[FEATURE] claude.ai: let users set a default permission mode, incl. "Skip all approvals"

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Problem

In the new combined chat and Cowork experience on claude.ai, a task can stop at "Required input" and wait for my approval. The permission selector in the message box offers only Manual (the default) and Auto. It does not offer "Skip all approvals". There is no setting in account Settings to change the default.

This is most disruptive during research. A task that reads many websites can stop again and again, and I must return to the tab to approve access. I want to authorise routine tool use once and let the task finish without supervision.

Design concern

The current behaviour is very annoying to users and needs refining. Good interface design puts a user's standing preferences in Settings, sets once and applies everywhere. It does not make the user repeat the same decision in every conversation. A control that interrupts work, cannot be switched off, and must be managed per conversation adds friction without adding value for users who have already made an informed choice. It also teaches users to click "Allow" without reading, which reduces the safety the prompt is meant to give. Safe defaults for new users are correct. But experienced users need a clear, permanent way to change that default. Other assistants and Claude Code itself already give users this control; claude.ai should do the same.

Requested change

1. Add a "Default permission mode" preference in account Settings, with three choices: Manual, Auto and Skip all approvals.
2. Apply this default to all new conversations. Keep the message-box selector as a per-conversation override.
3. Make "Skip all approvals" available in the new combined experience. It already exists in the Chrome side panel and was available in Cowork before the merge.
4. Show a clear, one-time confirmation when a user first enables Skip. Do not show it again after the user confirms.
5. Keep the active mode visible in the message box at all times.
6. Keep the existing controls that let Team and Enterprise admins disable Skip and Auto for their organisation.

If Skip cannot be offered for all actions, please at least provide a persistent way to approve read-only website access across conversations. I understand that some protected actions (for example deleting files, or entering sensitive information) will still need explicit approval.

Why this matters

For a multi-step research task, even a small number of approval prompts defeats the purpose of leaving Claude to work on its own. Auto reduces interruptions, but it can still pause for input, and its safety checks use more of the usage allowance. Users who accept the trade-off should be able to choose their default once, not set it again in each conversation.

Related issues

• #62076 – permission mode resets at the start of each Cowork session
• #76141 – "Always allow" does not persist across runs
• #92740 – "Skip all approvals" still demands an approval card for the first artifact publish
• #97969 – web session mode menu is limited

Environment

claude.ai web on Linux, new combined chat and Cowork experience (Pro/Max rollout from 16 September 2026).

Proposed Solution

1. Add a "Default permission mode" preference in account Settings (for example Settings > General), with three choices: Manual, Auto and Skip all approvals.
2. Apply this default to every new conversation on web, desktop and mobile. Keep the message-box selector as a per-conversation override.
3. Make "Skip all approvals" available again in the new combined chat and Cowork experience. It already exists in the Chrome side panel and was available in Cowork before the merge.
4. Show one clear confirmation, with the risks stated, the first time a user selects Skip. Do not show it again after the user confirms.
5. Keep the active mode visible in the message box at all times, so the user always knows which mode applies.
6. Keep the existing admin controls: Team and Enterprise admins can still disable Skip and Auto for their organisation.
7. Protected actions (for example permanent deletion, entering sensitive information, granting authorisations) continue to need explicit approval in every mode.

Alternative Solutions

1. Remember the last choice. Make the message-box selector keep the user's last selection across conversations, as the Chrome side panel already does. This is less work to build, but a Settings entry is easier to find, to audit and to change.

1. Persistent approval for read-only web access. Add an "Always allow" for web fetch and web search that persists across conversations. This would remove most prompts during research tasks, even if Skip is not offered.

1. Per-action "Always allow" that persists. When the user clicks "Always allow" on an approval card, store it at account level so the same action does not prompt again in later conversations.

1. Current workaround (not adequate). Select Auto in each conversation and set all connector tools to "Always allow" in Customize > Connectors. This reduces prompts, but Auto can still stop for input, uses more of the usage allowance, and must be selected again in each new conversation.

Priority

Critical - Blocking my work

Feature Category

CLI commands and flags

Use Case Example

_No response_

Additional Context

_No response_