feat: extended interactive prompt options (Yes, Always Allow, Text Box)

[VforVitorio]

What problem does this solve?

The interactive [Y/n/e]\ prompt falls a bit short in ask mode compared to Claude Code's capabilities.

We successfully integrated a spectacular diff viewer in the \eat/interactive-permissions\ branch, but currently, if the user rejects (
), it simply aborts execution or assumes a generic error. Furthermore, there's no way to give feedback or to say 'yes, but leave me alone from now on for this tool'.

Proposed solution

1. Expand the options to be more powerful (potentially adding keyboard shortcuts).
2. Yes and always allow: Allow the user to accept this tool invocation and automatically switch to \�uto\ mode so it stops asking for safe tools.
3. Feedback text box (No, do this instead): Instead of aborting right away, if the user chooses
   , deploy an inline \prompt_toolkit\ input box within the confirmation block so the user can instruct the agent on what to do instead and why the action was rejected.
4. This text correction should be injected into the history as a 'user' message to steer the agent back on track.

Alternatives considered

• Sticking with the pure \�sk\ flow and forcing the user to manually interrupt the agent to explain their disagreement, which adds friction to the workflow.

[VforVitorio]

Resolved by #102 (merged to dev). The extended interactive prompt now supports all four requirements: Yes / Always allow this tool (session-scoped via _always_allowed_tools) / No with feedback textbox / Esc to cancel. GitHub did not auto-close on merge because the PR landed on dev rather than main.