Detailed content
🤖 Subagents
Subagents are agents in background which run on parallel — each in its own context window. They can be invoked automatically or on request. The difference from Agent Teams: subagents don't talk to each other; they only report back to the main terminal that created them. It’s the cleanest way to do several things at once without cluttering your context.
🧩 Subagents as files
In your .claude you don’t just have skills, rules, and workflows—you also have agents. These are Markdown files that define what that agent does, which tools it can use, and which skills it contains. You invoke them much like a skill.
- •Each subagent has isolated context — doesn't fill up your main window.
- •Run in parallel: great for research, scanning files, or independent tasks.
- •Report only to the terminal that created them — no cross-talk.
Multiple tasks at once.
A separate window for each agent.
They don’t debate among themselves.
Markdown in .claude/agents.
🔁 /loop
O /loop makes a prompt if repeat at an interval until a condition is met. Under the hood, it uses the cron creation tool — so the loops stay tied to the session: each terminal has its own, and closing the terminal kills the loop. They also shut down automatically after ~7 days with the terminal open.
/loop remind me to drink water every 1 minute # or in natural language, without the slash: set up a recurring loop to check the queue every 20 minutes
💡 Practical tip
You don’t need the /loop to invoke—just ask in natural language ("run this every hour"). Use short loops for reminders and light checks; for something that needs to run while your computer is off, the right place is Routines (#3), not the loop.
Repeats every X amount of time.
Uses the cron tool.
Closed. Gone.
~7 days of terminal use.
📱 Remote Control
Remote Control lets you control your local Claude Code session from your phone or the web. Once enabled, you get a link; open it on your phone (or in the Claude → Code app), and you're controlling the same session— synced in real time. What you type on your phone appears on your computer, and vice versa.
Enable for the session
Ask for “enable remote control” in the session you want to take with you. It generates a unique link.
Open on your phone
Use the link or the Claude app (Code tab). The same session appears in the palm of your hand.
Continue from anywhere
Go for a walk and keep the session alive — brainstorming or building. Every message echoes on both devices.
It's not a copy.
Syncs on both sides.
Link or Claude app.
Step away from your desk without stopping.
⏰ Routines
Routines is the ability to schedule a real agent — not a deterministic Python script. You schedule your Claude Code terminal exactly as you would talk to it. It can be local or in the cloud (running with the laptop closed). Create with /schedule; manage them from the desktop app, with calendar and list views.
✓ Good use of Routine
- ✓Daily summary of what changed in your projects.
- ✓Weekly scan of one source and a report.
- ✓Cloud routine that runs overnight, with your laptop closed.
✗ Don’t confuse
- ✗It's not the /loop (this one lives only in the open session).
- ✗It's not "just an API call" — it's a complete agent.
- ✗You don't need to keep the terminal open if it's running in the cloud.
💡 Practical tip
The real benefit of Routines is having a agent on the schedule, not a dumb robot. It can research, decide, and write—so put as much care into the routine prompt as you would into a regular instruction.
Not a fixed script.
Laptop closed, all good.
Create with one command.
Calendar and list.
📊 Status Line
Probably the feature most underrated of all. The status line is a bottom bar scriptable: shows the model, effort level, and — most importantly — how much of the context window has already been used. It's the easiest fix there is: /statusline and describe what you want (you can even send a screenshot).
✓ Why it matters so much
- ✓You can tell right away when you need to hand off, /clear, or /compact.
- ✓Knows the model and effort level without guessing.
- ✓It was the reason many people switched from the extension to the terminal.
✗ Work without it
- ✗Exceeding the context window without noticing and losing quality.
- ✗Not knowing whether you're using the right model/effort for the task.
- ✗Discover too late that you should have compacted.
💡 Practical tip
Combine the status line with /context (what's filling up the window) and /usage (tokens and cost). Together, they give you full control over the session's scarcest resource: context.
Shows whatever you want.
% and tokens in real time.
Describe it, and you're done.
High impact, low cost.
⚡ Skills
The number-one feature — what gives more leverage in your day-to-day. A skill is essentially a recipe: just as you open a cookbook to make lasagna instead of improvising from memory, the agent opens the skill and follows the steps. The result: it does the task much more consistent than if you repeated the same prompt every time.
📖 The agent’s recipe
Skills are markdown files that live in your .claude. That's why they're so powerful:
- •You create your own e pulls in other people’s (plugins bundle skills, hooks, and MCP together).
- •They're versatile: they can have many steps and tools, or be a single 4-sentence prompt.
- •They chain together: one skill can route to another.
- •Work in Claude Code, in Chat e in Cowork—and you can share it with the team.
✓ When to turn it into a skill
- ✓A prompt you find yourself repeating (e.g., session handoff).
- ✓A process with fixed steps (thumbnails, presentations, video cuts).
- ✓Something that needs to come out the same every time — consistency > improvisation.
✗ Pitfalls
- ✗Thinking a skill needs to be complex — it can be 4 sentences.
- ✗Re-explain the same process with a prompt every session.
- ✗Don't leave skills unreviewed or unrefined over time (they evolve).
💡 Practical tip
Start with the simplest skill you have: that four-sentence prompt you type all the time. Turn it into /comando. It's the fastest way to see why Skills is #1.
Ready-to-follow steps.
The same every time.
Markdown, with the team.
Code, Chat, and Cowork.
🏁 Module summary
You completed the Top 12 🎉
Review the visual tier list in the deck, or revisit any feature from the foundation to the podium.