[FEATURE] Add "Accept/Deny with comment/reason" option to permission prompts (alongside Allow / Deny)
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
When Claude Code Desktop asks for permission to run a tool, the options are: Allow once / Always allow / Deny. A plain Deny gives Claude zero signal about why it was denied. Claude then either retries the same action, asks a vague clarifying question, or stalls. The user ends up typing the reason in the chat box anyway — this is a redundant, broken step that breaks flow.
Proposed Solution
Add a fourth option to permission prompts: "Deny with comment". When selected, a short text input appears inline where the user types the reason (e.g. "wrong file", "use Edit tool not sed", "find another approach"). Claude receives the comment as context immediately and adjusts its next action accordingly — no separate chat message needed.
Claude wants to run: rm -rf ./tmp
› Allow once
Always allow
Deny
Deny with comment...
Alternative Solutions
Instead of a fourth option, a text field could appear below the existing Deny button automatically after it is clicked (a two-step: click Deny → optional comment box appears → submit). This keeps the initial prompt clean and still lets users add context when they want to.
Priority:
Medium-High — this is a UX quality-of-life improvement that meaningfully reduces back-and-forth when Claude is heading in a wrong direction. The current flow forces an unnecessary chat roundtrip for every corrective denial.
Priority
High - Significant impact on productivity
Feature Category
Interactive mode (TUI)
Use Case Example
Claude asks permission to run git reset --hard HEAD. User wants to deny but explain why — they want to stash instead, not discard. Current flow: Deny → type explanation in chat → Claude re-plans. With this feature: "Deny with comment: use git stash instead of reset --hard" → Claude immediately re-plans with that context in one step.
Additional Context
This pattern already exists in other AI coding tools and code review systems where inline feedback on a rejected action is standard. It would make the permission prompt a two-way communication surface rather than a one-bit signal.