ModApi gap: mods cannot read or write the live input field
Summary
The ModApi gives mods no way to read the current text in the input field or set/prefill it. transformInput only intercepts submitted text — it never sees what's being typed, and there's no cmd.ui.setInputText-style API. This blocks a whole class of UX (ghost-text suggestions, inline completion, input prefill) that pi-prompt-suggester and similar tools rely on.
Context
I built a prompt-suggester mod (https://github.com/burningportra/cmd-prompt-suggester) that aims to suggest the user's next prompt after each response — the pi-prompt-suggester UX. The core interaction I cannot implement:
- Show a suggestion as ghost text inside the input box — impossible; rendering is line-based only (
cmd.ui.notify / cmd.ui.setStatus / cmd.ui.widget, and widget is explicitly "not wired into the TUI yet").
- Prefill the input field so the user can edit before sending — impossible; no input-write API.
- Press Space to accept the suggestion into the field — impossible;
transformInput fires on submit, not on keystroke, so I can't fill the field or bind a key.
What I had to ship instead
- Suggestion rendered as a
💡 next: notice (line-based).
- Accept via
/j slash command or typing 1 — because there's no way to inject text into the field.
This works, but it's a strict downgrade from pi's UX (ghost text + Space-to-accept), and it's the single most-requested feature for this kind of mod.
Proposed API (open to alternatives)
cmd.ui.getInputText(): string — read the current input field contents.cmd.ui.setInputText(text: string): void — replace/prefill the input field.- Optionally a keybinding hook (
hooks.onKeypress) so mods can respond to Space/Tab/arrows while typing.
Why it matters
The ModApi docs advertise "typed-input interception" as a first-class capability, but it only covers submitted input. Interactive suggestion/prefill is the natural next step, and it's what users expect from a "prompt suggester" mod.
Environment
- Command Code 1.15.0, macOS
- Docs read: commandcode.ai/docs/mods (ModApi reference, UI surface)
ModApi gap: mods cannot read or write the live input field
Summary
The ModApi gives mods no way to read the current text in the input field or set/prefill it.
transformInputonly intercepts submitted text — it never sees what's being typed, and there's nocmd.ui.setInputText-style API. This blocks a whole class of UX (ghost-text suggestions, inline completion, input prefill) that pi-prompt-suggester and similar tools rely on.Context
I built a prompt-suggester mod (https://github.com/burningportra/cmd-prompt-suggester) that aims to suggest the user's next prompt after each response — the pi-prompt-suggester UX. The core interaction I cannot implement:
cmd.ui.notify/cmd.ui.setStatus/cmd.ui.widget, andwidgetis explicitly "not wired into the TUI yet").transformInputfires on submit, not on keystroke, so I can't fill the field or bind a key.What I had to ship instead
💡 next:notice (line-based)./jslash command or typing1— because there's no way to inject text into the field.This works, but it's a strict downgrade from pi's UX (ghost text + Space-to-accept), and it's the single most-requested feature for this kind of mod.
Proposed API (open to alternatives)
cmd.ui.getInputText(): string— read the current input field contents.cmd.ui.setInputText(text: string): void— replace/prefill the input field.hooks.onKeypress) so mods can respond to Space/Tab/arrows while typing.Why it matters
The ModApi docs advertise "typed-input interception" as a first-class capability, but it only covers submitted input. Interactive suggestion/prefill is the natural next step, and it's what users expect from a "prompt suggester" mod.
Environment