Tool approval 'Yes + comment' — comment silently dropped, never reaches the model

Problem

When a user approves a tool call with Yes + an additional comment (e.g. \"Yes, but don't ask me about read-only operations\"), the comment is silently dropped. The model only receives that the tool was approved — the comment never arrives as a conversation message.

By contrast, when a user rejects a tool call with \"No, [comment]\", the comment DOES come back to the model as a rejection message. This asymmetry is inconsistent and means constructive feedback on approvals is lost.

Impact

• Users cannot correct or guide the model inline while approving an action
• The model keeps repeating the same mistakes despite the user commenting on them alongside approvals
• The user is forced to send a separate follow-up message after approving — breaking flow
• Categorize the types of actions for the model to ask or not-ask on is needed. But seems the commonly showing option like : "2. Yes, but skip next time when ..." are a bit limited.

Steps to Reproduce

1. Claude Code prompts for tool approval
2. User types a comment alongside "Yes" (e.g. "Yes, and stop asking about read-only things")
3. Tool runs — but the comment never reaches the model
4. Model repeats the same behaviour next time

Expected Behavior

One of:

1. "Yes + comment" delivers the comment back to the model as a follow-up conversation message (same behaviour as the rejection flow)
2. A distinct UI option — "Approve + note" — that explicitly passes the note through to the model

Workaround

User sends the correction as a separate message after approving. Works but breaks flow and is easy to forget.

Environment

• Claude Code CLI
• macOS