Last updated 13 October 2026
Switch between quiet and interactive modes
Controlling the level of manual intervention
Manual approval for every single action slows down your development cycle. You want the agent to be helpful and autonomous but you also need to ensure it does not execute destructive shell commands without your oversight. Balancing these needs requires a clear understanding of how permissions work in the terminal environment. You can toggle between different levels of trust to match the specific task you are currently working on. This control prevents you from becoming a bottleneck in your own workflow while maintaining a safety net for your project.
Every session begins with a set of rules about what the agent can do on its own. If these rules are too strict, you spend more time typing responses than writing code. If they are too loose, you risk the agent making changes that are difficult to track or revert. You can resolve this by mastering the four permission modes that govern how tools like Bash and Write are used during your interaction with the command line agent. Choosing the right mode at the right time ensures that you stay productive without sacrificing the integrity of your code.
The security of the default permission mode
The default permission mode is designed for maximum safety and developer awareness. In this mode, the agent must ask for your permission before it uses tools that could change your file system or execute code. Tools such as Write and Bash require a manual confirmation for every single instance of their use. This is the baseline experience for new users because it prevents accidental damage while you are still learning how the agent interprets your prompts.
While the default mode is secure, it is also the most interactive. You will see a prompt in your terminal every time the agent wants to perform an action. You can use this opportunity to review the proposed shell command or the content of the file being written. This mode is best for sensitive tasks where you need to verify every line of code or every system configuration change before it is applied to your environment. It ensures that no major changes happen without your direct consent.
Using the plan mode for architectural changes
When you need the agent to think through a complex problem without taking any action, you should switch to plan mode. This is the most restrictive permission level available. In plan mode, the agent is restricted from using tools that modify your project. It can still use the Read and Glob tools to explore your codebase, but it cannot use Edit or Write to make changes. This forces the agent to provide a conceptual explanation of how it would solve a problem instead of just doing it.
You might use plan mode when you are designing a new feature or planning a major migration. It allows the agent to scan your repository and build a mental model of the existing architecture. You can then discuss the pros and cons of its proposed solution without worrying about it touching your source files. Once you are satisfied with the strategy, you can switch to a more permissive mode to begin the actual implementation work. This separation of planning and execution reduces the risk of incorrect implementations.
Speeding up development with accept edits mode
The acceptEdits mode is a common choice for developers who want a quieter experience. This mode allows the agent to use the Read, Glob, Grep, and Edit tools without asking for your approval. It significantly reduces the number of interruptions during a typical coding session. The agent can search through your files and apply code modifications autonomously, which is perfect for fixing bugs or adding small features across multiple files without constant back-and-forth prompts.
Crucially, acceptEdits still protects your system by requiring manual approval for the Bash tool. If the agent needs to run a script, install a package, or execute a command in your shell, it will still prompt you for permission. This provides a high level of productivity while maintaining a safety barrier against arbitrary code execution. It is the ideal middle ground for daily development work where you trust the agent with your code but not necessarily with your entire terminal environment.
Full automation using the bypass permissions mode
For tasks that are repetitive or low risk, you can use the bypassPermissions mode to remove all barriers. In this mode, the agent has full autonomy to use every tool in its arsenal, including Bash and Write, without ever asking for your consent. This turns the agent into a truly independent worker. You can give it a high-level goal and let it work through the entire process from file discovery to final testing and deployment without touching your keyboard.
You should use bypassPermissions with caution. It is best suited for isolated environments, such as a new branch where you can easily revert changes if something goes wrong. This mode is extremely powerful for running test suites, performing large-scale data migrations, or setting up new project scaffolding. Because there are no prompts, the session will continue until the agent believes the task is complete or it encounters an unrecoverable error. It is the fastest way to get work done when you are confident in your instructions.
Switching modes with keyboard shortcuts and commands
You can change your permission level at any point during a conversation without restarting the session. The fastest way to do this is by pressing Shift+Tab. This keyboard shortcut cycles through the four modes: default, acceptEdits, plan, and bypassPermissions. Your terminal interface will display the name of the active mode so you always know how much autonomy the agent currently has. This allows you to start in a safe mode and switch to a faster mode as you gain confidence in the task.
