- Published on
Claude Code 2.1.290: AI Watching AI β The "You Should Know" Mode and Managed Agents Onboarding
The most dangerous moment in coding is when concentration is at its peak. Deep focus on one file makes it easy to overlook side effects elsewhere, and the more promising a solution looks, the harder it is to notice when it has drifted from the original goal. Claude Code 2.1.290 tackles this problem head-on.
1. "You Should Know" β A Watchdog That Sees the Whole Picture
You Should Know is not just a plugin β it is a side agent. While you work alongside Claude, it runs a separate process in the background and surfaces important information in real time.
What it catches:
- Other files that may be unintentionally affected by a code change
- When the direction of work is drifting from the original goal
- Choices that conflict with patterns or settings you have defined
Enabling it takes one line:
/plugin enable cc-plugin-you-should-know@builtin
Think of it as a teammate who keeps watching the big picture while Claude focuses on writing code. Even in a solo development environment, you can have a "reviewer''s perspective" running alongside you.
2. Managed Agents Onboarding: One Command to Bootstrap Multi-Agent Architecture
Introducing the Managed Agents pattern into a project used to be laborious β finding the right documentation, manually writing configuration files, and understanding the structure from scratch. Version 2.1.290 simplifies this:
/claude-api managed-agents-onboard <url>
/claude-api managed-agents-onboard <quickstart-name>
For example, specifying the deep-researcher template instantly generates a research-oriented agent architecture. Instead of designing from zero, you start from a proven pattern.
3. VS Code Agent Map: Every Agent on One Screen
If you use Claude Code inside VS Code, you will notice a new "N agents" pill at the bottom of the screen. Clicking it opens the Agent Map, which shows all sub-agents running in the current session.
- View each agent''s current status as an individual card
- Stop a specific agent (Stop agent)
- Open read-only transcripts for any agent
The need to dig through logs to answer "what agent is doing what, and where?" disappears.
4. Session Management Commands
claude attach <name> and claude logs <name> have been added. You no longer need to remember a full session ID β typing part of the name is enough to attach or retrieve logs.
Tips for Getting the Most Out of These Features
You Should Know + Refactoring: Enable this mode when refactoring across multiple files. It will flag side effects in connected files β particularly effective in structures where test files and configuration files are tightly coupled.
Starting with Managed Agents: The deep-researcher template is a great entry point for automating research workflows. Rather than designing a multi-agent system from scratch, analyze the template structure and adapt it to your project.
Sources: