Claude Code ships with several built-in commands and keywords that improve how you interact with it day-to-day. Here are the ones worth knowing.
ultrathink — Deeper Reasoning on Demand
Typing ultrathink anywhere in your message tells Claude to reason more thoroughly before responding. It’s useful when a task is genuinely complex — architecture decisions, tricky bugs, multi-file refactors — and a shallow first pass would waste time.
1 | ultrathink Can you provide suggestions to improve this project? |
There is a family of these keywords, each triggering a progressively deeper thinking budget:
| Keyword | Reasoning depth |
|---|---|
think |
Light extended thinking |
think hard |
Moderate |
think harder |
Deep |
ultrathink |
Maximum |
Use them selectively. On straightforward tasks they add latency with no benefit.
/effort — Control How Hard Claude Works
/effort sets the effort level Claude applies to responses. Useful when you want faster, lighter answers for exploratory work, or maximum thoroughness for critical changes.
1 | /effort low # faster, lighter responses |
Running /effort with no argument shows the current setting. Toggle it back to auto when you want Claude to self-pace again.
/insights — Analyze Your Usage Patterns
/insights analyzes your recent sessions and generates a usage report — both a console summary and a full HTML report saved locally.
1 | /insights |
Output example:
1 | 32 sessions · 592 messages · 337h · 88 commits |
The report covers:
- Project areas — which projects got the most attention and what was done
- Interaction style — how you batch tasks, redirect Claude, and delegate
- Friction analysis — recurring failure modes with examples from your own sessions (wrong scope, regressions, auth blockers)
- Suggestions — CLAUDE.md additions, features to try, copyable prompts
- On the horizon — ambitious autonomous workflows matched to your patterns
The friction analysis is the most actionable part. A wrong-scope mistake in a single session feels like a one-off; seeing it happen 16 times across a month makes it worth fixing in CLAUDE.md.
/clear — Reset the Conversation Context
/clear wipes the current conversation context and starts fresh. Use it when:
- The context has grown too large and responses are slowing down
- You are switching to an unrelated task in the same session
- Claude has gone off-track and a clean slate is faster than correcting it
1 | /clear |
A related command is /compact, which summarizes the conversation history instead of discarding it — useful when you want to continue a long session without losing the thread entirely.
Combining Them
These commands complement each other in practice. A typical flow might look like:
- Start a complex task with
ultrathinkto get a thorough plan upfront - Switch to
/effort autofor the implementation work - Use
/clearbetween unrelated tasks to keep context clean - Run
/insightsperiodically to spot patterns and improve your workflow
Small adjustments — adding the right keyword, clearing context at the right moment — add up over dozens of sessions.