Someone made $22 million in 31 days on Polymarket. Another lost $7.2 million total but flipped $550k in just one week. And someone quietly earned $540k in BTC threshold markets.
This data is all public. But Polymarket only shows you leaderboards and transaction flows — you can't see what strategies these people are actually running.
I built a tool that takes an address and has AI automatically profile it for you.
What is Polymarket Toolkit
An open-source AI Agent Skill. Works with Claude Code, OpenClaw, Cursor, or any LLM that can run shell commands.
Give it a Polymarket address, get a complete trading profile in seconds.
No API keys, no database, no configuration. Public APIs + AI Agent handles the entire workflow.
GitHub: https://github.com/runesleo/polymarket-toolkit
What You Get
Input: Any 0x address, username, or profile link.
Output:
- PnL overview (total PnL + 7-day/30-day trends)
- Win rate (based on settlement data, not estimates)
- Current positions + unrealized PnL
- Trading behavior breakdown (TRADE / SPLIT / MERGE / REDEEM counts)
- Category distribution (Crypto, Politics, Sports, Weather...)
- Top 5 biggest wins/losses
- Automatic strategy pattern recognition (market maker, SPLIT arbitrage, whale concentration, diversified investing...)
Real Test: Top Performers Across 9 Categories This Week
I pulled the #1 weekly PnL performer from each of Polymarket's 9 categories:
Just looking at this table reveals patterns. The crypto category leader has volume 114x their PnL — clearly high-frequency trading. The tech category leader only made $660 on $178 volume, suggesting extremely low liquidity in tech markets. Overall and sports are the same person.
But this is just surface data. Drop these addresses into the profiling tool and you see the deeper story.
Deep Dive: Two Case Studies
Politics #1: How.Dare.You
92,398 trading records (tool automatically paginated through 185 pages), 1,181 markets, 48k buys + 43k sells. Trades every political market, heavy on Ukraine ceasefire and Trump-related bets. 81% win rate, but total PnL is negative (-$59k). Wins often but loses hard when wrong — risk management didn't keep up.
Culture #1: cynical.reason
Tool automatically identified as SPLIT arbitrage type. $929k SPLIT + $1.12M MERGE across 91 markets, heavy on Fed chair nominations and Iran situation. Not betting on direction — using SPLIT to create YES+NO pairs, sell one side, wait for settlement or MERGE the other back. This operation is invisible in raw transaction records, but the profiling tool makes it obvious immediately. Total PnL +$210k.
Same leaderboard, completely different money-making methods. One relies on broad diversification with high win rate (but poor risk control), the other makes quiet money through SPLIT arbitrage. The leaderboard tells you who made money, the profiling tool tells you how they made it.
How It Works
The Skill orchestrates 4 public APIs through AI Agent:
- lb-api — PnL and leaderboard data
- data-api — Positions and complete trading history (auto-pagination)
- gamma-api — Market metadata and category mapping
- CLOB API — Real-time orderbook prices
7-step execution flow:
- Pull PnL snapshot (totals + 7-day/30-day trends)
- Pull current positions, calculate win rate
- Paginate through complete trading history (handles 10K+ records)
- Map market categories via Gamma API
- Analyze trading behavior, identify strategy patterns
- Sort biggest wins/losses
- Assemble complete profile output
Username auto-resolution — Agent searches leaderboards to convert usernames to addresses, no manual lookup needed.
Why a Skill, Not an App?
Traditional tools need frontend, servers, databases. A Skill is a Markdown file with clear instructions that AI Agents read and execute themselves. Zero infrastructure, any LLM that can run curl can use it, modify however you want.
The entire tool is just a SKILL.md file. That's it.
Roadmap
This is v0.1 — address profiling. The toolkit will continuously expand:
Analysis Tools
- PnL Calculator — Position-level profit breakdown
- Brier Score Rating — Address prediction quality scoring
- Trading Style Tags — Conservative/aggressive/event-driven classification
Market Intelligence
- Market Scanner — Discover markets by category, volume, liquidity
- Liquidity Dashboard — Spread, depth, market maker concentration analysis
- LP Incentive Scanner — Find active liquidity incentive programs
Tracking & Alerts
- Leaderboard Tracker — Daily snapshots, ranking changes, streak detection
- Whale Movement — Large player position change alerts
API (Planned)
- REST API for all tools — Integrate into your own applications
Try It Out
Installation Method 1: Just tell your AI
Drop this article or the GitHub link to your AI (Claude Code, ChatGPT, Cursor all work), say "help me install this Polymarket profiling tool" and it'll handle everything automatically.
Installation Method 2: Manual two steps
git clone https://github.com/runesleo/polymarket-toolkit.git
cp -R polymarket-toolkit/skills/polymarket-profile ~/.claude/skills/
After installation, ask your AI: "Help me analyze this Polymarket address: 0x63ce342161250d705dc0b16df89036c8e5f9ba9a"
GitHub: https://github.com/runesleo/polymarket-toolkit
Feedback & Contributing
This is v0.1, there are definitely calculations that miss or cases we don't cover. If you find issues while using it, please give feedback anytime:
- GitHub Issues: https://github.com/runesleo/polymarket-toolkit/issues
- X DM: @runes_leo
PRs are also welcome. This toolkit is open source, and if you're interested in any features on the roadmap, we can build them together.
About the author: Leo (@runes_leo), independent builder at the intersection of AI and Crypto. Trading prediction markets on Polymarket with AI agents, open-sourcing the tools along the way. More: leolabs.me