Claude Code Efficiency Doubling Guide - Do You Know These Commands

Introduction

“Pressing Esc twice can revert code? There’s such a thing?”

This was my friend’s reaction last week when I was chatting about Claude Code. He’d been using it for over half a year and didn’t know this shortcut.

This isn’t really his fault. Claude Code’s features are buried deep, and the update speed rivals AI industry news - just when you learn one trick, new features are already online. The latest versions added quite a few practical features: /btw, /loop, /simplify…

This article organizes the 12 most worth mastering features from the past six months. Whether you’re a beginner or veteran, you should find at least one or two “I didn’t know it could do that” surprises.

1. Agent Teams (Swarm Mode) - Biggest Update ⭐

Your Pain Point

Having AI do a big task, but:

Halfway through it forgets what was said before
Gets stuck in an infinite loop at some step
You have to keep watching it like a babysitter

What is Agent Teams?

Simplest analogy:

Traditional AI = You hired one employee, they do one thing when you say one thing
Agent Teams = You started a small company, hired several employees

You just need to say “I want to build a website”, then:

Employee A handles design
Employee B handles coding
Employee C handles testing
Employee D handles documentation

The key is, these people will figure out how to work together on their own, without you constantly watching!

How to Enable?

Add environment variable in settings.json:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Core Architecture

Team Leader (Project Manager): Coordinates globally, assigns tasks, quality control

Team Mate (Members): Independent Claude Code instances, each with their own responsibilities

Three Collaboration Mechanisms

1. Shared Task List

Like a company task board, shared by all Agents
Leader automatically splits tasks, assigns to different members
Updates status when complete, everyone knows progress

2. Message Mailbox

Agents can communicate directly with each other
No need to relay through you as the boss
Efficient, smooth collaboration

3. Independent Context

Each Agent has their own workspace
Don’t interfere with each other, won’t “forget” what was said before
No overflow no matter how complex the task

Official Case: C Compiler in Two Weeks

Anthropic officially used 16 Agents working in parallel:

2000 Claude Code interactions
100,000 lines of code
Two weeks
$20,000 cost

Compared to traditional compiler project development cycles, this speed is absurdly fast.

6 Usage Templates

Template 1: Prompt Optimization (Beginners Must Learn)

Simplest approach, just modify the prompt:

Use Agent Swarm to help me write a [your project].

Execution requirements:
1. Use skills to handle tasks as much as possible
2. Split tasks so multiple sub agents can execute in parallel
3. Prioritize using find-skills to search for the most suitable Skill

Template 2: Cross-Model Agent Army (Save Money)

Not all tasks need the most expensive model:

Complex tasks (architecture design, core logic) → Use Opus
Simple tasks (write docs, write tests) → Use Haiku

Template 3: Community Skill Ecosystem Reuse

Community has tons of open source Skills: paper analysis, code review, bug fixes… Let Agents automatically search for the most suitable ones.

Template 4: Plan Mode + Review Loop

Complex tasks fear rework. The flow is:

Planning Agent produces detailed Todo List
Review Agent checks, points out gaps and risks
Execution Agents work in parallel
Quality check at each milestone

Template 5: Lead-Member Architecture

Best for large projects:

Lead coordinates globally
Frontend Agent handles frontend
Backend Agent handles backend
Testing Agent handles testing
Documentation Agent handles docs

Template 6: Advanced Feature Combination

Persistent memory: Share information across sessions
Hooks mandatory rules: Automatically check code standards
Dynamic tool discovery: Automatically search for available tools

2. /btw - Interject Without Polluting Context

Release Date: March 11, 2026

Your Pain Point

Having Claude Code refactor a large module, and suddenly you want to ask: “Hey, which directory is that test file in?”

One question, Claude stops to answer you, and an irrelevant conversation segment gets added to the context window. When it starts working again, it might go off track.

This is context pollution.

Solution

/btw your question

For example:

/btw What's the scraping flow for this project?

Why Is It Good?

Doesn’t interrupt current task: Two processes run in parallel
Doesn’t pollute conversation history: Press space, enter or Esc to collapse after answered
Almost no token cost: Reuses current prompt cache

The kind you can’t go back from once you use it. I now use it several times in every long session.

3. /rewind - Undo Button (Press Esc Twice)

Your Pain Point

Having Claude try a new approach, changed it and found it doesn’t work, can only rewind the entire conversation together, losing your previous discussions too.

Solution

After the February 2026 upgrade, /rewind can separately revert code and conversation.

Four Modes

After entering /rewind, a menu pops up:

Revert code and conversation - Undo everything
Revert conversation but keep code - Code stays, only undo discussion
Revert code but keep conversation - Best for experiments! Claude remembers this path doesn’t work
Compress to release context - Release context window space from this point

Best Practice

Let Claude try a new approach, if it doesn’t work:

Revert code
Keep conversation

This way Claude still remembers what you just discussed, knows this path doesn’t work, can directly change direction without re-explaining requirements.

4. /insights - Usage Habit Analysis Report

Feature Introduction

Enter /insights, it generates an HTML report analyzing your Claude Code session habits:

Which commands you use most frequently
What repetitive operation patterns you have
Recommended custom commands and Skills
Records of failed attempts

Why Is It Underrated?

It’s actually Claude Code observing you in reverse.

It will tell you:

Which operations can be automated
Which habits are wasting time
Which errors keep appearing

Recommendation: Run /insights periodically to re-understand your habits, very interesting.

5. /simplify - Three-in-One Code Review

Release Date: Integrated late February 2026

Feature Introduction

Enter /simplify, Claude Code simultaneously starts three parallel Agents:

Code reuse perspective: Which code can be reused?
Code quality perspective: Which patterns can be optimized?
Runtime efficiency perspective: Which places affect performance?

