Anthropic publishes Claude's full system prompts as public docs

Anthropic has started publishing Claude’s full system prompts as public documentation. These are the instructions that run behind the scenes before your chat even starts, telling the model how to behave, what to avoid, and how to format answers. Early Claude prompts were around 300 words. Today’s are running past 3,000. That is a big jump, and it is now out in the open for anyone to read at the official release notes.

An active Hacker News thread, sitting at 373 points and 157 comments, is picking through the details. People are flagging what they see as bloat, wasted tokens, and instructions that contradict each other in places. Some are pointing out that Pi, a competing assistant, uses a system prompt of roughly 300 words and reportedly holds up just fine, or even performs better in some comparisons. The takeaway floating around is that longer is not automatically better. More instructions can mean more room for the model to get confused about which rule wins.

Here is why this matters if you run a business. If you’re building anything on top of Claude or other models, whether that’s a support bot, an internal tool, or a client-facing assistant, your own system prompt is doing the same job. Bloated instructions cost you money on every single call and can quietly make your outputs less consistent. Anthropic publishing their prompts gives you a real reference point for what good structure looks like versus what just adds noise. It’s worth treating your prompt library the same way you’d treat any other piece of business infrastructure, worth reviewing and trimming on a schedule. That kind of ongoing prompt management is exactly the sort of thing we build into an AI command centre, so nothing gets bloated and forgotten in Omni.