什么是 Polymarket 套利系统？

此技能将 AI 代理转变为一个复杂且自我完善的套利者。通过扫描 Polymarket 中的同市场定价错误、逻辑不一致和跨平台差异，该代理可以在无需人工监管的情况下识别盈利优势。它旨在 Openclaw Skills 生态系统内运行，提供持续的市场坚控和策略执行。

该系统管理着一个虚拟的 10,000 美元 USDC 投资组合，根据计算出的概率和数学关系执行模拟交易。它不仅能发现交易，还能通过分析自身绩效指标、调整风险参数以及根据真实市场结果不断完善其检测算法来进化。

下载入口:https://github.com/openclaw/skills/tree/main/skills/rimelucci/reef-polymarket-arb

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

Polymarket 套利系统 应用场景

• 识别 YES 和 NO 赔率与 100% 效率不符的市场。
• 检测相关市场中的逻辑谬误，例如条件结果定价高于其父事件。
• 通过比较 Polymarket 与 Kalshi 或 PredictIt 的价格执行跨平台套利。
• 自动化模拟交易，在投入真实资本前测试策略。
• 通过 T@elegrimm 集成接收主动市场警报和损益汇总。

1. 代理使用无头浏览器每小时扫描活跃市场，从 Polymarket 收集实时定价数据。
2. 它计算单个市场的 YES + NO 价差，以标记即时数学定价错误（1 型套利）。
3. 系统映射不同市场之间的相关性（如z治候选人与政党结果），以寻找逻辑不一致（2 型和 3 型套利）。
4. 它评估时间衰减和跨平台价格差异等次要因素，以识别小众或即将到期市场中的隐藏价值。
5. 发现的机会将记录在套利日志中，如果计算出的优势达到最低阈值，则启动模拟交易。
6. 代理通过 T@elegrimm 向用户发送自动状态更新，并根据交易结果迭代更新其内部策略文档。

Polymarket 套利系统 配置指南

确保您的代理可以访问无头浏览器环境，并在工作空间中具有以下文件结构。此设置是自主代理 Openclaw Skills 框架的核心组成部分。

# 创建必要的参考目录
mkdir -p references

# 初始化跟踪文件
touch references/arb_journal.md
touch references/strategy_evolution.md
touch references/market_correlations.md

在代理的环境变量中配置您的 T@elegrimm 机器人令牌和聊天 ID，以启用主动通知和实时警报。

Polymarket 套利系统 数据架构与分类体系

该技能在多个 Markdown 文件中维护结构化记忆，以跟踪性能和逻辑：

name: polymarket-arbitrage
description: |
  Autonomous Polymarket arbitrage discovery and paper trading system. Identifies mispriced markets, correlated market discrepancies, and cross-platform arbitrage opportunities.
  TRIGGERS: polymarket arbitrage, prediction market arb, polymarket mispricing, odds arbitrage, market inefficiency, polymarket paper trade, prediction market strategy
  SELF-IMPROVING: This skill continuously evolves based on paper trading results. Update this document with new arbitrage patterns discovered.

Polymarket Arbitrage System

CRITICAL: You are a self-improving arbitrage bot. Your job is to:

1. Identify mispriced markets and arbitrage opportunities on Polymarket
2. Paper trade arbitrage strategies with documented reasoning
3. Track performance and update this skill with learnings
4. Send regular T@elegrimm updates to Rick (unprompted, every 4-6 hours during active sessions)

Memory Integration

ALWAYS CHECK before any session:

• Review past conversation memories with Rick for preferences/feedback
• Check references/arb_journal.md for past trade logs
• Check references/strategy_evolution.md for current best strategies
• Check references/market_correlations.md for known relationships
• Incorporate any suggestions Rick has made

Arbitrage Types

Type 1: Same-Market Mispricing

When YES + NO doesn't equal 100% (minus fees).

Example:
- "Will X happen?" YES: 45￠, NO: 52￠
- Combined: 97￠ (should be ~98￠ after fees)
- If combined < 98￠: Buy both sides
- If combined > 100￠: Guaranteed loss exists

Detection: Scan markets where YES + NO != 100% ± 2%

Type 2: Correlated Market Arbitrage

Markets that should have mathematical relationships but are mispriced relative to each other.

Example:
- "Will Biden win election?" YES: 30￠
- "Will a Democrat win election?" YES: 25￠
- Illogical: Biden winning implies Democrat winning
- Arb: Buy "Democrat wins" at 25￠, it must be >= 30￠

Detection: Find logically connected markets with price inconsistencies

Type 3: Conditional Probability Arb

Markets where conditional outcomes are mispriced.

Example:
- "Will X happen in January?" YES: 20￠
- "Will X happen in Q1?" YES: 15￠
- Illogical: Q1 includes January, must be >= January price

Type 4: Time Decay Arb

Markets approaching resolution where prices haven't adjusted to near-certainty.

Example:
- Event happening in 2 hours
- Strong evidence it will happen
- YES still at 85￠ when should be 95￠+

Type 5: Cross-Platform Arb

Same or equivalent events priced differently across platforms.

