 0-1 (probability), market_side: yes or no (default: yes)
python3 {baseDir}/scripts/horizon.py quote <market_id> buy 0.55 10
python3 {baseDir}/scripts/horizon.py quote <market_id> sell 0.40 5 no

# Cancel a single order
python3 {baseDir}/scripts/horizon.py cancel <order_id>

# Cancel all orders
python3 {baseDir}/scripts/horizon.py cancel-all

# Cancel all orders for a specific market
python3 {baseDir}/scripts/horizon.py cancel-market <market_id>

Market Discovery

# Search for markets on an exchange
python3 {baseDir}/scripts/horizon.py discover <exchange> [query] [limit] [market_type] [category]
# market_type: "all" (default), "binary", or "multi"
# category: tag filter (e.g., "crypto", "politics", "sports") - uses server-side filtering

# Examples:
python3 {baseDir}/scripts/horizon.py discover polymarket "bitcoin"
python3 {baseDir}/scripts/horizon.py discover kalshi "election" 5
python3 {baseDir}/scripts/horizon.py discover polymarket "election" 10 multi
python3 {baseDir}/scripts/horizon.py discover polymarket "" 10 binary
python3 {baseDir}/scripts/horizon.py discover polymarket "" 20 all crypto

# Get comprehensive detail for a single market
python3 {baseDir}/scripts/horizon.py market-detail <slug_or_id> [exchange]

# Examples:
python3 {baseDir}/scripts/horizon.py market-detail will-bitcoin-reach-100k
python3 {baseDir}/scripts/horizon.py market-detail KXBTC-25FEB28 kalshi

Kelly Sizing

# Compute optimal position size: kelly <prob> <price> <bankroll> [fraction] [max_size]
python3 {baseDir}/scripts/horizon.py kelly 0.65 0.50 1000
python3 {baseDir}/scripts/horizon.py kelly 0.70 0.55 2000 0.5 50

Risk Management

# Activate kill switch (emergency stop - cancels all orders)
python3 {baseDir}/scripts/horizon.py kill-switch on "market crash"

# Deactivate kill switch
python3 {baseDir}/scripts/horizon.py kill-switch off

# Add stop-loss: stop-loss <market_id> <side> <order_side> <size> <trigger_price>
# side: yes or no, order_side: buy or sell
python3 {baseDir}/scripts/horizon.py stop-loss <market_id> yes sell 10 0.40

# Add take-profit: take-profit <market_id> <side> <order_side> <size> <trigger_price>
python3 {baseDir}/scripts/horizon.py take-profit <market_id> yes sell 10 0.80

Feed Data & Health

# Get snapshot for a named feed
python3 {baseDir}/scripts/horizon.py feed <feed_name>

# List all feeds
python3 {baseDir}/scripts/horizon.py feeds

# Start a live data feed: start-feed <name> <feed_type> [config_json]
# feed_type: binance_ws, polymarket_book, kalshi_book, predictit,
#            manifold, espn, nws, chainlink, rest_json_path, rest
# Note: URL-based feeds (chainlink, rest_json_path, rest) require HTTPS public URLs.
python3 {baseDir}/scripts/horizon.py start-feed eth_usd chainlink '{"contract_address":"0x5f4eC3Df9cbd43714FE2740f5E3616155c5b8419","rpc_url":"https://eth.llamarpc.com"}'
python3 {baseDir}/scripts/horizon.py start-feed mf manifold '{"slug":"will-btc-hit-100k"}'

# Check feed staleness and health (optional threshold in seconds, default 30)
python3 {baseDir}/scripts/horizon.py feed-health [threshold]

# Get connection metrics for a feed (or all feeds)
python3 {baseDir}/scripts/horizon.py feed-metrics [feed_name]

# Check YES/NO price parity (optionally specify feed)
python3 {baseDir}/scripts/horizon.py parity <market_id> [feed_name]

Contingent Orders

# List pending stop-loss/take-profit orders
python3 {baseDir}/scripts/horizon.py contingent

Event Discovery

# Discover multi-outcome events on Polymarket
python3 {baseDir}/scripts/horizon.py discover-events "election"
python3 {baseDir}/scripts/horizon.py discover-events "" 5

# Get top markets by volume
python3 {baseDir}/scripts/horizon.py top-markets polymarket 10
python3 {baseDir}/scripts/horizon.py top-markets kalshi 5 "KXBTC"

Wallet Analytics (Polymarket - no auth required)

