anthropic just published claude's system

anthropic just published claude's system prompts officially. and they update them with every release.

now you can see exactly what instructions a model gets before you even type your first message. and more interesting — you can track how those instructions changed release over release.

why it's worth reading: → you see why the model answers the way it does, not some other way → you understand the actual logic behind its limits and permissions → you write better prompts yourself, just by reading how anthropic writes theirs

rare case of a company showing you the kitchen, not just the finished dish.

docs.claude.com/en/release-notes/system-prompts