Usage Recommendation

After many rounds of conversation with Claude Code and writing several large feature updates, run /simplify.

AI-written code often has:

Extra imports
Repeated logic
Places that could use more concise syntax

/simplify can basically pick them all out, equivalent to having three colleagues review your code simultaneously.

6. /branch - Parallel Universe (formerly /fork)

Feature Introduction

Fork the current conversation into a new session, original session unaffected.

A bit like ChatGPT’s branch feature.

/rewind vs /branch

/rewind = Undo Button (go back)
/branch = Parallel Universe (fork)

Use Case

Claude just helped you work out a solution’s approach, and you want to try two different implementation methods along this line:

/branch once
Two sessions go different ways
Pick the one with better results in the end

7. /loop - Periodic Tasks

Usage

/loop 5m check deployment status

Runs once every 5 minutes, no need to watch it yourself. Default interval is 10 minutes.

Features

Results come directly into current conversation
Claude can make judgments and follow-up operations based on these results
Better suited for continuous checking within current session

Suitable Scenarios

Monitor deployment status
Periodically check task progress
Continuously observe certain metrics

8. /remote-control - Phone Remote Control

Release Date: Late February 2026 Applicable Users: Official docs currently state All plans can use; Team/Enterprise may need admin to enable

Usage

Enter /rc or /remote-control, it generates a URL.

Open this link on your phone, your entire Claude Code session appears on the phone.

Why Is It Good?

Fully synced: Send commands from phone, terminal sees them too; terminal operations, phone updates in real-time
Secure: Code always runs locally, phone is just a remote
Convenient: File system, MCP servers, project config, all still local

Continue your work even when you’re out, very practical.

9. /export - Export Conversation

Usage

Enter /export to export the entire current conversation as a plain text file, convenient for archiving and sharing.

Why Is It Needed?

Many valuable things in Claude Code are produced during conversation:

You discussed an architecture plan with Claude for half an hour
Lots of back-and-forth deliberation and judgment in between
Finally reached agreement

If these discussions aren’t saved, they’re very hard to find later.

After exporting, you can also share with other AI tools (like Codex) for collaborative use.

10. /model opusplan - Money-Saving Mode

Why Is It Needed?

If your Opus quota is limited, using Opus for all coding might get rate-limited halfway through.

Solution

/model opusplan

This mode will:

Complex reasoning → Use Opus 4.6 (plan mode)
Specific execution → Use Sonnet 4.6 (faster, cheaper)

What Is It Really?

More accurately, opusplan is a model alias configuration, not some “hidden entrance”. You can use /model opusplan directly, or set it as default in settings, environment variables, or startup parameters.

Planning and coding have completely different model capability requirements:

Planning needs deep thinking, understanding entire project architecture
Coding is actually completely fine with Sonnet, and much faster

Best of both worlds, highly recommend saving Opus mainly for planning phases.

11. /fast - Fast Mode

Feature Introduction

Fast mode is Claude Opus 4.6’s low-latency configuration, suitable for scenarios where you want faster results.

Usage

/fast          # Toggle switch
/fast on       # Enable
/fast off      # Disable

Suitable Scenarios

Rapid iteration: Frequently modifying code, need to see results quickly
Real-time debugging: Modifying and testing, don’t want to wait too long
Simple tasks: Daily work that doesn’t need deep reasoning

Note

Fast but slightly higher cost per token. Whether to think more mainly depends on /effort; /fast is more like switching gears between speed and price.

12. /compact - Context Compression

Your Pain Point

Conversation getting longer, Claude starts “forgetting”:

Forgets requirements mentioned before
Repeatedly asks questions already answered
Response quality drops

This is because the context window is almost full.

Solution

/compact                              # Basic compression
/compact focus on the database work   # Focused compression

Function

Compress conversation history, release context space:

Keep key information
Delete redundant content
Let Claude “travel light”

Best Practice

Don’t wait until Claude starts forgetting to use it!

Recommendations:

After completing a work phase, proactively /compact
Before starting a big new task, compress first
Long sessions: compress every 30-60 minutes

Even if you don’t use it, Claude will eventually auto-compress. But active use lets you better control what to keep.

Bonus: Practical Shortcuts

Shortcut	Function	Note
Ctrl+V	Directly paste screenshot	Common in macOS terminal is Ctrl+V, iTerm2 also supports Cmd+V
Ctrl+J	New line	Mac can also use Option+Enter
Ctrl+R	Search history prompts	Find previously asked questions
Esc Esc	Open rewind menu	Can quickly enter rewind / summarize menu

Summary

People who know how to use these features get their time multiplied.

People who don’t will just keep being chased by tasks.

Core value of these 12 features:

Command	One-Line Summary	Note
Agent Teams	Multi-Agent collaboration, you be CEO	Experimental feature
/btw	Interject without polluting context	Available to all
/rewind	Undo button, four modes	Available to all
/insights	Usage habit analysis report	Available to all
/simplify	Three-in-one code review	Available to all
/branch	Parallel universe, fork exploration	Available to all
/loop	Periodic tasks, auto monitoring	Available to all
/remote-control	Phone remote, anywhere anytime	All plans; Team/Enterprise may need admin enable
/export	Export conversation, easy archiving	Export as plain text
/model opusplan	Plan with Opus, execute with Sonnet	Can be set as default model
/fast	Low-latency mode	More of a speed/price toggle
/compact	Context compression, prevent forgetting	Available to all

Starting tonight, pick the feature you need most and try it. Tomorrow you’ll feel the workflow upgrade.

Welcome to follow the WeChat Official Account FishTech Notes for more discussions!