I keep a small watchlist of tickers I check most mornings, and for years my process was the same: scroll the chart, eyeball the trend, guess a stop, take the trade or talk myself out of it. No structure, no paper trail, no way to tell if a loss was bad luck or a bad process. A few weeks ago I rebuilt that entire routine as a Claude workflow instead, using Claude Fable 5, Anthropic's latest model, though the same six prompts work just as well with Sonnet or Opus. It does not place trades and it never will. What it does is turn "should I take this?" into a structured, evidence based decision I can actually review.
This is not financial advice, and Claude is not a broker. It is a decision support system: every setup still ends with a human clicking approve, review, or reject. That last part matters more than any of the prompts below, so keep it in mind as you build your own version.
Why a Single Prompt Was Never Going to Work
My first attempt was one long prompt: "Here's a chart, tell me if I should buy." It gave me an opinion, not an analysis. No memory of my risk rules, no pushback on my own bias, no record of why the trade made sense at the time. Writing a better prompt starts with the same formula I use for every Claude task, the Role, Context, Task, Format, Constraint structure, just applied six times instead of once. The fix was splitting the job into six focused stages, each with a narrow task and a clear handoff to the next one, the same way you would structure a multi agent Claude pipeline for any other business process. Each stage below is a prompt you can copy straight into Claude.
Stage 1: Scan the Market for Setups
Before I even look at a chart, Claude scans my watchlist for anything worth a closer look, weighing price action, trend, volatility, volume, and news context together instead of one signal at a time.
You are a market scanner. Given this watchlist: [list your tickers], and today's price action, identify the top 3 setups worth investigating further. For each one, score price movement, trend direction, volatility (range expansion or contraction), volume participation, and any recent news catalyst. Return a ranked list with one clear sentence explaining why each setup made the cut.
Stage 2: Read the Full Chart
For each setup that survives the scan, I drop in a chart screenshot and have Claude write a proper chart brief instead of a vague impression.
Here is a chart of [ticker] on the [timeframe] timeframe. Read the trend direction and strength, map the key support and resistance zones, and describe current momentum as accelerating, decelerating, or flat. Summarize this as a Chart Brief with four labeled sections: Trend, Support, Resistance, Momentum.
Stage 3: Build the Trade Plan
A setup is not a plan. This stage forces Claude to commit to specific numbers, not just a directional opinion.
Based on the Chart Brief above, build a Trade Plan. Define the entry zone, the target price with the structural or momentum reason behind it, the stop level, and the exact invalidation condition that would cancel this idea entirely. Include the risk to reward ratio.
Stage 4: Let the Risk Agent Block It
This is the stage that has saved me from the most trades I would have regretted. Claude checks the plan against my actual rules, not general best practice.
Act as my risk manager. Given this Trade Plan, an account size of $[X], and a maximum risk of [Y]% per trade, calculate the correct position size, confirm the stop distance is reasonable for current volatility, and check that this trade would not push my total portfolio exposure or maximum loss past my limits. Return PASS or FAIL for each check individually, then an overall verdict.
Stage 5: Challenge the Setup Before I Believe It
Claude is agreeable by default, which is exactly the problem when you want an honest second opinion. This prompt forces it to argue against itself.
Play devil's advocate on this Trade Plan. Give me the strongest bull case for taking it, the strongest bear case against it, and the single biggest unknown that could change the outcome. Be objective, not agreeable, and do not soften the bear case just because I already like this setup.
Stage 6: One Clear Decision Memo
Everything from the five prior stages gets folded into a single memo I can actually act on, or file away and learn from later.
Combine the Chart Brief, Trade Plan, Risk Gate result, and Bull/Bear Review into one Final Decision Memo: setup summary, chart evidence, risk level, entry plan, invalidation, and a confidence score out of 100%. End with exactly one label: Approved (execute), Review (needs more context), or Rejected (do not trade). State clearly that this is decision support only, not financial advice, and that I am the one who decides whether to act on it.
What Changed Once I Ran This for Two Weeks
The biggest shift was not accuracy, it was accountability. Every setup now has a paper trail: what Claude saw, what it flagged as risky, what the bear case was, and what I actually did about it. When a trade goes wrong, I can read back through the memo and tell whether the process failed or I just ignored good advice. That distinction alone is worth more than any single good call Claude has made.
If you want to build the same shape for a business process instead of trading, the pattern still holds: scan, analyze, plan, gate on risk, challenge the output, then produce one clear decision a human signs off on. I connect this same workflow to my own tools through the Claude API when I need it running outside the chat window, and it is just as useful outside the markets as it is inside them.