If you prefer using commands, you can type /permissions to manage your current settings. This slash command shows you the status of all permission levels and allows you to make adjustments through the interactive interface. You can also use /config to view or change your global settings during a session. Being able to toggle these modes on the fly means you can adjust the friction of your workflow as the complexity of your work changes from minute to minute.
Fine tuning tool access in settings files
For more permanent control over your workflow, you can define your permission preferences in configuration files. The .claude/settings.json file in your project root allows you to set specific rules for that repository. You can also use ~/.claude/settings.json to apply global permission settings across all your projects. By adding a permissions section to these JSON files, you can specify exactly which tools require manual approval and which ones are allowed to run automatically.
You can even create a .claude/settings.local.json file for settings that are specific to your local machine. This is useful for overriding project-wide permissions without affecting other developers on your team. For example, you might want to allow the Bash tool to run freely on your local development machine but keep it restricted in the main project settings. This layered approach to configuration gives you total control over the balance between safety and speed in every environment you work in.
Monitoring multiple automated sessions with adhdTerminal
Managing several sessions in bypassPermissions or acceptEdits can become difficult to track in a standard terminal. This is where adhdTerminal becomes useful. It is a native app for macOS and Windows, made by MINDBALANCER OU in Estonia, not affiliated with Anthropic. It runs many Claude Code sessions side by side, each one a tile on a wall with its own colour, number, and name. This layout allows you to see whether a tile is working, waiting for you, done, or idle, and it even shows the last exchanges of the conversation so you can monitor progress at a glance.
The app also features a free iPhone app that mirrors your wall over an end to end encrypted link, allowing you to answer a prompt or send the next task while away from your desk. Tiles reopen in the same place after a restart and resume the same session, ensuring your workflow remains consistent. It costs 99 dollars once, where one key runs one computer at a time and can be moved, and it comes with a 7 day free trial and no account required. This setup is ideal for developers who need to supervise multiple autonomous agents without constantly context switching.
Recovering from errors in quiet modes
Even when you are using a quiet mode like bypassPermissions, you still have the ability to intervene if the agent makes a mistake. If you see the agent performing an incorrect action in your terminal, you can press the Escape key once to immediately interrupt its current task. This stops the agent from continuing down a wrong path and allows you to provide new instructions or correct the error manually. This manual override is an essential safety feature for any automated workflow.
If a series of automated edits has already caused a problem, you can press the Escape key twice. This opens a history of your session and allows you to select an earlier message to go back to. It effectively rewinds the conversation, letting you discard the incorrect steps and try a different approach. You can also use the /clear command to start fresh in the same folder if the context has become too messy, or /compact to summarize the session and free up space in the context window while keeping the core progress.
Questions people ask
Can I restrict specific tools while using a quiet mode?
Yes, you can define tool-specific settings in your .claude/settings.json file. Within the permissions object, you can set specific tools like Bash or WebSearch to always require manual approval, even if the overall session is in a more permissive mode. This allows you to build a custom quiet mode that fits your specific security needs.
How do I see my current spending and token usage?
You can track your costs and token usage by using the /cost and /usage slash commands at any time. These commands provide a detailed breakdown of your spend for the current session and your overall account usage. This is helpful when running in bypassPermissions mode, as it ensures you are aware of the financial impact of long-running automated tasks.
What happens to permissions when I restart a session?
When you carry on a session using claude --continue, the agent typically remembers the settings of the previous interaction. However, it is a good practice to use /status or /permissions to verify your current mode upon resuming. If you want a specific mode to be the default for every new session, you should configure it in your global settings.json file.
Read next
- Add build instructions to your CLAUDE.md fileImprove project context by adding specific build commands and environment requirements to your CLAUDE.md file for the AI to follow.
- Add common error fixes to CLAUDE.mdImprove project memory by adding a section for common errors and their fixes to your CLAUDE.md configuration file.
- Add custom shell aliases for Claude CodeSpeed up your workflow by creating shell aliases for common Claude Code invocations and project-specific startup commands.
- Add test commands to a CLAUDE.md fileUnderstand how to define specific test runners and environment flags in your CLAUDE.md file so the tool runs the correct test suite.