Dangerous mode
Full Lesson Reference
Claude Code has a flag called --dangerously-skip-permissions. As the name suggests, it removes all permission prompts. Every action Claude wants to take, it just takes - no prompts, no confirmations, no rails.
You'll see this flag in other people's videos, blog posts, and workflows. They make it look cool. This lesson explains what it actually does, why it exists, and why it's the wrong tool for marketing work.
What it does
With --dangerously-skip-permissions enabled, Claude Code
- Never asks before editing files (any file, any location)
- Never asks before running commands
- Never asks before installing packages
- Never asks before pushing to GitHub
- Never asks before deleting anything
- Never asks before calling external APIs
- Never asks before anything
This is different from auto mode. Auto mode keeps rails on destructive actions. Dangerous mode removes every rail.
Why the flag exists
It's there for developers running Claude Code in isolated environments where permission prompts would block automation:
- CI/CD pipelines (automated build-and-deploy systems)
- Docker containers with nothing sensitive inside
- Sandbox VMs that get destroyed after the run
- Throwaway test repos where mistakes don't matter
The common theme: isolated environments where a mistake can only affect a temporary system, not your real work, real clients, or real machine.
Why you shouldn't use it
You run Claude Code on your actual laptop, inside your actual work folders, connected to your actual client platforms. Mistakes in this environment are not cheap:
- Delete the wrong folder - real files gone
- Push wrong content to GitHub - real URLs update
- Mutate a campaign - real spend hits the client's card
- Send a draft email - real customers see it
- Run a destructive database operation - real data gone
The permission prompts you'd skip with dangerous mode are the things that would catch any of these before they happen. Skipping them turns a minor confusion into an expensive recovery job.
When you'll see it recommended
Lots of online content suggests dangerous mode as a "productivity hack" or "how power users run Claude Code". These posts usually come from:
- Developers working in throwaway sandbox environments
- YouTube demos on fresh VMs that don't represent real daily work
- Tutorials focused on how fast Claude can work, not how safely
The speed benefit is real in a sandbox. The risk profile is completely different in a marketer's actual workflow .
If you think you need it
You almost certainly don't. Here's the decision tree
Are the permission prompts slowing you down?
- For routine work - use auto mode (Lesson 3). It removes friction without removing safety.
- For batch work - approve once with option 2 ("don't ask again this session"). Keeps you in control for the session.
- For repetitive patterns in one project - approve with option 3 ("don't ask again for this project"). Persistent but bounded to the project.
These three options cover 99% of the "I want fewer prompts" situations. The remaining 1% is work that probably SHOULD have prompts.
If you already enabled it accidentally
Dangerous mode is a flag you pass when starting Claude Code. It's not persistent - end the session and it's gone.
If you started a session with claude --dangerously-skip-permissions, just type: Close the terminal, then start a new session with just claude (no flag). You're back in safe mode.
The framing to remember
Every other mode in Claude Code is designed to help you. The word dangerously is in the flag name for a reason - Anthropic is telling you directly that this is risky. They don't use that word anywhere else in the product.
When a tool has dangerously in its name, believe it.
Action items
☐ Understand what --dangerously-skip-permissions actually removes
☐ Never start Claude Code with this flag for real work
☐ If you see it recommended online, check the author's context - it's usually for sandboxes, not marketing work
☐ Use auto mode or session-level approvals instead when you need fewer prompts
☐ If you ever accidentally enabled it, close the terminal and start fresh without the flag
Next lesson: What never to send through Claude Code.
Exercises
- Review the concepts covered in this lesson: Dangerous mode.
- Write down your key takeaway from this lesson.
- Practice running any commands or prompts mentioned above inside your terminal.