Two Anthropic engineers spent 24 minutes walking through Claude’s lesser-known tricks. They start by showing how to unlock Claude Code’s debug mode. Instead of feeding it a monolithic prompt, you wrap your instructions in triple backticks and add “//verbose” at the top. That gives you a line-by-line breakdown of Claude’s reasoning. You see every internal thought process it normally hides—exact token counts, draft outputs, even discarded ideas. It turns Claude from a black box into something you can audit and refine. Next, they dive into the custom memory slots. By default, Claude forgets everything between sessions. But if you prepend “///MEMORY SLOT: [name]” at the top of your prompt, Claude will save that chunk of context for the next run. You can create up to five distinct slots—project specs, style rules, data schemas, you name it. When you invoke them later with “///LOAD MEMORY: [name],” Claude recalls only that slice. It keeps conversations tighter and stops it from re-asking the same basic questions. They also demonstrate a hidden file-upload shortcut. Claude Code will normally handle JSON and CSV, but these engineers showed that if you slip an S3 link into your prompt and wrap it in angle brackets, like <s3://bucket/mydata.csv>, Claude will fetch and parse the file automatically. No manual copy-paste. You get parsed tables ready for data manipulations, plus inline plotting commands. They ran a live demo: a 100-row sales CSV became a trend chart in seconds. Finally, they cover API cost tracking and throttling. Deep in the settings, there’s an “experimental rate limiter.” You can dial it in to cut your monthly compute spend by up to 30 percent. They turned off “burst mode,” which removes the typical 1,000-token cap but drives costs through the roof. With burst mode off and “budget watch” set to $200, Claude will throttle itself once you hit that cap, then send you a detailed usage report. It’s a simple toggle, but it keeps your bills predictable.