[Feature Request] Add `promptStyle` config option for rate-limit/decision prompts
Bug Description
Title: Add config to choose between modal and chat-message style for rate-limit /
decision prompts
Body:
The current rate-limit dialog (and similar decision prompts) is a blocking TUI modal.
This causes three compounding problems during active coding sessions:
- Harness freeze: Background tasks (e.g. run_in_background watchers) cannot deliver
their completion signals while the modal is open. This breaks automation patterns like
auto-resuming after a 5h window reset.
- Context obscured: The modal overlays the in-progress work. Users must scroll up to
read the full task context before making a meaningful choice.
- Poor readability for rich choices: Development decisions often involve multi-line
descriptions, code snippets, and option rationale. A modal with limited space and no
markdown rendering makes these hard to read and impossible to copy.
Proposed change:
Add a settings.json key to let users choose how prompts are delivered:
{
"promptStyle": "chat"
}
┌─────────┬──────────────────────────────────────────────────────────────────────────┐
│ Value │ Behavior │
├─────────┼──────────────────────────────────────────────────────────────────────────┤
│ "modal" │ Current behavior (blocking TUI dialog). Default to preserve backwards │
│ │ compatibility. │
├─────────┼──────────────────────────────────────────────────────────────────────────┤
│ │ Deliver the prompt as a regular message in the conversation stream. │
│ "chat" │ Harness stays live; background tasks keep running; user replies in chat │
│ │ as normal. │
└─────────┴──────────────────────────────────────────────────────────────────────────┘
A settings.json option like "rateLimitAction": "wait" to silently default to waiting
(without even a chat prompt) would also address the full automation case.
Environment Info
- Platform: darwin
- Terminal: iTerm.app
- Version: 2.1.159
- Feedback ID: b50ac897-d0df-4428-921f-4da1d2c122d8
Errors
[]This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