[
    {
        "role": "user",
        "content": "tell%20me%20about%20modes%20in%20claude%20code%20such%20as%20accept%20edits%20vs%20plan%20mode%20vs%20any%20others"
    },
    {
        "role": "assistant",
        "content": "Here\u2019s a breakdown of the different *permission modes* in Claude Code\u2014what each one does, when to use it, and how they differ. If you're working with Claude Code, you'll mostly see: **Default Mode**, **Accept Edits Mode**, **Plan Mode**, and **Bypass Permissions**. There\u2019s also a new **Auto Mode** in preview. ([claudelog.com](https://claudelog.com/faqs/how-to-set-claude-code-permission-mode?utm_source=openai))\n\n---\n\n## Modes in Claude Code\n\n| Mode | What It Means / Behavior | Common Use Cases | What\u2019s Restricted / Allowed |\n|---|---------------------------|-------------------|-----------------------------|\n| **Default Mode** | Standard behavior: Claude asks you for permission before doing anything \u201crisky\u201d \u2014 editing files, running shell commands, modifying folder structures, etc. ([medium.com](https://medium.com/%40sandrasoulful/claude-code-guide-for-beginners-a86d1517ad83?utm_source=openai)) | When you're in an unfamiliar codebase, or you want tight control over what changes are made. Good for learning or when safety is a priority. | All actions require approval: edits, file writes, executing commands. ([repovive.com](https://repovive.com/roadmaps/claude-code/getting-started/permission-modes?utm_source=openai)) |\n| **Accept Edits Mode** | Claude automatically edits files without asking for approval each time. It still asks the user when it needs to run shell commands or do significantly risky operations. ([lilys.ai](https://lilys.ai/notes/en/claude-code-20260210/claude-code-tutorial-2026?utm_source=openai)) | When you're confident in what you want done (e.g. rapid prototyping, small tweaks, trusted project), and want fewer prompts slowing you down. | Editing files = auto-approved. Shell commands / folder creation / possibly version control actions still need permission. ([repovive.com](https://repovive.com/roadmaps/claude-code/getting-started/permission-modes?utm_source=openai)) |\n| **Plan Mode** | Read-only mode. Claude can explore and analyze files, ask clarifying questions, generate detailed plans (often saved as markdown files). But it won\u2019t make any changes until you explicitly approve a plan. ([blakecrosley.com](https://blakecrosley.com/guides/claude-code?utm_source=openai)) | When the change is large, across many files, architectural or involves dependencies. Useful in code review, complex feature work, or exploring an unfamiliar codebase. ([repovive.com](https://repovive.com/roadmaps/claude-code/getting-started/permission-modes?utm_source=openai)) | No file edits, no running commands, no destructive actions. Only exploration tools like read, grep, file discovery. ([blakecrosley.com](https://blakecrosley.com/guides/claude-code?utm_source=openai)) |\n| **Bypass Permissions Mode** | Less common and riskier: Claude skips or aggressively minimizes permission prompts altogether. Meant for sandboxed or experimental environments. ([claudelog.com](https://claudelog.com/faqs/how-to-set-claude-code-permission-mode?utm_source=openai)) | Automated tasks, CI/CD, being able to trust the scope and content fully. Only when you know what you're doing. | Prompts are mostly suppressed \u2014 could allow risky operations if not careful. Not generally safe for sensitive codebases. ([claudelog.com](https://claudelog.com/faqs/how-to-set-claude-code-permission-mode?utm_source=openai)) |\n| **Auto Mode** _(Preview)_ | Launched recently as a \u201cmiddle path.\u201d Claude automatically grants safe permissions, but still blocks requests it considers risky \u2014 all without constant user prompting. ([techradar.com](https://www.techradar.com/pro/anthropic-gives-claude-code-new-auto-mode-which-lets-it-choose-its-own-permissions?utm_source=openai)) | When you want speed but still want safety guards. Good for trusted projects where you still want some oversight. | Depends on a permissions classifier: safe actions = automatic, risky ones = still need your OK. Limited rollout so may still behave conservatively or over-restrict. ([techradar.com](https://www.techradar.com/pro/anthropic-gives-claude-code-new-auto-mode-which-lets-it-choose-its-own-permissions?utm_source=openai)) |\n\n---\n\n## Switching Modes & Controls\n\n- You can cycle between **Default \u2192 Accept Edits \u2192 Plan** (and sometimes Bypass) by pressing **Shift + Tab** during a session. ([docs.claude.com](https://docs.claude.com/s/claude-code-worktrees?utm_source=openai))  \n- You can also start sessions in Plan mode via a flag/command-line, e.g. `claude --permission-mode plan` or using a shortcut like `-p`. ([docs.claude.com](https://docs.claude.com/s/claude-code-worktrees?utm_source=openai))  \n- In Plan Mode, after the plan is ready, Claude usually gives you options for how to execute, for example:\n  1. Clearing context and auto-accepting edits  \n  2. Manually approving each edit  \n  3. Just auto-accept edits (keeping context)  \n  etc. ([reddit.com](https://www.reddit.com/r/ClaudeAI/comments/1qf91g4/clear_context_new_option_when_plan_finishes/?utm_source=openai))\n\n---\n\n## Things to Be Aware Of\n\n- **Context clearing**: After Plan Mode finishes, there's often an option to \u201cclear context\u201d so the execution phase doesn't carry extra chat context or unused tokens. This option used to be visible by default, but in newer versions (e.g. v2.81.1), it\u2019s hidden unless you enable a setting like `\"showClearContextOnPlanAccept\": true` in your `.claude/settings.json`. ([reddit.com](https://www.reddit.com/r/ClaudeCode/comments/1s0813s/no_more_auto_accept_edits_and_clear_context/?utm_source=openai))  \n- Some users have reported that in Plan Mode Claude sometimes dead-ends and tries to edit anyway (bugs or things edge cases), especially if toggled mid-task. It\u2019s not supposed to happen in theory; Plan Mode should be read-only until you approve. ([reddit.com](https://www.reddit.com/r/ClaudeCode/comments/1puvvww/claude_code_making_edits_while_in_plan_mode_this/?utm_source=openai))  \n- For simple or single-file modifications, using Plan Mode might introduce unnecessary overhead. In those cases Accept Edits or even Default Mode can be faster. ([claude-ai.chat](https://claude-ai.chat/blog/plan-mode-in-claude-code-when-to-use-it/?utm_source=openai))  \n\n---\n\n## What\u2019s Best When\n\n- Use **Plan Mode** for anything large, complex, or where mistakes are costly.\n- Use **Accept Edits** when you want speed, trust the task, or you're iterating quickly.\n- Use **Default** when exploring unfamiliar territory or learning.\n- Use **Bypass** only in trusted sandbox environments.\n- Use **Auto Mode** when you want a mix: speed + some safety.\n\n---\n\nIf you want, I can show you the **latest behaviors** (as of *today*) with some examples, or help pick which mode fits your current workflow. Want that?"
    }
]