How Permissions Work by Default
Every time Claude Code wants to run a terminal command, edit a file, or create something new, it asks for your permission first. You see a prompt, you approve it, and then Claude continues.
This is a good safety net. It means Claude can't accidentally delete your files, run a dangerous command, or make changes you didn't expect. For most situations, this is exactly what you want.
But sometimes, it gets in the way.
The Problem with Constant Approval
Imagine you ask Claude to build a full web app. Claude needs to create directories, write files, install packages, run build commands, fix errors, and test things. Each of those steps triggers a permission prompt. You end up clicking "approve" dozens of times, breaking the flow of what should be an autonomous process.
For personal projects where you trust Claude to work freely, all that clicking is just friction.
The Command
Claude Code has a special option that skips all permission prompts:
claude --dangerously-skip-permissionsWhen you run Claude with this option, it won't ask before running commands, editing files, or executing scripts. It just does the work.
The word "dangerously" is intentional. It's a reminder that you're removing a safety layer. Claude will still try to do the right thing, but it won't pause to check with you first.
When to Use It
This mode makes sense when:
- Personal projects. You're working on your own code, on your own machine, and there's nothing sensitive at stake.
- Rapid prototyping. You want Claude to build something fast without interruptions. You'll review the output after, not during.
- Trusted environments. You're in a sandbox, a container, or a fresh project folder where nothing important can break.
- Learning and experimenting. You're following along with a tutorial or trying things out, and the constant prompts slow you down.
When NOT to Use It
Do not use this mode when:
- Production systems. If Claude runs a bad command on a live server, there's no approval step to catch it.
- Shared projects. Other people's work deserves the extra caution of permission checks.
- Sensitive data. If the project has API keys, database credentials, or personal data, keep the safety net on.
- Unfamiliar projects. If you don't fully understand the project yet, let Claude explain what it wants to do before it does it.
Running It
Open your terminal, go to your project folder, and run:
claude --dangerously-skip-permissionsThat's it. Claude Code starts in full autonomy mode. Every command, file edit, and script execution happens without a permission prompt.
When you're done, just exit Claude Code normally. The option only applies to that session. Next time you run claude without it, permissions are back to normal.
Combining with Plan Mode
A powerful workflow is to combine this with Plan Mode. Start Claude in plan mode to review the approach, then restart with skip permissions to let it execute the plan without interruptions:
# First: review the plan
claude
# Then: execute with full autonomy
claude --dangerously-skip-permissionsThis gives you the best of both worlds: thoughtful planning with uninterrupted execution.
What You Learned
The --dangerously-skip-permissions option removes the approval step so Claude can work on its own. Use it for personal projects and quick experiments. Skip it for important projects, shared work, and sensitive files.
Now that you've unlocked Claude's full speed, it's time to build something real.
