Draft an alert-only workflow for my watchlist. Notify me when a stock moves more than 3% from its previous close. Include the ticker, the move and the time. One alert per ticker per day. Show me the workflow before activation. Do not place trades.
It watches. It never trades. Change the 3% to whatever actually gets your attention.
Buy $500 of BTC over the next 4 hours in equal slices. Max slippage 0.3%. If the spread goes above 0.2%, pause and ask me. Show me the plan before you execute.
Prompt makeover. The vague version was: "Trade BTC for me and make money."
For each ticker in [list], give me the overnight news in one line, the next earnings date, and one risk to watch. Link every source. No price predictions.
Swap [list] for your tickers. Ask it to keep the same format every morning.
Summarise the three biggest market-moving stories since yesterday's close. One line each, with the source link. No forecasts.
Prompt makeover. The vague version was: "What's happening in the market?"
Before you do anything: max position size is [amount]. Do not open new positions I haven't named. If the price moves more than [X]% against the plan, the spread widens past [Y]%, or anything is unclear, pause and ask me. Show me the full plan and wait for my go-ahead before you act.
Fill in the three blanks, then paste this ahead of whatever you actually want the agent to do.
Here is an earnings call transcript. Tell me: 1) how guidance changed versus last quarter, 2) the three things analysts pushed hardest on, 3) any question management didn't really answer. Quote the exact lines. No opinion on the stock.
Works with the transcript pasted in or attached as a file. Keep the same three headings every quarter so you can compare.
Here is my thesis: [paste thesis]. Give me the strongest case against it in five bullets, then tell me the one piece of evidence that would prove me wrong. Do not tell me whether to buy or sell.
Use it before you size up, not after. If you can't answer the fifth bullet, that's the point.
Attached is a CSV of my trades with date, ticker, side, size, entry, exit and a setup tag. Group by setup. For each one show count, win rate, average hold time and the largest loss. Then list every trade where I broke one of these rules: [your rules]. Tables only, no advice.
Export the fills from your broker. Add a setup column by hand before you upload; that's what makes the output useful.
Attached are this year's and last year's 10-K for [company]. Compare the risk factors and outlook sections. List what was added, what was removed and what was reworded, with the old and new sentence side by side. Skip boilerplate that didn't change.
Also works on two 10-Qs or two proxy statements. The reworded list is usually where the story is.
I hold [positions]. List only this week's economic releases, earnings and scheduled events that could move those names. For each: day, time in ET, consensus, prior and one line on why it matters to my book. Link every source.
Run it Sunday night. Swap in your positions each week and keep the format identical.
Build a strategy that holds [tickers] equal weight, rebalances monthly, and moves fully to cash when SPY closes below its 200-day average. Backtest it over the last five years and show me total return, max drawdown and the number of trades before anything goes live.
The point is the drawdown number, not the return. Change the tickers and the exit rule to match what you'd actually hold.
Attached are two files: [system A export] and [system B export]. Match the rows on [reference field] and amount. Give me three tables: matched, in A but not B, in B but not A. For each unmatched row, suggest the most likely reason (timing, rounding, duplicate, missing) in a few words.
Use sample or anonymised data. Good for bank recs, intercompany and vendor statements.
Write variance commentary for this actual-versus-budget table. Five bullets, biggest variances first, each with the amount, the percentage and the likely driver. Flag anything over [threshold]% for review. Match this tone: [paste one bullet from last month].
Pasting last month's bullet is what stops it sounding like a robot wrote it.
Turn the attached analysis into a one-page memo. Put the decision you need from the reader in the first line. Then: context in three sentences, the numbers that matter in one table, risks in three bullets, and the recommendation. Keep everything that's in my notes and add nothing that isn't.
Works for IC memos, budget asks and board pre-reads. The last sentence stops it inventing numbers.
These examples aren't tested configurations. Review before activating.