The Claude Code prompt that catches Board Director signals...

In May 2024, a director at eBay decided to take his board fees in stock instead of cash.

The director was Logan Green, the co-founder of Lyft, who has sat on eBay’s board since 2016. He did not buy a single share on the open market. He just elected to receive his retainer as fully vested eBay stock rather than a check.

A year later, eBay was up 40% (vs S&P 500 +12%)

So choosing stock instead is a deliberate trade. He would rather own eBay than hold the cash..... Did he know something the rest of the market didn't?

And a bigger question for us as investors.... Can we use AI to screen for these cash-for-equity elections by board directors systematically?

That's what we are answering in this week's newsletter.

The challenge is that not every election is a bullish signal. Taking stock over cash is not, by itself, a buy. So how can we create a filtering mechanism using AI to find the cash-for-equity conversions that are actually worth our time studying?

And here is the catch that makes it hard. None of this shows up on your insider feed. There is no open-market buy, so nothing triggers. The only record is a footnote on a Form 4, and every insider-data vendor I have used strips footnotes out before you see them.

I wanted to make it systematic using the last 30 days as an example on how you would screen for this. Here is the plan:

1. Pull every Form 4 that mentions taking pay in stock
2. Read the footnote, not the data fields
3. Throw out the look-alikes that fool a naive screen
4. Rank what survives and watch it from here

The Prompt

The prompt has to do two jobs. First find the filings. Then prove they are real, because most are not.

I built the logic around five decisions, and each one exists to kill a specific kind of fake.

1. Search the text, not a database. There is no “took stock instead of cash” field anywhere. The election exists only as written English in a footnote. So the prompt starts with EDGAR full-text search, not a data feed. That is the same reason your vendor misses it.

2. Run several phrases, not one. Directors word it differently. Some write “in lieu of cash,” others “in lieu of director fees” or “elected to receive.” One phrase misses half the filings, so the prompt searches a small bundle and dedupes the results.

3. Open the filing and read the footnote. The search only tells you which Form 4 to look at. Whether the election is real sits in the footnote text inside the raw filing, the part the convenience tools strip out. So step two is to go read it.

4. Make it prove the election before it counts. This is the core of the whole thing. A filing survives only if the footnote explicitly says equity instead of cash, the shares were acquired and not sold, it settles in real stock rather than “payable only in cash,” and it is not a Rule 10b5-1 auto-trade. Each of those clauses deletes one look-alike: routine grants, sales, phantom units, automatic plans.

5. Treat a board sweep as one event. When a whole board elects on the same day under a standing plan, the prompt collapses it into a single line. Eight directors on one date is one decision, not eight signals.

That is the logic. Here is the prompt I used Claude Code as the foundation before iterating on the outputs. CashtoEquityInsiderPrompt.docx​

The Result

I ran it on the last 30 days. Full screen and the dossier are here.

• Screen: Cash_to_Equity_Insider_Screen.xlsx​
• Dossier: Cash_to_Equity_Insider_Screen_report.pdf

The wide net pulled 986 filings. After the filters, 228 were genuine cash-to-equity elections. The other 758 were noise.

Most of what gets thrown out falls into three buckets:

• It is just the standard board grant. Directors get paid partly in equity every year by policy. That is the pay plan, not a view on the stock.
• Some of the “equity” pays out in cash. These are phantom units that look like stock and settle like a paycheck. 86 of them were this.
• The clusters are one decision, not many. When a whole board elects on the same day under a standing plan, that is one boardroom vote. 156 of the 228 were exactly this kind of same-day sweep.

The difference is right there in the text once you read it.

A real one, from a recent filing:

A fake one, the kind that fools a naive screen:

Same “equity” label on the filing. One is ownership. The other is a deferred paycheck.

Strip all of it out and what is left is a short list. Here is some of what cleared the filter this month.

A $CTSH (Cognizant) director took fully vested stock for 100 percent of his retainer, with the stock down 29 percent over the prior three months. A $BGS (B&G Foods) director took $165,000 of his fees in equity, with that one down 21 percent. And at $IMNN (Imunon), six insiders took pay in stock while the shares sat down about 38 percent.

The biggest by dollar value, roughly $3.5 million at $DTSS (Datasea), is a tiny company tied to a recapitalization. That is the kind of name you glance at and pass on.

None of these is a slam dunk, and that is the point. The ones worth a second look are the directors electing equity into a falling stock, because an election plus weakness is closer to conviction than a board-wide grant in a year the stock already ran.

One caveat on the ceiling. A director electing stock is a softer signal than an insider buying shares with their own cash. The backtest is the proof. It is a tilt, not a green light.

That is fine. A real filter mostly tells you to pass. The job is not to manufacture ten ideas a month. It is to find the one eBay hiding in the 98.

And that is the part that is actually new. You do not have to read all filings to catch the next one....Claude Code does this for you.

The same prompt runs on a schedule. It watches EDGAR, throws out the phantom units and the board sweeps on its own, and pings you the moment a director genuinely swaps cash for stock. The work stops being “go read filings” and turns into “a name just showed up worth looking at.”

One more thing...

If you want to build things like this screen yourself, that is the whole point of what I do.

In a few weeks I am running the second cohort of my Claude Code for Investors bootcamp.

The first cohort went better than I hoped. More than 200 portfolio managers, CIOs, and senior analysts came through, from places like Point72, Millennium, Goldman Sachs Asset Management, and HIG Capital. The room ran from associates trying to make the jump all the way up to the CTO of Pacific Equity Partners and the COO of a $5 billion asset manager.

Two of them put it better than I will:

Cohort 2 runs July 13 to 17, and I set the time for both sides of the Atlantic: 7pm London, 2pm New York, 11am San Francisco.

You learn how to build the actual workflows like the ones you see here: screens like this one, diligence that used to take a junior analyst a week, investment memos, and creating elegant Claude Skills automations that run research for you overnight.

I am opening early-bird seats to this newsletter first, before registration goes public. They are limited.

The price is $497 right now, and it doubles to $997 once the early-bird seats are taken. This window is only open for the next few days.

If you have any questions about the cohort, just hit reply and ask me.