This is the version to keep within reach. Twelve Claude Code features in ranking order, one line each: what it does and how to run it. The ranking is subjective and leans toward knowledge work and automation — treat it as a map, not a verdict, and build your own from it.
Two terms come back across the whole list:
- Context — the model's short-term memory inside one conversation: everything Claude can see at a given moment. It's finite, so you have to manage it.
- Agent — Claude acting on its own against instructions you leave it, instead of answering each request one at a time.
The top five — these change the most
| # | Feature | What it does | How to run it |
|---|---|---|---|
| 01 | Skills | A ready recipe for a task you do often — a markdown file with steps Claude follows the same way every time. | Save a command you repeat as a skill; write your own, pull someone else's, or install a ready-made pack. |
| 02 | Status line | A bar at the bottom of the screen — shows the model, the effort level, and how much context you've used. | Type the status line command and describe what should be on it; you can also show Claude a screenshot for it to match the layout. |
| 03 | Routines (scheduled agents) | A task scheduled for a specific time — an agent runs, not a rigid script. | The schedule command; in the desktop app you browse routines on a calendar or a list and choose local or cloud. |
| 04 | Remote control from your phone | Driving a local session from your phone or browser, synced both ways. | Turn on remote control mode, you'll get a link, open it on your phone. |
| 05 | /loop — recurring commands | Repeats one terminal command at fixed intervals (a cron-like mechanism). | /loop, or just ask in plain language: "set up a recurring task every twenty minutes." |
Status line — how to read it: 27% means 274,000 of a million available tokens. Seeing how much context is left tells you when to hand work to a fresh session, when to clear it, and when to compact it, before the model starts to lose the thread.
/loop — worth knowing: loops are tied to the session (each terminal has its own, and the loop disappears when the terminal closes) and expire by default after seven days of continuous running. Nothing keeps running forever behind your back.
The next six — you'll reach for these constantly
| # | Feature | What it does | How to run it |
|---|---|---|---|
| 06 | Sub-agents | A helper working in the background; several can run in parallel, and each reports only to the main session, they don't talk to each other. | Define one with a markdown file: what it does, which tools it can use, which skills it carries. Called automatically or on request. |
| 07 | /rewind — checkpoints | Rewinds the code and the conversation to an earlier point, instead of piling fixes onto the context. | /rewind; you pick the exact command and the point in the conversation to return to. |
| 08 | Agent teams | Team members talk to each other and can disagree — panels and debate boards. Expensive (heavy on tokens). | Turn it on in settings first (an experimental feature); then call it in plain language. |
| 09 | Auto-memory | Claude improves its own memory on its own — at a set interval it reviews what you worked on together and updates itself. | Check memory settings to see if it's on; if not, turn it on. |
| 10 | /insights — usage report | A report covering the last thirty days: message and session counts, what's working, what's slowing you down, quick fixes and more ambitious workflows. | /insights. |
| 11 | Ultra plan — cloud planning | Moves planning to the cloud and splits it across multiple planning agents, while you keep working in the terminal. | Type the command, pick cloud mode, you'll get a link; the plan comes back to you or you run it straight from the browser. |
| 12 | /goal — work until the goal is met | Claude doesn't stop until it meets a condition; it verifies itself as it goes, so you can walk away. | /goal with an unambiguous definition of done. |
/goal — the trick to making it work: give a criterion the agent can check itself. "Find three very hard puzzles" is weak, because "very hard" is subjective. "Optimize this page until a given element loads instantly on refresh" is good, because it's measurable.
What's deliberately left off the list
Useful isn't the same as day-changing. The following are worth having, but didn't make the top twelve.
- The foundation you can't do without — the
CLAUDE.mdfile,/clearand/compact, resuming a conversation, web search. - Great, but used less often — voice mode, co-work mode, interactive artifacts.
- Support that ranks above all of the above — office-suite integration from the command line (one tool for spreadsheets, documents and mail), the desktop app for managing routines and reviewing many projects, and prompt caching that quietly saves tokens in the background.
Where to start
Don't try to master everything at once. Pick one feature and work it into your routine long enough that it becomes a habit, then add the next.
- Skills — take one command you type over and over and save it as a skill. That's enough to start.
- Status line — so you stop working blind and can see how much context is left.
- Then keep adding features from the ranking, one at a time, once the previous one is already a habit.
The best tool isn't the flashiest one, it's the one that changes the most in your actual day. Build your own ranking, not someone else's.