How to use the Perpetual Funding & Basis Trading agent skill

What does the perp-funding-basis skill make the agent do?

Perpetual futures have no expiry date. Instead, a funding rate is exchanged between longs and shorts every 8 hours on most exchanges, which keeps the perpetual price anchored to the spot price. When the perp trades above spot, longs pay shorts; when it trades below, shorts pay longs.

The skill teaches the agent to convert funding rates into annualized figures, classify funding into market states, compute annualized basis between spot and dated futures, and lay out a delta-neutral cash-carry trade. It also covers funding spreads between exchanges and divergence between price and funding.

• Annualize a per-8-hour funding rate (multiply by 3 windows per day and 365 days)
• Classify funding into regimes such as overheated_long, bullish_carry, neutral and bearish_carry
• Compute annualized basis for a dated future from its price, the spot price and days to expiry
• Describe how a cash-carry trade is built and what can go wrong
• Compare funding rates across exchanges for the same asset

How do you install the perp-funding-basis skill?

Download the skill ZIP from the directory page where you found this guide. The skill's only listed file is SKILL.md, so the unzipped folder should contain that file. Agent skills are read from a skills folder, and you can install them at two levels.

For a personal install that applies across your projects, unzip the folder into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP
• Check that the skill folder contains SKILL.md after unzipping

How do you trigger the skill with a natural request?

You do not need a special command. Describe the task in plain language that mentions funding rates, basis, carry trades or perpetual futures, and the agent can match the request to the skill's description. Naming the exchange, the asset and the time window in your request gives the agent more to work with.

The skill does not fetch market data. Paste the funding rates, prices or open interest numbers into the conversation, or tell the agent where you got them, so it can apply the formulas to real figures.

• "Annualize a funding rate of 0.0002 per 8 hours and tell me what regime that is."
• "BTC spot is 65,000 and the quarterly future is 66,500 with 45 days to expiry. What is the annualized basis?"
• "Walk me through building a delta-neutral cash-carry trade on OKX and list the risks."
• "Here are the last 7 days of BTC-USDT funding rates from three exchanges. Where is the cross-exchange spread?"

What examples does the skill contain?

The skill includes worked calculations you can check by hand. The cash-carry example uses a $100,000 position with an average funding rate of +0.015% per 8 hours over 30 days, which the file calculates as $1,350 of revenue, or 16.4% annualized.

The basis example uses a BTC spot price of $65,000 and a quarterly future at $66,500 with 45 days to expiry, giving a 2.31% basis and 18.7% annualized. Working through these yourself is a good way to confirm the agent is applying the formulas correctly.

What are the safety notes and risks?

The skill runs no scripts and uses no network access, according to its listing. The Python in the file is formula and example logic that the agent reads and applies. It does not place orders, connect to an exchange or store credentials.

The file itself lists major risks of the cash-carry trade. The short perpetual leg can be liquidated if margin is insufficient, and the file suggests using 3-5x maximum leverage. Funding can flip negative and turn the carry into a cost. Holding positions across two or three exchanges reduces counterparty risk. The basis can widen before it reverts, which causes mark-to-market losses on the short leg.

• No scripts run and no network calls are made by the skill itself
• Liquidation risk on the short perpetual leg if margin is thin
• Funding can turn negative, so the carry becomes a cost
• Exchange counterparty risk, which the file suggests reducing by spreading positions
• Signal thresholds are heuristic rather than proven rules

What tips help you get reliable results?

Give the agent the units explicitly. The skill treats funding as a per-8-hour decimal, so 0.0001 means 0.01%. The tables show percentages, so divide a table's percentage by 100 before comparing it with code output. Mixing these up is the most common source of wrong numbers.

The thresholds are not identical everywhere. The overheated-long level in the signal table is above +0.05%, while the regime code uses +0.03% per 8 hours. Ask the agent which threshold it used and state the one you want. The historical BTC probabilities in the file are given without method or data, so treat them as claims to verify, not as tested results.

• Always state whether a funding rate is per 8 hours or annualized
• Divide table percentages by 100 before feeding them into code
• Ask the agent which threshold set it used, since the table and code differ
• Verify the historical probability figures before relying on them