These seven commands are built into Claude and cost nothing to use, but most people never touch them. Each one solves a specific problem you've probably worked around manually: burning through usage limits, losing accuracy on long sessions, or opening a second window every time you have a quick question.
Two of them (/loop and /schedule) do overlapping jobs, so read those together — the difference between them is whether the task needs files on your computer.
ultrathink — force maximum reasoning on a single prompt
Type ultrathink anywhere in a prompt and Claude immediately switches to maximum reasoning effort for that prompt only, regardless of your default setting. When the prompt is done it reverts to your default. You never have to touch your settings.
Overrides your reasoning setting for one prompt
If your default reasoning effort is set to low or medium, typing ultrathink tells Claude to treat this specific prompt as maximum effort. It switches back afterwards automatically.
When to use it
Run your normal day-to-day prompts on low or medium. When you hit something genuinely complex — a strategy, a technical plan, an important decision — add ultrathink to that one prompt.
This matters most on the $20/month plan, where you're otherwise burning high-reasoning tokens on simple tasks. Good candidates:
- Complex campaign planning
- Technical architecture decisions
- Detailed analysis
- Anything where you need Claude to think through every angle before responding
/caveman — cut output tokens without losing quality
/caveman is an open-source Claude skill that significantly cuts the number of output tokens Claude uses. If you're on a lower-tier plan and watching your usage limits, this is one of the most practical things you can install.
How to install it
Open Claude and type: "Install the open-source skill called caveman."
Claude installs it automatically. No downloads, no setup.
How to use it
Add /caveman to any prompt where you want to reduce output token usage. Claude responds more concisely without losing the substance of the answer.
Useful for long research tasks, bulk content generation, and any session where you're producing a lot of output and want to stretch your usage further.
/insights — get a monthly report on how you're using Claude
/insights tells Claude to analyse how you've been using it across all your sessions in the past 30 days. It generates a detailed HTML report showing what you're doing well, what's holding you back, and specific ideas to try — based on your actual usage, not generic tips.
What the report covers
- What's working well in your sessions
- Anti-patterns and things that need improvement
- Specific commands, prompts, and features to try next
- Recommendations based on tasks you've actually been doing, not random suggestions
The output is a full HTML report you can browse through and reference.
How to use it
Just type /insights. Claude does the rest.
Run it once a month for a reality check on whether you're getting full value from Claude. Most people find at least two or three things they didn't know existed that are directly relevant to what they're trying to do.
/loop — schedule recurring tasks in your Claude session
/loop lets you schedule a task to repeat on an interval inside your current Claude Code session. Tasks run on your computer, so your machine needs to be on.
The syntax
/loop [interval] [what to do]
Example: /loop 5m /write draft a LinkedIn post on this topic
This runs the /write skill every 5 minutes. You can call other skills inside the loop command.
Key limitations
- Minimum interval: 1 minute
- Maximum duration: 3 days — tasks expire
- Session-specific: if you close the session, the loop stops
- Your computer must be on for tasks to run
Use /loop for short-duration recurring tasks that need access to your local files. For longer-running automations that need to run without your computer on, use /schedule instead.
/schedule — run automated tasks in the cloud
/schedule runs tasks in the cloud on a timer, completely independent of your computer. It clones your project repo and runs the task on schedule. Your machine doesn't need to be on.
/loop vs /schedule
| /loop | /schedule | |
|---|---|---|
| Runs on | Your computer | The cloud |
| Minimum interval | 1 minute | 1 hour |
| Local file access | Yes | No |
| Expiry | 3 days | None |
| Tied to your session | Yes | No |
Use /loop when the task needs files on your computer. Use /schedule when it doesn't, and you want it to keep running regardless of whether you're at your desk.
Good use cases
- Daily standup summaries
- Recurring performance reports
- Scheduled content generation
- Any automation where you don't need local file access
Plan limits
Pro plan: 5 runs per day. Next tier: 15 per day. Team/Enterprise: 25 per day.
These limits may change — check current plan details in your Claude settings.
/btw — ask a question without interrupting Claude
When Claude is in the middle of a long task, type /btw followed by a question and Claude answers it without interrupting the main task. It keeps running what it was doing while also handling your side question.
You don't have to open a new session
Normally, if Claude is running a long task and you have a quick question, you'd open a new session — but a new session has none of the context from the current one.
/btw lets you ask inside the same session, so Claude has all the context and answers accurately, without breaking the main task.
Parallelise your work
While Claude is running a long research task, building something, or processing a large output, use /btw to get quick answers, check something, or run a small parallel task.
Example: /btw what's the word count limit for LinkedIn posts?
Claude answers that while continuing to build your campaign brief in the main task.
/clear — wipe context between tasks to keep Claude sharp
/clear wipes the context from your current session. Use it when you switch between unrelated tasks. AI becomes less accurate as context builds up — clearing it gives Claude a clean slate and saves tokens.
More context isn't always better
Research shows that AI accuracy decreases as the amount of content already in the context window grows. Old, irrelevant context from previous tasks makes Claude work harder and perform worse on the new task.
If you've been writing a YouTube script and now you want to write a cold email, that entire script conversation is now noise Claude has to process through.
Use it between unrelated tasks
Any time you finish one task and start something with a different context, run /clear first. For example:
- Switching from writing content to analysing data
- Moving from one client to another
- Going from research to execution
Don't use it when you're continuing the same project or need Claude to remember what was just discussed. Use it when you're starting something new and the old context would just be noise.
It also saves tokens
Every prompt Claude processes includes all the context from the current session. The longer the session, the more tokens every new prompt costs.
Running /clear between tasks means each new prompt starts lean. On a limited plan, this directly extends how much you can get done before hitting usage limits.
