[FEATURE] Feedback should be a conversation, and the model should know what was sent

What the user wants

To talk feedback through in the chat: the model shows the text, the user says "change X", the model edits it, the user says "yes, send", and it's sent. Then, when asked "what did you send?", the model answers.

What happens today

• The model can only queue drafts. It can't edit, withdraw or send them.
• The model can't see what was sent or dismissed. ~/.claude/feedback/drafts/<id>.json has only "status": "queued". Sent and dismissed drafts just disappear, with no record of which happened. When the user asked "didn't you send some already?", the honest answer was "I can't tell".
• Each correction becomes another draft, and the user has to decode 1 to review · 2 to send · 0 to dismiss.

The user's words: "I just wanna talk to you … and like, yes, send. Like that is a smooth workflow. This workflow is ridiculous." And: "YOU CANT SEE WHAT WAS SENT THATS TERRIBLE TOO."

Workaround we are using

We stopped using /feedback for anything we need to track. Reports now go in as GitHub issues on this repo, filed by the model after the user says "yes". The model can then list them, comment on them and check their state, so the feedback is manageable end to end. That is what the built-in feedback flow should offer.

Request

• Let the model send a draft after explicit user confirmation in chat, and let it edit or withdraw its own drafts.
• Record the outcome (sent, with an id or timestamp, or dismissed) where the model can read it.

Claude Code 2.1.280, Linux.