A Claude skill for deep analysis of Polymarket prediction market trader accounts. Fetches live on-chain data, computes performance metrics, and classifies trading strategies.
Give it a Polymarket wallet address and it will:
- Fetch all trading history via the public Polymarket API (no key required)
- Compute 25+ performance, behavioral, timing, and risk metrics
- Classify the trader's strategy (one of 6 canonical Polymarket strategies)
- Generate a structured analyst report with evidence-backed conclusions
It can also scan the leaderboard and compare multiple accounts side by side.
| Strategy | Description |
|---|---|
| Information Arbitrage | Enters before the market prices in private/early information |
| Cross-Platform Arbitrage | Exploits price gaps between Polymarket and other platforms |
| High-Probability Bond | Buys near-certain outcomes (>88¢) for yield |
| Market Making | Posts bid/ask to capture spread, often automated |
| Domain Specialist | Deep expertise in one category (sports, politics…) |
| Speed / News Trading | Enters within minutes of breaking news |
Drop the polymarket-analyst-skills/ folder into your Claude skills directory:
~/.claude/skills/polymarket-analyst-skills/
Same as above — place the folder under ~/.claude/skills/ and Claude Code will
pick it up automatically on next startup.
Run the scripts directly:
# 1. Fetch data for a trader
python3 scripts/fetch_trader_data.py \
--address 0xYourAddressHere \
--output /tmp/poly_data/trader1
# 2. Analyze and get strategy attribution
python3 scripts/analyze_trader.py \
--data-dir /tmp/poly_data/trader1 \
--output /tmp/poly_data/trader1/analysis.json \
--print
# 3. Fetch the leaderboard
python3 scripts/fetch_trader_data.py \
--leaderboard \
--limit 100 \
--output /tmp/poly_leaderboardRequirements: Python 3.9+ with pip install requests python-dateutil
Analyze this Polymarket wallet: 0xabc123...
Who are the top 10 traders on Polymarket and what's their strategy?
Compare these two wallets and tell me which has better risk-adjusted returns.
Is this trader using insider information? Their win rate seems too high.
Find me Polymarket accounts that specialize in political markets.
polymarket-analyst-skills/
├── SKILL.md ← Main skill instructions (for Claude)
├── README.md ← This file (for humans)
├── references/
│ ├── api.md ← Full Polymarket API documentation
│ ├── metrics.md ← Definition of all 25+ metrics
│ └── strategies.md ← Strategy classification logic & thresholds
├── scripts/
│ ├── fetch_trader_data.py ← Fetches data from Polymarket APIs
│ ├── analyze_trader.py ← Computes metrics & produces analysis.json
│ └── requirements.txt ← Python dependencies (requests, python-dateutil)
└── evals/
└── evals.json ← 5 test cases for skill evaluation
| Source | URL | What it provides |
|---|---|---|
| Data API | data-api.polymarket.com |
Activity, positions, portfolio P&L |
| Gamma API | gamma-api.polymarket.com |
Market metadata, leaderboard |
| CLOB API | clob.polymarket.com |
Price history, order book data |
All data lives on Polygon (chain ID 137). You can verify everything on Polygonscan or Dune Analytics.
Performance: total P&L, win rate, profit factor, expectancy, ROI, largest win/loss, position size coefficient of variation.
Behavioral: avg entry probability, YES/NO ratio, hold time, repeat market rate, market concentration (Herfindahl index).
Timing: entry lag vs. market creation, hour-of-day distribution, day-of-week distribution, bot detection (chi-squared uniformity test).
Risk: max drawdown, approximate Sharpe ratio, max concurrent open positions, monthly P&L buckets.
Each of the 6 strategies receives a score (0–100) based on weighted signal combinations. The highest-scoring strategy becomes the primary label. A confidence level (High/Medium/Low) and any secondary strategies are also reported.
- This skill uses public on-chain data only. No private APIs or proprietary data.
- Strategy classification is probabilistic, not definitive. A high win rate alone does not prove insider trading.
- The skill includes a lucky variance warning for accounts with fewer than 30 closed positions.
- When the insider timing flag triggers, the report presents the observation without accusation — on-chain data cannot prove intent.
- Copy-trading based on this analysis carries significant risk. Past performance does not guarantee future results.
Pull requests are welcome. Key areas for improvement:
- Cross-platform arbitrage detection (requires Kalshi / Manifold API integration)
- News event correlation (requires external news API)
- Portfolio-level Kelly fraction estimation
- Multi-wallet portfolio consolidation (for traders using multiple addresses)
Please open an issue before starting major work.
MIT — see LICENSE
Polymarket is a decentralized prediction market on Polygon where users trade binary outcome contracts priced between $0 and $1. A $0.72 YES contract implies a 72% chance of the outcome. All trades settle in USDC via a hybrid Central Limit Order Book (CLOB): orders are matched off-chain, settled on-chain.
As of 2026, only ~7.6% of wallets are profitable, and the top 0.51% account for the vast majority of gains. The six strategies in this skill are derived from an analysis of 95 million on-chain transactions by Polymarket's own research team.