Most mornings I run a fixed options due-diligence routine against a shortlist of tickers: names I have thoughts on, a rumor, something a friend sent. An AI agent (Claude Code) pulls live prices from my broker, event statistics from an earnings service, and price history from a cloud research platform, then hands back a well-defined trade structureTrade structureThe specific combination of options to buy and sell, at named strike prices and a named expiration date. with entry and exit rules, a risk profileRisk profileA chart of profit and loss at every possible stock price, with the strikes and the breakeven marked., and a list of whatever could not be verified. I review the details and make the decision. This piece names every tool I use, walks two winning trades and one loss through the routine, and ends with how to build your own version.
01 · What a run looks likeOne line of input becomes a sized trade thesis in about twenty minutes.
The most expensive input in retail options trading is “sounds interesting.” The agent turns that into evidence before it turns into a position.
Here is the trigger prompt from one August morning:
That one line provided the key specifics: direction, risk budget and structure constraint. The rest is the agent’s job.
About twenty minutes later came a well-defined trade structure, with entry and exit rules, a risk profile, and a list of whatever could not be verified. I reviewed the details and made my decision.
Two rules never bend. Every position must be closed within two weeks of opening it, win or lose, so nothing becomes a long-term hold by accident. And the agent never places an order: it can only recommend, and I place any trade myself.
Sections 02 and 03 cover the tools and the steps they follow. Sections 04 to 06 walk two winning trades and one loss through those steps.
02 · The toolsEach platform answers one kind of question, and none of them is allowed to guess.
A broker cannot tell you what volatility did last year. A research platform cannot tell you what a contract costs right now. So each tool is trusted for exactly what it can actually see.
- Robinhood owns every live fact about the instrument itself: the current price, the full option chainOption chainThe full list of options available on a stock, every strike and every expiration, with their prices., the implied volatilityImplied volatilityThe price of insurance the market is charging on a stock, expressed as the size of move option buyers are paying for. and quote at every strikeStrikeThe price at which an option lets you buy or sell the stock., and the earnings date.
- Earnings Watcher, optional, owns what a broker cannot supply about earnings: how far a stock has jumped on past earnings days and how its option prices behave around a report. Not needed for the two trades in this piece.
- QuantConnect owns history and simulation: backtestsBacktestReplaying a trading rule over past prices to see how it would have done., and research notebooks that pull a year of prices and old option chains.
- Web search answers the one question no price feed can: why is it moving, and what could reverse it overnight. News is treated as a hypothesis, never a verdict.
- Barchart supplies IV rankIV rankWhere today’s implied volatility sits between this stock’s lowest and highest readings of the past year. A high rank means options are expensive by this name’s own standards.: are this stock’s options expensive by its own standards?
- Market Chameleon supplies a second, independent read of the same IV rank; if the two disagree across the decision line, a QuantConnect computation breaks the tie.
- Claude Code orchestrates all of it: the routine, the adversarial panelsAdversarial panelA simulated debate the agent runs among five trader personas whose job is to attack the current answer. It happens three times per run., the pricing, the chart. It never places an order.
03 · The routineEvery run moves through four phases: gather, test, decide, commit.
The sequence never changes. What changes is how far down it the evidence lets a trade get.
The real work is the volatility surfaceVolatility surfaceThe whole map of implied volatility across every strike and expiration, read for where the insurance is dearest and cheapest.: term structureTerm structureHow option prices compare across different expiration dates on the same stock., skewSkewHow option prices compare across different strike prices for the same expiration. With term structure, it is how you find the single most overpriced option on the board., IV rank against the name’s own history, realizedRealized volatilityHow much the stock has actually moved. Implied above realized means the insurance looks expensive relative to the risk. versus implied volatility. That read decides whether the agent should recommend buying or selling premiumPremiumThe price of an option. Selling premium means being the insurer: you take money up front (a credit) and keep it if the stock behaves. Buying premium means being the insured: you pay (a debit) and need a move to profit. before any structure is considered.
Three times in every run the agent convenes a simulated panel of five trader personas (momentum trader, volatility quant, risk engineer, event trader, skeptic) whose only job is to argue against the current answer: what still needs measuring, which structure to pick, whether the case holds up.
Conviction is a size, not a verdict. Only two things kill a trade outright: no expirationExpirationThe date an option stops existing and settles at whatever it is worth. date fits the idea, or the risk cannot be capped and the legs cannot actually be filledFilledActually getting an order executed at a sensible price. A strike with no buyers and sellers cannot be filled, so any edge there is imaginary.. Everything else is scored 0, 1 or 2 on six questions.
The total, out of a maximum of 12, sets the position size. A score of 9 or more takes the full risk budget; 6 to 8 takes half; 4 or 5 takes a quarter, as a starter position; 3 or under means no trade.
The last step is a checkpoint: the payoff chart is not drawn until all three panels agree the trade is worth taking, because drawing it earlier would dress up an idea that has not earned the confidence. The scoring thresholds are admittedly guesses, so every run, no-trades included, appends a row to a log that will eventually say whether they were good ones.
04 · Case A, ServiceNow$NOW: QuantConnect confirmed the options were expensive, so the routine sold premium instead of buying it.
ServiceNow ($NOW), early July, after a 34% monthly fall, with no earnings report due before the trade’s expiration. The whole question was whether its options were expensive by its own standards.
Robinhood gave the live picture: a share price near $104.72 and implied volatility around 57 to 64%. Barchart put that at the 92nd percentile of the trailing year. But one vendor’s percentile is a claim, not a measurement, and a broker cannot answer a history question. So the agent stood up a QuantConnect project and wrote a notebook with a one-sentence job: is $NOW’s implied volatility rich, fair, or cheap versus its own trailing year?