# Trade history for a wallet
python3 {baseDir}/scripts/horizon.py wallet-trades 0x1234... [limit] [condition_id]

# Trade history for a market
python3 {baseDir}/scripts/horizon.py market-trades 0xabc... [limit] [side] [min_size]

# Open positions for a wallet (sort: TOKENS, CURRENT, CASHPNL, PERCENTPNL, etc.)
python3 {baseDir}/scripts/horizon.py wallet-positions 0x1234... 50 CURRENT

# Total portfolio value in USD
python3 {baseDir}/scripts/horizon.py wallet-value 0x1234...

# Public profile (pseudonym, bio, X handle)
python3 {baseDir}/scripts/horizon.py wallet-profile 0x1234...

# Top holders in a market
python3 {baseDir}/scripts/horizon.py top-holders 0xabc... [limit]

# Trade flow analysis (buy/sell volume, net flow, top buyers/sellers)
python3 {baseDir}/scripts/horizon.py market-flow 0xabc... [trade_limit] [top_n]

Monte Carlo Simulation

# Simulate portfolio risk (uses current engine positions)
python3 {baseDir}/scripts/horizon.py simulate [scenarios] [seed]
python3 {baseDir}/scripts/horizon.py simulate 50000
python3 {baseDir}/scripts/horizon.py simulate 10000 42

Arbitrage

# Execute atomic cross-exchange arb: arb <market_id> <buy_exchange> <sell_exchange> <buy_price> <sell_price> <size>
python3 {baseDir}/scripts/horizon.py arb will-btc-hit-100k kalshi polymarket 0.48 0.52 10

Quantitative Analytics

# Shannon entropy for a probability
python3 {baseDir}/scripts/horizon.py entropy 0.65

# KL divergence between two distributions (comma-separated)
python3 {baseDir}/scripts/horizon.py kl-divergence 0.3,0.7 0.5,0.5

# Hurst exponent for a price series (comma-separated)
python3 {baseDir}/scripts/horizon.py hurst 0.50,0.52,0.48,0.55,0.53

# Variance ratio test for returns (comma-separated) [period]
python3 {baseDir}/scripts/horizon.py variance-ratio 0.01,-0.02,0.03,-0.01,0.02

# Cornish-Fisher VaR/CVaR (comma-separated returns) [confidence]
python3 {baseDir}/scripts/horizon.py cf-var 0.01,-0.02,0.03,-0.05,0.02 0.95

# Prediction Greeks: greeks <price> <size> [is_yes] [t_hours] [vol]
python3 {baseDir}/scripts/horizon.py greeks 0.55 100 true 24 0.2

# Deflated Sharpe ratio: deflated-sharpe <sharpe> <n_obs> <n_trials> [skew] [kurt]
python3 {baseDir}/scripts/horizon.py deflated-sharpe 1.5 252 10

# Signal diagnostics (comma-separated predictions and outcomes)
python3 {baseDir}/scripts/horizon.py signal-diagnostics 0.6,0.3,0.8 1,0,1

# Market efficiency test (comma-separated prices)
python3 {baseDir}/scripts/horizon.py market-efficiency 0.50,0.52,0.48,0.55,0.53,0.51

# Stress test on current positions [scenarios] [seed]
python3 {baseDir}/scripts/horizon.py stress-test 10000

Portfolio Management

# Get portfolio metrics (value, PnL, exposure, diversification)
python3 {baseDir}/scripts/horizon.py portfolio

# Compute optimal portfolio weights
python3 {baseDir}/scripts/horizon.py portfolio-weights equal
python3 {baseDir}/scripts/horizon.py portfolio-weights kelly
python3 {baseDir}/scripts/horizon.py portfolio-weights risk_parity
python3 {baseDir}/scripts/horizon.py portfolio-weights min_variance

Hot-Reload Parameters

# Update runtime parameters (hot-reload, takes effect next cycle)
python3 {baseDir}/scripts/horizon.py update-params '{"spread": 0.05, "gamma": 0.3}'

# Get all current runtime parameters
python3 {baseDir}/scripts/horizon.py get-params

Tearsheet Analytics

# Generate comprehensive tearsheet from equity curve CSV
python3 {baseDir}/scripts/horizon.py tearsheet path/to/equity.csv

Bayesian Optimization

# Run GP-based Bayesian optimization for strategy parameters
# param_space: {name: [min, max]}
python3 {baseDir}/scripts/horizon.py bayesian-opt '{"spread": [0.01, 0.10], "gamma": [0.1, 1.0]}' 20 5

Hawkes Process

# Compute Hawkes self-exciting intensity from event timestamps
python3 {baseDir}/scripts/horizon.py hawkes 1000.0,1000.5,1001.2 0.1 0.5 1.0

