Describe the feature or problem you'd like to solve
When the agent streams a long response, the terminal auto-follows the bottom of the output. This forces me to either (a) wait for streaming to finish and then scroll back up, or (b) manually scroll up mid-stream to break the auto-follow. Both interrupt reading flow.
Proposed solution
Request: A setting (e.g. /footer-style toggle, or a config key like responseAnchor: "top" | "bottom") that pins the viewport to the first line of the agent's response when
streaming starts, so the user can read top-to-bottom at their own pace while new tokens arrive below the visible area. Similar to how less/git log open at the top of their
output.
Why terminal-level settings don't solve it: Windows Terminal (and most terminals) only break auto-follow when the user manually scrolls. There's no "anchor on new output
region" primitive at the terminal layer — it has to come from the agent's renderer, which knows where each response begins.
Nice-to-haves:
- Keybinding to jump to the top of the current/last response (e.g. Ctrl+Shift+R for "response top").
- Same toggle should work in autopilot mode where streams can be very long.
Example prompts or workflows
No response
Additional context
No response
Describe the feature or problem you'd like to solve
When the agent streams a long response, the terminal auto-follows the bottom of the output. This forces me to either (a) wait for streaming to finish and then scroll back up, or (b) manually scroll up mid-stream to break the auto-follow. Both interrupt reading flow.
Proposed solution
Request: A setting (e.g. /footer-style toggle, or a config key like responseAnchor: "top" | "bottom") that pins the viewport to the first line of the agent's response when
streaming starts, so the user can read top-to-bottom at their own pace while new tokens arrive below the visible area. Similar to how less/git log open at the top of their
output.
Why terminal-level settings don't solve it: Windows Terminal (and most terminals) only break auto-follow when the user manually scrolls. There's no "anchor on new output
region" primitive at the terminal layer — it has to come from the agent's renderer, which knows where each response begins.
Nice-to-haves:
Example prompts or workflows
No response
Additional context
No response