The lead structure was a bull put spreadBull put spreadSell a put at one strike and buy a cheaper put below it. You collect the difference up front and keep it if the stock stays above the sold strike; the bought put caps your loss., sold at $100 and bought at $98, for a $0.55 credit against $145 of risk, with one exit rule: a close below $100. It never fired, and the spread expired worthless on July 10 for the full $55. The secondary call debit spreadCall debit spreadBuy a call at one strike and sell a cheaper call above it. You pay the difference and profit if the stock rises past breakeven; the sold call caps the gain and lowers the cost. also finished in the money, but its modeled entry was $0.43 better than the natural priceNatural priceWhat you actually pay when you cross the market in a hurry: buying at the ask and selling at the bid. It is worse than the mid, the midpoint between the two. and it carried no exit rule, so the process cannot take credit for that 59%.
05 · Case B, United States Oil Fund$USO: The evidence argued against this trade, and the exit rule banked it anyway.
A crude-oil ETFETFA fund that trades like a stock. $USO holds oil futures contracts, so it has no earnings reports. up 13% in nine sessions, a bullish hunch from me, and a history that said the hunch was wrong.
A web search established the driver first: a genuine supply shock, with active US-Iran diplomacy running alongside it. That read is why an overnight peace-deal headline became the panel’s first caveat, since no stop protects against a gap. Then the agent measured what my hunch was implicitly betting on, the base rateBase rateWhat has actually happened in the past after setups that look like this one.. After every prior burst of +10% in nine sessions, $USO’s median return over the next eight sessions was −4.3% across eleven such setups, and −5.9% in a six-setup holdoutHoldoutA second, older slice of history kept aside to check that a pattern found in recent data is not a fluke. from 2022 to 2024. Only 18 to 33% of them cleared a move the size of this trade’s breakevenBreakevenThe stock price at which the trade makes nothing. Past it, profit.. On “does the history support the direction” the candidate scored zero.
Four of five candidate structures were rejected, including my own idea of selling puts, which would have meant selling insurance priced at 44% on a stock actually moving 51 to 66%.
What survived was a Tier CTierThe size bracket the conviction score assigns: A takes the full risk budget, B half, C a quarter, as a starter position. starter, one contract, with three exit rules. On the exit date the spread was worth $2.78, a gain of $145 on $133 at risk. The analysis said the trade was unlikely to work, the sizing kept it small, and the exit banked what showed up. A discipline result, not a forecasting one.
06 · Case C, CoreWeave$CRWV: Ignoring one exit rule turned a $48 loss into a $145 one.
The same agent, six weeks before the $USO run, on a name where the exit rule was ignored.
CoreWeave ($CRWV), July 8: a bull put spread, sold at $78 and bought at $76, for a $0.55 credit against $145 of risk. The chart called it a bet that a fear-driven selloff would hold the $80 support, not a volatility trade, and named the live risk it was taking: competition-from-Meta headlines, which the steep put skew was already pricing. It wrote the exit in advance: a daily close below about $79 kills the thesis; close the spread, do not hope. On July 15 it closed at 77.12. Obeyed, the exit cost $48. Held to expiry instead, it lost the full $145, after briefly showing a $25 profit a week past the stop, which is why “it was up at one point” is not evidence of anything.
The caveats that apply to all three
No order was placed; every price here is a market quote, not a fill.
Entry prices are modeled at the midMidThe midpoint between the best bid and the best ask. A common modeling price, and usually better than what you actually get filled at., and the natural price is worse. Of the four entries here, two held, one was better than modeled, one was worse.
$NOW’s call debit had no exit and won by expiring in the money. The notebook priced 104/108; the trade was 104/109.
$USO’s IV rank came from one vendor only, and no historical test of the exact spread was run.
07 · Build your ownYou can rebuild this with any broker, and six habits matter more than the tools do.
You need one source of live option prices you trust, one way to ask whether a name’s implied volatility is high or low against its own history, and somewhere to write your rules down before you trade.
- 01Write the routine before you automate it. One page: what you check, in what order, and what would make you pass. Mine is the four phases in Figure 2. Run it by hand for a month before any code touches it.
- 02Give every platform one job. A broker for live prices, a vendor or a notebook for IV history, and nothing fills a gap from memory. When two sources disagree, write the disagreement down instead of averaging it away.
- 03Treat your favourite structure as the thing to disprove. If you always reach for calls, make the routine argue for the credit spread first. Buy calls only when the short-premium case has lost on evidence.
- 04Score conviction, then size from the score. A thin edge is a small position, not a pass. Only an impossibility (no expiration fits, the loss cannot be capped, or a leg cannot be filled) gets a veto.
- 05Write the exit before the entry, then obey it. A price that proves you wrong, a profit target, and a date. In this piece that single habit was worth $97 on a $145 position: the difference between $USO and $CRWV.
- 06Log every run, including the no-trades. Your thresholds are guesses until the log says otherwise.
The agent is not a forecasting machine, and I do not use it as one. What it does is make every decision auditable: I can open any morning’s chart and see what was known, what was assumed, what the exit was, and whether I obeyed it. That is the part worth copying. An agent will now produce a confident thesis for any ticker you hand it; generation is cheap. The value is in the referee: the written threshold, the exit rule, and the human who still has to say yes. Build that first. The tools can come later.
Research, not financial advice. Options can lose most of their value fast even when the direction is right. Trade only risk capital. Platform names and marks belong to their owners; no affiliation or endorsement is implied.