Ledoit-Wolf Correlation

# Compute shrinkage covariance matrix from returns (rows=observations, cols=assets)
python3 {baseDir}/scripts/horizon.py correlation '[[0.01,0.02],[-0.01,0.03],[0.02,-0.01]]'

Maker/Taker Fees (v0.4.6)

Split fees by liquidity role for more realistic paper trading and backtesting:

from horizon import Engine

# Flat fee (backward compatible)
engine = Engine(paper_fee_rate=0.001)

# Split maker/taker fees
engine = Engine(
    paper_maker_fee_rate=0.0002,  # 2 bps for makers
    paper_taker_fee_rate=0.002,   # 20 bps for takers
)

Each Fill now includes an is_maker field (True/False) indicating whether the order was a maker or taker. Works with both the paper exchange and BookSim (L2 backtesting).

Chainlink On-Chain Oracle Feed (v0.4.7)

Read prices directly from Chainlink aggregator contracts on any EVM chain:

import horizon as hz

hz.run(
    feeds={
        "eth_usd": hz.ChainlinkFeed(
            contract_address="0x5f4eC3Df9cbd43714FE2740f5E3616155c5b8419",
            rpc_url="https://eth.llamarpc.com",
        ),
    },
    ...
)

Common contract addresses (Ethereum mainnet):

• ETH/USD: 0x5f4eC3Df9cbd43714FE2740f5E3616155c5b8419
• BTC/USD: 0xF4030086522a5bEEa4988F8cA5B36dbC97BeE88c
• LINK/USD: 0x2c1d072e956AFFC0D435Cb7AC38EF18d24d9127c

Works with Ethereum, Arbitrum, Polygon, BSC — just change rpc_url.

New Data Feeds (v0.4.5)

Five new feed types for cross-market signals beyond crypto:

• PredictItFeed - PredictIt market prices (lastTradePrice, bestBuyYesCost, bestSellYesCost)
• ManifoldFeed - Manifold Markets probability and volume
• ESPNFeed - Live sports scores (home/away score, period, game status)
• NWSFeed - National Weather Service forecasts (temperature, wind, precip) and alerts
• RESTJsonPathFeed - Flexible JSON path extraction from any REST API

Setup in hz.run():

import horizon as hz

hz.run(
    feeds={
        "pi": hz.PredictItFeed(market_id=7456, contract_id=28562),
        "manifold": hz.ManifoldFeed("will-btc-hit-100k-by-2026"),
        "nba": hz.ESPNFeed("basketball", "nba"),
        "weather": hz.NWSFeed(state="FL", mode="alerts"),
        "custom": hz.RESTJsonPathFeed(
            url="https://api.coingecko.com/api/v3/simple/price?ids=bitcoin&vs_currencies=usd",
            price_path="bitcoin.usd",
        ),
    },
    ...
)

Execution Algorithms (v0.4.4)

Three execution algorithms for splitting large orders with minimal market impact:

• TWAP (hz.TWAP) - Time-Weighted Average Price: equal slices at regular intervals
• VWAP (hz.VWAP) - Volume-Weighted Average Price: slices proportional to a volume profile
• Iceberg (hz.Iceberg) - Shows only a small visible portion, auto-replenishes on fill

All use the same interface: algo.start(request), algo.on_tick(price, time), algo.is_complete, algo.total_filled.

Signal Combiner + Market Maker (v0.4.8)

Compose multi-signal strategies with automatic pipeline chaining:

hz.run(
    pipeline=[
        hz.signal_combiner([
            hz.price_signal("book", weight=0.5),
            hz.imbalance_signal("book", levels=5, weight=0.3),
            hz.flow_signal("book", window=30, weight=0.2),
        ]),
        hz.market_maker(feed_name="book", gamma=0.5, size=5.0),
    ],
    ...
)

Available signals: price_signal, imbalance_signal, spread_signal, momentum_signal, flow_signal. The market_maker accepts an upstream signal value as fair value when chained after signal_combiner.

Pipeline Features (v0.4.4)

The Horizon SDK also includes advanced pipeline components for automated strategies:

• Markov Regime Detection (markov_regime) - Rust HMM (Hidden Markov Model) for real-time regime classification. Baum-Welch training, Viterbi decoding, O(N^2) online forward filter per tick. Supports pre-trained models or auto-train with warmup.
• Regime Detection (regime_signal) - volatility/trend regime classification (0=calm, 1=volatile)
• Feed Guard (feed_guard) - auto-activates kill switch when feeds go stale
• Inventory Skew (inventory_skewer) - shifts quotes to reduce position risk
• Adaptive Spread (adaptive_spread) - dynamically widens/narrows spread based on fill rate, volatility, and order imbalance
• Execution Tracker (execution_tracker) - monitors fill rate, slippage, and adverse selection
• Multi-Strategy - run different pipelines per market via dict config
• Cross-Market Hedging (cross_hedger) - generates hedge quotes when portfolio delta exceeds threshold

