Arbitrage套利分析

When to Use

User is evaluating an apparent price gap, hedge, surebet, basis trade, multi-leg basket, or cross-venue spread. Agent handles fee-aware arbitrage math, execution planning, settlement checks, and fast rejection of fake edge.

Architecture

Memory lives in ~/arbitrage/. If ~/arbitrage/ does not exist, run setup.md. See memory-template.md for structure.

CODEBLOCK0

Quick Reference

Topic	File
Setup guide	INLINECODE4
Memory template

memory-template.md | | Locked Spread Protocol | workflow.md | | Fee-aware formulas | calculator.md | | Venue and settlement checks | venue-checks.md | | Scenario playbooks | playbooks.md | | Safe language and disclaimers | legal.md |

Requirements

• - No credentials required
• No extra binaries required
• Live data only when the user provides it or explicitly asks you to fetch it

Locked Spread Protocol

Use the full protocol in workflow.md. At minimum, every opportunity passes these gates:

1. 1. Confirm the legs really describe the same economic outcome
2. Normalize prices, fees, financing, transfer, and timing costs with INLINECODE12
3. Decide whether the position is fully locked, soft-locked, or only expected value
4. Check venue rules, fill depth, and settlement mechanics with INLINECODE13
5. Classify the trade and explain the remaining failure modes before discussing size

Core Rules

1. Define the Exact Arbitrage

• - Write each leg with venue, instrument, side, price, size limit, timestamp, and settlement rule
• If the legs do not resolve to the same outcome, currency, or unit of risk, it is not arbitrage yet

2. Normalize Every Dollar of Friction

• - Always include fees, spread, gas, borrow, financing, FX, transfer, and any user-stated tax or settlement drag
• Use calculator.md to convert headline edge into net edge before calling anything profitable

3. Sequence for Fill Risk, Not for Hope

• - Identify the constrained leg first: shallow book, max bet, slow venue, borrow dependence, or promo cap
• Prefer structures that can be locked immediately; otherwise label them soft lock or directional exposure

4. Treat Settlement as First-Class Risk

• - Compare rules, expiry, void conditions, collateral type, withdrawal limits, transfer time, and counterparty exposure
• Use venue-checks.md and playbooks.md whenever the opportunity crosses venues or products

5. Reject Fake Edge Fast

• - Stale quotes, hidden min or max size, promo-only pricing, mismatched markets, and low depth are default failure modes
• If the residual risk is execution, model drift, or rule mismatch rather than a true lock, say so explicitly

6. Output a Decision Memo, Not Hype

• - Return the classification, net edge, max realistic size, required sequence, open checks, and kill conditions
• Good output ends with a clear next action: proceed, verify one item, downgrade to watchlist, or reject

7. Analysis Never Becomes Advice

• - Never call a trade risk-free, guaranteed, or suitable for the user personally
• Provide analysis and trade structure only; do not execute trades or give portfolio-specific recommendations

Opportunity Types

Type	Typical Setup	What Makes It Real
Cross-venue spot	Buy cheaper on venue A, sell higher on venue B	Enough depth, transferable inventory, fees covered
Surebet or matched book

Opposite sides across books or exchanges | Implied probability sum below 100 percent after fees | | Prediction market basket | Buy outcomes whose total cost is below guaranteed payout | Resolution rules and outcome partition match exactly | | Basis or carry | Spot versus future, perp, or synthetic hedge | Funding, carry, borrow, and expiry all modeled | | Retail or pricing mismatch | Same item or service priced differently by channel | Shipping, returns, fraud, tax, and inventory verified |

Arbitrage Traps

Trap	Why It Fails	Better Move
Matching names instead of resolution rules	Similar labels can settle differently	Compare exact payout and settlement language
Using top-of-book only

Apparent edge disappears after first fill | Calculate size against real depth | | Ignoring transfer time | Edge can vanish before hedge arrives | Price latency as a cost, not a note | | Forgetting limits and KYC | One leg fills, the other is capped or blocked | Check limits, region, and account status first | | Treating rebates as guaranteed | Rebates or promos can change net economics | Separate hard edge from conditional incentives | | Mixing currencies loosely | FX and spread can erase edge | Convert every leg into one base currency | | Calling EV arbitrage | Positive EV is not a locked spread | Label it expected value, not arbitrage | | Overstating size | The shallowest leg defines realistic capacity | Size to the weakest link |

Scope

This skill ONLY:

• - Analyzes arbitrage structures and net economics
• Stores user-stated constraints in INLINECODE17
• Uses its own auxiliary files for formulas, venue checks, and playbooks
• References timestamps and missing inputs when data may be stale

This skill NEVER:

• - Executes real trades, transfers, or withdrawals
• Uses exchange keys, broker credentials, or wallet secrets
• Calls an opportunity risk-free or guaranteed
• Gives personalized financial, tax, or legal advice

Security & Privacy

Data that leaves your machine:

• - None by default
• If the user explicitly asks for live public data, only the requested symbols, markets, or venue pages needed for the analysis

Data that stays local:

• - Preferences, venue notes, and working opportunity notes in INLINECODE18

This skill does NOT:

• - Store credentials
• Read unrelated files
• Make undeclared network requests

Related Skills

Install with clawhub install <slug> if user confirms:

• - trading - technical analysis, risk framing, and trade planning language
• INLINECODE21 - disciplined execution mindset and position management basics
• INLINECODE22 - pricing logic when the opportunity is a commercial mismatch rather than a market trade
• INLINECODE23 - long-horizon investing context when the user is mixing arbitrage with portfolio decisions

Feedback

• - If useful: INLINECODE24
• Stay updated: INLINECODE25