What changed
Claude Code can now run a second assistant that points out what you might miss
On October 1st, Claude Code, the version of Claude built for writing software, got a new kind of add-on called mods. Mods change how Claude Code behaves and what it shows you, and they come packaged inside plugins, the bundles Claude Code already installs.
Several mods come built in, and the one worth trying first is called "You should know". While Claude works through a longer task, a second assistant watches alongside Claude. When that second assistant spots something you might miss, it puts a short note above the box where you type.
"You should know" is off by default, and it only works if you sign in with an Anthropic account and leave usage reporting to Anthropic switched on. If your team turned that reporting off, this assistant isn't available to you, and your company's admin can also hide it.
Anthropic's own blog warns that mods have the same access to your computer as Claude Code, and nothing walls them off. On Team and Enterprise plans, a built-in safety mod loads first and stops installed mods from overriding the things your team has told Claude Code never to do. Your admin can choose which plugin stores your team installs from.
Try: in Claude Code, type /plugin enable cc-plugin-you-should-know@builtin, then give it a job with several steps. A short task won't produce a note. If your team uses Claude Code, also ask your admin which plugins are installed, since that's where mods come from.
Sources: Claude Code docs, built-in mods · Claude Code changelog, version 2.1.287, October 1st · Claude blog, "Customize Claude Code with mods", October 1st
Claude Code can now tell you which of your instructions were written for an older model
Claude Code reads an instruction file in each project, called CLAUDE.md, that says how your team works. Teams also save skills, which are reusable instructions for a job they do often.
A new check, /doctor prompt-audit, reads your instruction files and skills and flags anything written for older Claude models. If your team wrote its instructions for an earlier model, this shows you which parts to look at again.
Try: run /doctor prompt-audit in a project you use every week, and read what it flags before you change anything.