A Claude Code sideagent now flags what you might miss

Claude Code is getting a second reader for Claude's output: a sideagent that watches what Claude writes and picks out the important information you might skim past, the Claude Code team announced on X. The built-in plugin is called "You should Know."

At a glance

• The plugin is labeled builtin, and you turn it on with a single command, /plugin enable cc-plugin-you-should-know@builtin, typed into a Claude Code session like any other slash command.
• It is built as a mod, the extension type Claude Code recently opened up, and it works by spinning off a sideagent whose only job is to observe Claude's output.
• The announcement does not say what the sideagent treats as important, how it flags what it finds, or what running a second agent next to Claude costs in usage.

If you haven't been following, Claude Code has been adding extension points for more than a year. According to GitHub, it launched in research preview as an agentic coding tool in February 2025, added remote MCP server support in June 2025 and subagents for specialized assistants in July 2025. Per AI/TLDR, mods arrived in Claude Code 2.1.287, released 2026-10-01, and the built-in "You should know" mod shipped with them.

"You should Know" is switched on with one /plugin command

The announcement is brief. The Claude Code team says "You should Know" scans Claude's output for important information you might miss, to help keep you in the loop, and that it works by spinning off a sideagent to observe that output. The post calls the plugin a mod and points readers to the team's newest blog post on mods for more detail.

To turn it on, you type /plugin enable cc-plugin-you-should-know@builtin in a session. The enable step matters. According to AI/TLDR, the built-in mod is off by default and works in first-party sessions with telemetry on. If you run Claude Code through another provider or keep telemetry switched off, check that condition before you expect the sideagent to show up.

"You should Know" ships alongside Claude Code's new mods

According to Claude Code Docs, some of Claude Code's own features are already mods, such as /diff. Other mods install the way plugins do, from a marketplace, with a command like /plugin install token-chart@your-org. The docs also point to sample mods in the claude-code/mods directory of the claude-code-playground repository, which are shared as they are.

Per AI/TLDR, mods open Claude Code's behavior to the same plugin marketplace that already carries skills and commands, so teams can package review or safety habits as a plugin. The community moved quickly. According to StashBase, on October 2 the awesome-claude-code-mods list covered 359 mods. Of those, 127 start processes, 95 write files, 75 reach the network and 111 see every prompt.

How does a mod watch Claude's output?

Through events. Per Claude Code Docs, a mod is a plugin made of JavaScript or TypeScript event handlers. Claude Code calls a handler when something happens, such as a tool call, a submitted prompt or part of the interface being drawn, and the handler can watch the event, change it or take it over. A sideagent that only observes Claude's output sits at the watching end of that range.

The bigger difference is where the code lives. The docs say settings hooks, skills, status lines and MCP servers work from outside Claude Code, while a mod runs inside it. From there it can draw panes and bands above the prompt, restyle Claude Code's own interface, hold or answer tool calls, and send a request to a different model. Think of a hook as a note slipped under the door, while a mod is a colleague sitting at your desk.

What can a mod reach on your machine?

Quite a lot. StashBase quotes Anthropic as saying mods aren't sandboxed: a mod "is code that runs with your permissions. It can read and write your files, start processes, and make network requests." Per StashBase, a mod can also see every prompt, rewrite it, approve tool calls before you are asked and spend your usage on model calls. Deny rules don't cover a mod's own file and process calls.

The same outlet lists two safeguards. You can check a mod without running it using claude plugin validate ./some-mod, and you can start a session with no installed mods using claude --safe-mode. StashBase also notes that the interface mods draw only shows in the terminal and Desktop, while hooks run in the terminal, the Desktop Code tab, VS Code and claude -p.

The announcement leaves the most useful details out: what the sideagent counts as important, how it shows what it finds, and what a second agent running next to Claude costs you. Mods can spend your usage on model calls, according to StashBase, so the cost question deserves a real answer. In our view, the telemetry requirement that AI/TLDR reports is the condition to read twice, because it decides whether the feature runs for you at all.

What the sideagent will flag

The team points to its newest blog post on mods. The announcement itself gives no examples of what "You should Know" flags and no figure for its usage cost. It also says nothing about whether the plugin will ever work outside first-party sessions with telemetry on. Until those details come out, the practical test is to enable it on a real task and see what it surfaces.