The viral "1,500 dollars a day Claude trading bot" with a "68.4% win rate hidden in Anthropic's guide" is not real. It got community noted, and the trading screenshots were stapled onto an Anthropic skills guide that never mentions trading. So ignore the income numbers. What is real is much more useful: Claude can read prediction markets and analyze them for you.
Sites like Polymarket and Kalshi let people trade on real future events. A share pays one dollar if it happens and nothing if it doesn't, so the price is the crowd's probability. Thirty cents means the crowd thinks about thirty percent. When news breaks, the price moves, so it is a live, money-backed probability feed.
Polymarket did over nine billion dollars in volume in 2024 with 314,500 active traders, and combined prediction-market volume jumped from under five billion dollars a month to around twenty-four billion a month in 2026. That is a lot of real signal you can point Claude at.
This works in Claude Code, where skills have real network access.
Create a skill. In Claude Code, make a folder with a SKILL.md file. In the description write: "Pull and analyze live prediction-market odds from Polymarket." That description is what makes Claude load the skill automatically.
Give it the data. Polymarket's Gamma API is public and needs no key for reading. Point the skill at https://gamma-api.polymarket.com to fetch active markets, prices, and volume.
Pull and rank: "Pull the top 20 active markets by volume, show each market's implied probability, and flag any that moved more than 10 points in the last 24 hours."
Add the news layer: "For the 5 biggest movers, search the news and tell me whether the price has caught up to the story or not." This is where the edge-spotting happens.
Compare platforms: "Check whether Kalshi and Polymarket disagree on the same event, and show me the gap." Differences between platforms are the most interesting signal.
The real Anthropic skills guide, no trading in it
Anthropic Agent Skills doc
Polymarket's public API docs
Claude reading 1.3 billion Polymarket trades, only 20.6% of traders were profitable