Let me edit the agent's proposed content inline before authorizing the action

## Request

Let me edit the agent's proposed content inline in the chat, then authorize it — instead of spending a full turn describing a tiny fix.

## Example

I ask: **"Update my GitHub profile."** The agent replies with proposed bio text that's almost perfect, but the last sentence needs to go. Today I have to write a follow-up ("drop the last sentence") and wait for the agent to regenerate. Instead, I want to delete that sentence directly in the agent's message and just say **"make the update"** — and the agent uses my edited version as the input to its action.

## Why this is different from existing requests

- **Not** user-message editing (#1602) — that re-runs my prompt and discards the agent's output.
- **Not** diff editing (#1397) — here the content lives only in the chat message; nothing is on disk yet.

The agent's message is sometimes the deliverable itself. Treat it as an editable draft I can amend before authorizing the action (like editing an AI-drafted email before hitting send).

Start by tracing the chat message and action-authorization flow described in the request. Done means a user can edit proposed agent content inline, authorize the action, and have the action use the edited version rather than the original draft.

索引モデルが issue の本文から書いたものです。