[FEATURE] Context-aware auto-accept for repeatable, low-risk sub-tasks

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

There are certain repetitive tasks that are well suited for allowing ClaudeCode to take the reins during execution of a plan:

Say the below examples constitute a checkbox in claude's todo list:

[x] Replace all blabla -> Task 1: (Good for auto-accept):

• Refactoring a block of code that is duplicated among multiple files (i.e. changing a param in a func call)

[x] Implement different signing scheme for blablabla -> Task 2: (Bad for auto-accept, bespoke step which should be reviewed):

• As part of the larger todo list or claude plan, claude has made a todo sub-task to change signing scheme for blabla, a task of medium+ complexity to be implemented in only one part of the codebase or repeated twice but very differently in both locations for some contextual reasons.

Proposed Solution

A great flow would be something like hitting tab takes me to some kinda hybrid auto/manual accept mode:

For the first time Claude does a task (or an action of certain complexity), say Task 1, it would be nice if the app considered that the start of a contextual session belonging to that task i.e. the param change across files.

It would propose an initial change as normal, and then based on the repeatability of the change, we're prompted if we wanna let that auto-complete until that sub task is complete.

At which point Claude completes that and then moves on to Task 2 in which case it either only needs to do it once or the variability of the change (if a repeated task) is too high so I can appropriately supervise it.

This stops me from manually having to accept a bunch of donkey changes or needing spiderman reflexes/staring at the CC output so that I can turn off auto-accept until I think it's done with the menial work and I need to be an active reviewer again.

Alternative Solutions

Currently I just watch the edits and thought stream until I get the feeling like the menial work is done and quickly switch back to manual. I don't like having to sit there just to babysit very easy/menial work that I have high confidence CC can execute correctly.

Or I manually context manage: i.e. Copying and pasting the task list and enriching it with wider context of what I'm doing and then deciding myself if the task is simple enough to leave in auto-accept mode, this is a headache

Both of these add to my mental load and I have a gut feeling they reduce my CC coding sessions by ~30% because I get tired of supervision for simple tasks and/or manual context engineering.

Priority

High - Significant impact on productivity

Feature Category

Interactive mode (TUI)

Use Case Example

In proposed solution.

Additional Context

_No response_