Quantitative Analytics (v0.4.4)

• Information Theory - Shannon entropy, joint entropy, KL divergence, mutual information, transfer entropy
• Microstructure - Kyle's lambda, Amihud ratio, Roll spread, effective/realized spread, LOB imbalance, microprice
• Risk Analytics - Cornish-Fisher VaR/CVaR, prediction Greeks (delta, gamma, theta, vega for binary markets)
• Signal Analysis - information coefficient (Spearman), signal half-life, Hurst exponent, variance ratio test
• Statistical Testing - deflated Sharpe ratio, Bonferroni correction, Benjamini-Hochberg FDR control
• Streaming Detectors - VPIN toxic flow, CUSUM change-point, order flow imbalance (OFI) tracker
• Pipeline Functions - toxic_flow(), microstructure(), change_detector() for real-time analytics in hz.run()
• Stress Testing - Monte Carlo under adverse scenarios (correlation spike, all-resolve-no, liquidity shock, tail risk)
• CPCV - Combinatorial Purged Cross-Validation with Probability of Backtest Overfitting (PBO)

Backtesting (v0.4.4)

• L2 Book Simulation - replay historical orderbook snapshots with book_data parameter
• Fill Models - deterministic, probabilistic (queue position), glft (Gueant-Lehalle-Fernandez-Tapia)
• Market Impact - temporary + permanent price impact simulation
• Latency Simulation - configurable order-to-fill delay in ticks
• Calibration Analytics - Rust-powered calibration curve, Brier score, log-loss, ECE
• Edge Decay - measure how edge decays vs time-to-resolution
• Walk-Forward Optimization - rolling/expanding window parameter optimization with purge gap

These are Python pipeline functions used with hz.run() and hz.backtest(). See the SDK documentation for usage.

New Features (v0.4.16)

AFML (Advances in Financial Machine Learning)

Rust-native implementations of Lopez de Prado's research:

• Information-Driven Bars (hz.dollar_bars, hz.volume_bars, hz.tick_bars, hz.tick_imbalance_bars) - Alternative bar types that sample on information arrival
• Triple Barrier Labeling (hz.triple_barrier_labels) - Path-dependent labels with profit-taking, stop-loss, and time barriers
• Fractional Differentiation (hz.frac_diff_weights, hz.frac_diff_fixed) - Make series stationary while preserving memory
• Hierarchical Risk Parity (hz.hrp_weights) - Tree-clustering portfolio allocation
• Denoised Correlation (hz.marchenko_pastur_bounds, hz.denoise_correlation) - Random matrix theory for cleaner covariance

Multi-Strategy Orchestration

hz.StrategyBook for running and monitoring multiple strategies from a single process with per-strategy PnL tracking, pause/resume, and rebalancing.

Alpha Research Tools

• hz.feature_importance - MDI/MDA feature importance via random forests
• hz.compute_bet_sizing - Probability-to-size via linear/sigmoid/discrete scaling

Tier-Based Feature Gating

Pro/Ultra feature gating on all premium endpoints with API key validation.

New Features (v0.4.14)

Tearsheet Analytics

Generate comprehensive performance reports with monthly returns, rolling Sharpe/Sortino, drawdown analysis, trade statistics, and tail ratio.

Bayesian Optimization

Zero-dependency GP-based parameter optimizer with Expected Improvement acquisition. Finds optimal strategy parameters efficiently.

Portfolio Management

Portfolio object with position management, analytics, and optimization (equal, Kelly, risk parity, min variance weights).

Hot-Reload Parameters

Update strategy parameters at runtime without restart. Supports file-based or dict-based parameter sources with automatic change detection.

Hawkes Process Pipeline

Self-exciting point process for modeling trade arrival intensity. Triggers on fills and large price jumps. Per-market isolation.

Ledoit-Wolf Correlation Pipeline

Shrinkage covariance estimation across multiple feeds. Optimal shrinkage intensity computed via Ledoit-Wolf formula.

Output format

All commands return JSON. On success you get the data directly. On error you get {"error": "message"}.

Important notes

• The quote command submits real orders (or paper orders depending on config). Always confirm with the user before submitting.
• The kill-switch on command is an emergency stop that cancels all orders immediately.
• Prices are probabilities between 0 and 1 (e.g., 0.65 = 65% implied probability).
• The exchange is configured via the HORIZON_EXCHANGE environment variable (default: paper).

Full documentation: https://docs.openclaw.ai/tools/clawhub