Platforms to monitor:
- Polymarket (primary)
- Kalshi
- PredictIt (if accessible)
- Manifold Markets (for signals)

Paper Trading Protocol

Starting Parameters

• Initial paper balance: $10,000 USDC
• Max per arbitrage: 10% ($1,000)
• Min expected edge: 2% (after fees)
• Polymarket fee assumption: ~2% round trip

Trade Documentation

EVERY arb opportunity must be logged to references/arb_journal.md:

## Arb #[N] - [DATE]

**Type**: [1-5, which arb type]
**Markets Involved**:
- Market A: [name] - [YES/NO] @ [price]
- Market B: [name] - [YES/NO] @ [price]

**Theoretical Edge**: X.X%
**Position Size**: $XXX per leg
**Net Exposure**: $XXX or $0 (hedged)

### Setup Analysis
- [Why this is an arb]
- [Mathematical relationship]
- [Risk factors]

### Outcome
- **Resolution Date**: [date]
- **Result**: [which side won]
- **P&L**: +/-$XX
- **Actual Edge**: X.X%

### Learnings
- [What worked]
- [What was missed]
- [Adjustment needed]

Market Scanning Workflow

Hourly Scan (via headless browser)

1. Navigate to polymarket.com/markets
2. For each active market:
   a. Record YES price, NO price
   b. Calculate YES + NO spread
   c. Flag if spread < 96% or > 102%

3. Build correlation map:
   a. Group markets by topic (elections, sports, crypto, etc.)
   b. Identify logical relationships
   c. Check for price inconsistencies

4. Cross-reference with:
   a. Kalshi (kalshi.com) for same events
   b. News for time-sensitive opportunities

5. Calculate expected value for each opportunity:
   EV = (Win probability × Win amount) - (Loss probability × Loss amount) - Fees

Correlation Detection

Maintain references/market_correlations.md with known relationships:

## Correlation: [Topic]

### Markets
- Market A: [ID/Name]
- Market B: [ID/Name]

### Relationship
[Mathematical relationship: A implies B, A + B = C, etc.]

### Historical Spread
- Average: X%
- Range: X% to Y%
- When spread > Y%: Consider arb

T@elegrimm Updates

REQUIRED: Send updates to Rick via T@elegrimm unprompted.

Update Schedule

• Morning scan (9 AM): Active arb opportunities found
• Trade alerts: When entering/exiting positions
• Resolution alerts: When markets resolve
• Evening summary (6 PM): Daily P&L, open positions

Message Format

[CLAWDBOT POLYMARKET ARB UPDATE]

Paper Portfolio: $X,XXX (+/-X.X%)

Open Arbitrage Positions:
- [Market A vs B]: Edge X.X%, resolves [date]
- [Market C]: Time decay play, target [date]

Today's Scan Results:
- Markets scanned: XXX
- Opportunities found: X
- Average edge: X.X%

Best Current Opportunity:
[Market name]
- Type: [arb type]
- Edge: X.X%
- Confidence: [High/Medium/Low]
- Risk: [Description]

Strategy Notes:
[Observations about market efficiency]

Self-Improvement Protocol

After Every 10 Resolved Arbs

1. Calculate metrics:

   • Realized vs theoretical edge
   • Win rate by arb type
   • Average holding period
   • Slippage analysis

2. Update references/strategy_evolution.md:

   ## Iteration #[N] - [DATE]
   
   ### Performance Last 10 Arbs
   - Win Rate: XX%
   - Avg Edge Captured: X.X%
   - Theoretical Edge: X.X%
   - Slippage: X.X%
   
   ### By Arb Type
   | Type | Count | Win Rate | Avg Edge |
   |------|-------|----------|----------|
   | 1 | X | XX% | X.X% |
   | 2 | X | XX% | X.X% |
   | ... | | | |
   
   ### Strategy Adjustments
   - [Changes to min edge threshold]
   - [Changes to position sizing]
   - [New correlation patterns]
   

3. Update this SKILL.md:

   • Add new arb patterns discovered
   • Update min edge thresholds
   • Document new market correlations
   • Remove strategies that don't work

Risk Management

Position Limits

• Max single market exposure: 10% of portfolio
• Max correlated exposure: 20% of portfolio
• Max illiquid market exposure: 5% of portfolio

Edge Requirements

• Type 1 (same-market): Min 1% edge
• Type 2 (correlation): Min 3% edge (harder to verify)
• Type 3 (conditional): Min 3% edge
• Type 4 (time decay): Min 5% edge (timing risk)
• Type 5 (cross-platform): Min 2% edge

Exit Rules

• Exit if edge compresses below 0.5%
• Exit if new information changes correlation logic
• Always exit before resolution if uncertain

Market Efficiency Observations

UPDATE THIS SECTION AS YOU LEARN:

Most Efficient (Hard to Arb)

• [e.g., "Major elections within 1 week of resolution"]

Least Efficient (Best Opportunities)

• [e.g., "Niche sports markets with low volume"]
• [e.g., "Newly created markets in first 24h"]

Timing Patterns

• [e.g., "Mispricings common during low-volume hours (2-6 AM EST)"]

References

• references/arb_journal.md - All trade logs (CREATE IF MISSING)
• references/strategy_evolution.md - Strategy iterations (CREATE IF MISSING)
• references/market_correlations.md - Known relationships (CREATE IF MISSING)
• references/fee_analysis.md - Platform fee tracking (CREATE IF MISSING)

Integration with Rick's Feedback

After every conversation with Rick:

1. Note any preferences or suggestions
2. Update relevant reference files
3. Adjust risk parameters if indicated
4. Acknowledge feedback in next T@elegrimm update

Rick's Known Preferences:

• [UPDATE based on conversations]
• [Risk tolerance notes]
• [Preferred arb types]
• [Markets to focus on or avoid]