Prompt Thinking extension

Type ^level in the editor to pick a thinking level for a single prompt. The control token is removed before the prompt is sent.

Install

pi install npm:@siddr/pi-prompt-thinking

Usage

1. Type ^ anywhere in the prompt where whitespace-delimited tokens are allowed.
2. Pick a level from autocomplete.
3. Send the prompt normally.
4. pi temporarily applies that thinking level for the prompt, then restores the previous session level after the response finishes.

Examples

Typing:

^high summarize the tradeoffs

submits as:

summarize the tradeoffs

Typing:

Please ^minimal answer briefly

submits as:

Please answer briefly

Autocomplete behavior

• Typing ^ opens a thinking-level picker in the interactive TUI.
• The current thinking level is read when the picker opens and preselected when it is available for the active model.
• Up/down arrows move through lower/higher thinking levels because suggestions stay in natural order.
• Available levels depend on the current model:
  • non-reasoning models: ^off
  • reasoning models: ^off, ^minimal, ^low, ^medium, ^high
  • xhigh-capable models: also ^xhigh

Token rules

• ^level can appear anywhere in the prompt as a standalone whitespace-delimited token.
• Matching is case-insensitive (^HIGH works).
• Unknown tokens are left unchanged.
• If multiple recognized tokens appear, the first recognized level wins and all recognized ^level tokens are removed before submission.

Notes

• The one-prompt behavior is implemented by temporarily changing pi's session thinking level before the prompt runs and restoring the previous level afterward.
• Because pi does not currently expose a true per-turn thinking override API to extensions, this may still append thinking-level-change entries to the session history and briefly update the default thinking setting before it is restored.
• Submit-time stripping works even when custom editor UI is unavailable; autocomplete itself is interactive-TUI only.